Search

Depth estimation provides essential information to perform autonomous driving and driver assistance. A promising line of work consists of introducing additional semantic information about the traffic scene when training CNNs for depth estimation. In practice, this means that the depth data used for CNN training is complemented with images having pixel-wise semantic labels where the same raw training data is associated with both types of ground truth, i.e., depth and semantic labels. The main contribution of this paper is to show that this hard constraint can be circumvented, i.e., that we can train CNNs for depth estimation by leveraging the depth and semantic information coming from heterogeneous datasets. In order to illustrate the benefits of our approach, we combine KITTI depth and Cityscapes semantic segmentation datasets, outperforming state-of-the-art results on monocular depth estimation.

Avoiding vehicle-to-pedestrian crashes is a critical requirement for nowadays advanced driver assistant systems (ADAS) and future self-driving vehicles. Accordingly, detecting pedestrians from raw sensor data has a history of more than 15 years of research, with vision playing a central role. During the last years, deep learning has boosted the accuracy of image-based pedestrian detectors. However, detection is just the first step towards answering the core question, namely is the vehicle going to crash with a pedestrian provided preventive actions are not taken? Therefore, knowing as soon as possible if a detected pedestrian has the intention of crossing the road ahead of the vehicle is essential for performing safe and comfortable maneuvers that prevent a crash. However, compared to pedestrian detection, there is relatively little literature on detecting pedestrian intentions. This paper aims to contribute along this line by presenting a new vision-based approach which analyzes the pose of a pedestrian along several frames to determine if he or she is going to enter the road or not. We present experiments showing 750 ms of anticipation for pedestrians crossing the road, which at a typical urban driving speed of 50 km/h can provide 15 additional meters (compared to a pure pedestrian detector) for vehicle automatic reactions or to warn the driver. Moreover, in contrast with state-of-the-art methods, our approach is monocular, neither requiring stereo nor optical flow information.

Introduction: Enhancement of exercise capacity (EC) and physical activity (PA) are core components of preoperative programs for optimization of multimorbid patients undergoing major surgery. However, the relationship between these 2 components has never been explored in this population. Aim: To explore baseline relationship of EC & PA in multimorbid patients included in a preoperative optimization program. Methods: Main inclusion criteria: 1) Undergoing major surgery; 2) High-risk for surgical complications defined by age≥70 and/or American Society of Anesthesiologists risk scale 3-4. Six-minute walk distance Results: 106 patients were included in the study. Main univariate analysis results are shown in table 1. Multivariate analysis results, taking fit & active as a reference group, are shown in table 2. Conclusions: Age, morbidity and frailty are factors associated to EC and PA in multimorbid elderly patients undergoing major surgery. These findings may contribute to enhance personalisation of preoperative optimization programs.

Background Multimodal prehabilitation is a preoperative intervention with the objective to enhance cancer patients’ functional status which has been showed to reduce both postoperative morbidity and hospital length of stay in digestive oncologic surgery. However, in lung cancer surgery patients further studies with higher methodological quality are needed to clarify the benefits of prehabilitation. The main aim of the current protocol is to evaluate the cost-effectiveness of a multimodal prehabilitation program supported by information and communication technologies in moderate-to-high risk lung cancer patients undergoing thoracic surgery. Methods A Quadruple Aim approach will be adopted, assessing the prehabilitation program at the following levels: i) Patients’ and professionals’ experience outcomes (by means of standardized questionnaires, focus groups and structured interviews); ii) Population health-based outcomes (e.g. hospital length of stay, number and severity of postoperative complications, peak oxygen uptake and levels of systemic inflammation); and, iii) Healthcare costs. Discussion This study protocol should contribute not only to increase the scientific basis on prehabilitation but also to detect the main factors modulating service adoption. Trial registration NCT04052100 (August 9, 2019).

Corporate Social Responsibility (CSR) and its disclosure in the university environment is a topic of current relevance, as it makes the entities&rsquo, commitments visible and provides indicators that enable them to improve the institution management and communication with stakeholders. The goal of this study is to determine to what extent the structure and mechanisms for governance and the demands of stakeholders influence policy for disclosing CSR information, both in general (more related to a strategic perspective) and specifically (more focused on specific social, environmental, economic, and educational issues). The results of our analysis of a sample of the top 200 universities in the Shanghai Ranking show no association of the profile and gender of the university&rsquo, s rector and frequency of board meetings with CSR disclosure policy, but leadership team, the size of governance board, committees in the governance board and stakeholder participation are factors determining disclosure of information on matters of CSR. The results show that proximity to the day-to-day, diversification of functions, and communication with interest groups are crucial to transparency and disclosure of CSR information.

On-board vision systems may need to increase the number of classes that can be recognized in a relatively short period. For instance, a traffic sign recognition system may suddenly be required to recognize new signs. Since collecting and annotating samples of such new classes may need more time than we wish, especially for uncommon signs, we propose a method to generate these samples by combining synthetic images and Generative Adversarial Network (GAN) technology. In particular, the GAN is trained on synthetic and real-world samples from known classes to perform synthetic-to-real domain adaptation, but applied to synthetic samples of the new classes. Using the Tsinghua dataset with a synthetic counterpart, SYNTHIA-TS, we have run an extensive set of experiments. The results show that the proposed method is indeed effective, provided that we use a proper Convolutional Neural Network (CNN) to perform the traffic sign recognition (classification) task as well as a proper GAN to transform the synthetic images. Here, a ResNet101-based classifier and domain adaptation based on CycleGAN performed extremely well for a ratio &sim, 1 / 4 for new/known classes, even for more challenging ratios such as &sim, 4 / 1 , the results are also very positive.

The political influence on the financial condition of local government has been examined in various studies in the academic literature. However, no clear relationship has yet been established between such political factors and the constituent elements of financial condition sustainability in the context of the recent Great Recession. The aim of this study is to evaluate the dimensions of electoral size, defined for the chapter's purposes as the effective number of parties, and of transfer (i.e., how electoral gains and losses are related to the configuration of the party system) using a series of aggregate indices. Once these indicators are obtained, they are related to various indicators of financial condition for five Spanish cities. The results obtained show that financial condition either worsened or presented little change in the different cases considered and that various scenarios of aggregate volatility during the study period were detected.

Nanomaterials produced from the bottom-up assembly of nanocrystals may incorporate ~1020–1021 cm–3 not fully coordinated surface atoms, i.e., ~1020–1021 cm–3 potential donor or acceptor states that can strongly affect transport properties. Therefore, to exploit the full potential of nanocrystal building blocks to produce functional nanomaterials and thin films, a proper control of their surface chemistry is required. Here, we analyze how the ligand stripping procedure influences the charge and heat transport properties of sintered PbSe nanomaterials produced from the bottom-up assembly of colloidal PbSe nanocrystals. First, we show that the removal of the native organic ligands by thermal decomposition in an inert atmosphere leaves relatively large amounts of carbon at the crystal interfaces. This carbon blocks crystal growth during consolidation and at the same time hampers charge and heat transport through the final nanomaterial. Second, we demonstrate that, by stripping ligands from the nanocrystal surface before consolidation, nanomaterials with larger crystal domains, lower porosity, and higher charge carrier concentrations are obtained, thus resulting in nanomaterials with higher electrical and thermal conductivities. In addition, the ligand displacement leaves the nanocrystal surface unprotected, facilitating oxidation and chalcogen evaporation. The influence of the ligand displacement on the nanomaterial charge transport properties is rationalized here using a two-band model based on the standard Boltzmann transport equation with the relaxation time approximation. Finally, we present an application of the produced functional nanomaterials by modeling, fabricating, and testing a simple PbSe-based thermoelectric device with a ring geometry.

With the purpose of increasing the quality of public financial information, improving decision-making, transparency and accountability processes, the countries of the Central American region have made substantial efforts to improve their governmental accounting systems. For this reason, most countries have chosen to implement the International Public Sector Accounting Standards (IPSAS) issued by the International Federation of Accountants (IFAC). This process has been enhanced by the impulse of international financial organizations that aim to these countries advancing their government financial management practices. In this context, the objective of this research is to define the IPSAS implementation processes in the countries of Central America as well as to analyze the main elements of the environment that are in favor or against this process, as indicated by Professor Lüder's Contingency Model.

The analysis of e-Government is of particular importance in developing countries because it constitutes a central element in the process of modernizing public administrations and strengthening governance within democratic societies. This chapter seeks to analyze the research efforts and directions in the field of e-Government performed in developing countries as well as past experiences in the implementation of e-Government in developing countries to learn some lessons for improving future implementation of e-Government, which could help to achieve more transparent, participative and democratic societies. Analysis indicates that e-Government implementations need to be analyzed from a systemic thinking approach, which suggests the use of strategic planning for e-Government success. International bodies should fund activities that allow developing countries to achieve political and social changes needed to make e-Government implementations successful and should monitor e-Government efficiency in developing countries with the aim at improving economic, democratic and social development of these countries.

Deep video action recognition models have been highly successful in recent years but require large quantities of manually annotated data, which are expensive and laborious to obtain. In this work, we investigate the generation of synthetic training data for video action recognition, as synthetic data have been successfully used to supervise models for a variety of other computer vision tasks. We propose an interpretable parametric generative model of human action videos that relies on procedural generation, physics models and other components of modern game engines. With this model we generate a diverse, realistic, and physically plausible dataset of human action videos, called PHAV for "Procedural Human Action Videos". PHAV contains a total of 39,982 videos, with more than 1,000 examples for each of 35 action categories. Our video generation approach is not limited to existing motion capture sequences: 14 of these 35 categories are procedurally defined synthetic actions. In addition, each video is represented with 6 different data modalities, including RGB, optical flow and pixel-level semantic labels. These modalities are generated almost simultaneously using the Multiple Render Targets feature of modern GPUs. In order to leverage PHAV, we introduce a deep multi-task (i.e. that considers action classes from multiple datasets) representation learning architecture that is able to simultaneously learn from synthetic and real video datasets, even when their action categories differ. Our experiments on the UCF-101 and HMDB-51 benchmarks suggest that combining our large set of synthetic videos with small real-world datasets can boost recognition performance. Our approach also significantly outperforms video representations produced by fine-tuning state-of-the-art unsupervised generative models of videos., Comment: Pre-print of the article accepted for publication in the Special Issue on Generating Realistic Visual Data of Human Behavior of the International Journal of Computer Vision (IJCV). arXiv admin note: substantial text overlap with arXiv:1612.00881

A major question for public managers is whether municipal services should be rendered in-house or contracted out. In view of the negative perceptions often aroused by contracting out, this politica...

Background COVID-19 infection has been known to predispose patients to both arterial and venous thromboembolic events such as deep venous thrombosis, pulmonary embolism, myocardial infarction, and stroke. A few reports from the literature suggest that Cerebral Venous Sinus Thrombosis (CVSTs) may be a direct complication of COVID-19. Objective To review the clinical and radiological presentation of COVID-19 positive patients diagnosed with CVST. Methods This was a multicenter, cross-sectional, retrospective study of patients diagnosed with CVST and COVID-19 reviewed from March 1, 2020 to November 8, 2020. We evaluated their clinical presentations, risk factors, clinical management, and outcome. We reviewed all published cases of CVST in patients with COVID-19 infection from January 1, 2020 to November 13, 2020. Results There were 8 patients diagnosed with CVST and COVID-19 during the study period at 7 out of 31 participating centers. Patients in our case series were mostly female (7/8, 87.5%). Most patients presented with non-specific symptoms such as headache (50%), fever (50%), and gastrointestinal symptoms (75%). Several patients presented with focal neurologic deficits (2/8, 25%) or decreased consciousness (2/8, 25%). D-dimer and inflammatory biomarkers were significantly elevated relative to reference ranges in patients with available laboratory data. The superior sagittal and transverse sinuses were the most common sites for acute CVST formation (6/8, 75%). Median time to onset of focal neurologic deficit from initial COVID-19 diagnosis was 3 days (interquartile range 0.75–3 days). Median time from onset of COVID-19 symptoms to CVST radiologic diagnosis was 11 days (interquartile range 6–16.75 days). Mortality was low in this cohort (1/8 or 12.5%). Conclusions Clinicians should consider the risk of acute CVST in patients positive for COVID-19, especially if neurological symptoms develop.

During the last years, Transparency has become in a key element in the interaction between public administrations and society. However, this is a broad concept, which needs to be defined and measured carefully. Various authors have addressed this concept, from the perspective of different areas, through the creation of dichotomous indexes of transparency, but not measuring the quality of the information. In the present study, we propose a new method to measure the quality and content of municipal information disclosure, avoiding the creation of a synthetic index, highlighting the elements of transparency that provide optimal information, in quality and quantity. After that, we identify differences between local governments with different political ideologies, using multidimensional scaling analysis. The results obtained show that conservative governments grant more importance to transparency indicators, especially regarding legal and institutional information, than progressive ones.

The aim of this study is to identify, synthesise and evaluate the current state of research into transparency in the public sector, with respect to the period 2005–2014, in order to provide a critical overview of published trends in this field and to identify future areas of interest. Specifically, we focus on the questions of mandatory and voluntary transparency that arouse most interest among researchers, institutions and journals in this field, with particular attention to the methodological aspects of the studies considered. To achieve these aims, various scientometric approaches are applied to articles included in the 2014 edition of the ISI Journal Citation Reports database for the research areas “Public Administration”, “Political Science” and “Information and Library Science”. During the past ten years the number of studies published in this field has increased considerably, particularly as regards voluntary transparency.

A mTICI 2b or a mTICI 3 score are currently considered success following mechanical thrombectomy (MT) in acute stroke but is undetermined whether the two scores translate equivalent outcomes. We present a single-center, retrospective cohort of patients with anterior circulation stroke treated with MT and achieving a final mTICI score 2b or 3. A multimodal CT at baseline and a multimodal MRI at 24 hours assessed the growth of the infarct, and the modified Rankin Scale (mRS) assessed functional outcome at 90 days. The primary outcome was the shift analysis of the mRS at day 90 in ordinal regression adjusted for covariates (age, sex, pretreatment NIHSS score, target occlusion, infarct core, pretreatment alteplase), and the collateral score. Infarct growth was explored in a similarly adjusted multiple linear regression model. MT was started within a median of 285 minutes of symptom onset; 51 (41%) patients achieved a mTICI 2b, and 74 (59%), a mTICI 3. mTICI 3 resulted in better mRS score transitions than mTICI 2b (odds ratio 2.018 [95% CI 1.033–3.945], p = 0. 040), and reduced infarct growth (p = 0.002). We conclude that in patients with acute stroke receiving MT, success should be redefined as achieving a mTICI 3 score.

In a context of greater demands in terms of accountability and transparency in management, we analyse the extent to which Corporate Governance (CG) mechanisms have been adopted and we define the model of governance in universities in Colombia. In the field of higher education, there are usually no compulsory requirements related to CG. Nevertheless, governance mechanisms are nowadays crucial in determining the objectives, strategies and direction of universities. Based on a survey of 81 rectors of higher education institutions in Colombia, we analyse the models of CG adopted (managerial and stakeholder models). The results contribute to a better understanding of key CG concepts at universities. Points for practitioners There is an increasing awareness of the importance of improving the mechanisms of governance that may help to achieve greater effectiveness and efficiency in the management of resources and awareness of society. The analysis of the predominant CG models, the stakeholder model and the managerial model, and the way in which Colombian universities are introducing these models could be used to promote the establishment of a common legal framework. This model should be flexible and include different opinions and preferences. The results may be useful and could be transferred to other countries that have the same characteristics.

Various studies have analyzed the relationship between fiscal stress and contracting out, but have failed to achieve conclusive results. In this article, we take a broad view of fiscal stress, addressed in terms of financial condition and studied over a lengthy period (2000-2010). The relationship between fiscal stress and contracting out is studied using a dynamic model, based on survival analysis, a methodology that enables us to take into account the effect of time on this relationship. As this study period includes the years of the Great Recession (2008-2010), we also highlight the impact of this event on the fiscal stress–contracting out relation. The results obtained suggest that taking into account the passage of time and conducting a long-term assessment of financial condition enable a more precise understanding of this relation. We also find that the Great Recession reduced the probability of local governments’ contracting out public services.

The aim of this paper is to analyse the level of implementation by Spanish local authorities of the Transparency Act (19/2013), and their voluntary information disclosure in areas where such disclosure is obligatory in other EU countries. We identify the main factors that may influence the implementation of this Act and the self-regulation policies adopted. Our analysis highlights the existence of delays among some local authorities, with notable differences between municipalities with larger and smaller populations, in implementing the Act, and also the fact that many municipalities are committed to achieving transparency, often going beyond minimum legal requirements. Factors such as population size, the political ideology of the governing party and the education level and economic capacity of the population are related to the level of implementation of transparency and self-regulatory policies in this respect by local authorities. We acknowledge that the 2013 Act was still being implemented dur...

Autonomous driving systems require huge amounts of data to train. Manual annotation of this data is time-consuming and prohibitively expensive since it involves human resources. Therefore, active learning emerged as an alternative to ease this effort and to make data annotation more manageable. In this paper, we introduce a novel active learning approach for object detection in videos by exploiting temporal coherence. Our active learning criterion is based on the estimated number of errors in terms of false positives and false negatives. The detections obtained by the object detector are used to define the nodes of a graph and tracked forward and backward to temporally link the nodes. Minimizing an energy function defined on this graphical model provides estimates of both false positives and false negatives. Additionally, we introduce a synthetic video dataset, called SYNTHIA-AL, specially designed to evaluate active learning for video object detection in road scenes. Finally, we show that our approach outperforms active learning baselines tested on two datasets., Accepted at ICCVW 2019 (CVRSUAD-Road Scene Understanding and Autonomous Driving)

Driving requires reacting to a wide variety of complex environment conditions and agent behaviors. Explicitly modeling each possible scenario is unrealistic. In contrast, imitation learning can, in theory, leverage data from large fleets of human-driven cars. Behavior cloning in particular has been successfully used to learn simple visuomotor policies end-to-end, but scaling to the full spectrum of driving behaviors remains an unsolved problem. In this paper, we propose a new benchmark to experimentally investigate the scalability and limitations of behavior cloning. We show that behavior cloning leads to state-of-the-art results, executing complex lateral and longitudinal maneuvers, even in unseen environments, without being explicitly programmed to do so. However, we confirm some limitations of the behavior cloning approach: some well-known limitations (e.g., dataset bias and overfitting), new generalization issues (e.g., dynamic objects and the lack of a causal modeling), and training instabilities, all requiring further research before behavior cloning can graduate to real-world driving. The code, dataset, benchmark, and agent studied in this paper can be found at \url{github.com/felipecode/coiltraine/blob/master/docs/exploring_limitations.md}

One of the objectives of supreme audit institutions (SAIs) is to promote transparency and accountability of the public sector entities. But, at the same time, SAIs must improve their own transparency and accountability. The use of information and communication technologies plays an essential role in making this disclosure. The International Organization of the Supreme Audit Institution (INTOSAI) recognizes that communication is a strategic factor of SAIs. Studies have analyzed information dissemination practices carried out in developed countries. However, SAIs elsewhere, such as the Latin American and Caribbean, have not been the object of research. This chapter analyzes whether these SAIs are using the internet as a means to enhance transparency and interaction with stakeholders. The results indicate that, although in general the institutions analyzed publish the information required by INTOSAI, there is still ample room for improvement, especially regarding interaction with their stakeholders.

A crucial component of an autonomous vehicle (AV) is the artificial intelligence (AI) is able to drive towards a desired destination. Today, there are different paradigms addressing the development of AI drivers. On the one hand, we find modular pipelines, which divide the driving task into sub-tasks such as perception and maneuver planning and control. On the other hand, we find end-to-end driving approaches that try to learn a direct mapping from input raw sensor data to vehicle control signals. The later are relatively less studied, but are gaining popularity since they are less demanding in terms of sensor data annotation. This paper focuses on end-to-end autonomous driving. So far, most proposals relying on this paradigm assume RGB images as input sensor data. However, AVs will not be equipped only with cameras, but also with active sensors providing accurate depth information (e.g., LiDARs). Accordingly, this paper analyses whether combining RGB and depth modalities, i.e. using RGBD data, produces better end-to-end AI drivers than relying on a single modality. We consider multimodality based on early, mid and late fusion schemes, both in multisensory and single-sensor (monocular depth estimation) settings. Using the CARLA simulator and conditional imitation learning (CIL), we show how, indeed, early fusion multimodality outperforms single-modality., Comment: The paper has been accepted by IEEE Transactions on Intelligent Transportation Systems 2020

Prediction of walking turns allows to improve human factors such as comfort and perceived safety in human-robot interaction. The current state-of-the-art suggests that upper body kinematics can be used for that purpose and contains evidence about the reliability and the quantitative anticipation that can be expected from different variables. However, the experimental methodology has not been consistent throughout the different works and the related data has not always been given in an explicit form, with different studies containing partial, complementary or even contradictory results. In this paper, with the purpose of providing a uniform view of the topic that can trigger new developments in the field, we performed a systematic review of the relevant literature addressing three main questions: (i) Which upper body kinematic variables permit to anticipate a walking turn? (ii) How long in advance can we anticipate the turn from them? (iii) What is the expected contribution of walking turn prediction systems from upper body kinematics for human-robot interaction? We have found that head yaw was the most reliable kinematical variable from the upper body to predict walking turns about 200ms. Trunk roll anticipates walking turns by a similar amount of time, but with less reliability. Both approaches may benefit human-robot interaction in close proximity, helping the robot to exhibit appropriate proxemic behavior interacting at intimate, personal or social distances. From the point of view of safety, they have to be considered with caution. Trunk yaw is not valid to anticipate turns. Gaze Yaw seems to be the earliest predictor, although existing evidence is still inconclusive.

Rainfall is a problem in automated traffic surveillance. Rain streaks occlude the road users and degrade the overall visibility which in turn decrease object detection performance. One way of alleviating this is by artificially removing the rain from the images. This requires knowledge of corresponding rainy and rain-free images. Such images are often produced by overlaying synthetic rain on top of rain-free images. However, this method fails to incorporate the fact that rain fall in the entire three-dimensional volume of the scene. To overcome this, we introduce training data from the SYNTHIA virtual world that models rain streaks in the entirety of a scene. We train a conditional Generative Adversarial Network for rain removal and apply it on traffic surveillance images from SYNTHIA and the AAU RainSnow datasets. To measure the applicability of the rain-removed images in a traffic surveillance context, we run the YOLOv2 object detection algorithm on the original and rain-removed frames. The results on SYNTHIA show an 8% increase in detection accuracy compared to the original rain image. Interestingly, we find that high PSNR or SSIM scores do not imply good object detection performance.

This work presents and evaluates a novel compact scene representation based on Stixels that infers geometric and semantic information. Our approach overcomes the previous rather restrictive geometric assumptions for Stixels by introducing a novel depth model to account for non-flat roads and slanted objects. Both semantic and depth cues are used jointly to infer the scene representation in a sound global energy minimization formulation. Furthermore, a novel approximation scheme is introduced in order to significantly reduce the computational complexity of the Stixel algorithm, and then achieve real-time computation capabilities. The idea is to first perform an over-segmentation of the image, discarding the unlikely Stixel cuts, and apply the algorithm only on the remaining Stixel cuts. This work presents a novel over-segmentation strategy based on a fully convolutional network, which outperforms an approach based on using local extrema of the disparity map. We evaluate the proposed methods in terms of semantic and geometric accuracy as well as run-time on four publicly available benchmark datasets. Our approach maintains accuracy on flat road scene datasets while improving substantially on a novel non-flat road dataset., International Journal of Computer Vision, 127 (11-12), ISSN:0920-5691, ISSN:1573-1405

The cost of drawing object bounding boxes (i.e. labeling) for millions of images is prohibitively high. For instance, labeling pedestrians in a regular urban image could take 35 seconds on average. Active learning aims to reduce the cost of labeling by selecting only those images that are informative to improve the detection network accuracy. In this paper, we propose a method to perform active learning of object detectors based on convolutional neural networks. We propose a new image-level scoring process to rank unlabeled images for their automatic selection, which clearly outperforms classical scores. The proposed method can be applied to videos and sets of still images. In the former case, temporal selection rules can complement our scoring process. As a relevant use case, we extensively study the performance of our method on the task of pedestrian detection. Overall, the experiments show that the proposed method performs better than random selection. Our codes are publicly available at www.gitlab.com/haghdam/deep_active_learning., Comment: Accepted at ICCV 2019

Graphical abstractDisplay Omitted HighlightsWe present a new automatic approach for head yaw angle estimation of the driver.We rely on a set of geometric features computed from just three representative facial keypoints.The method has a confidence mechanism to decide the reliability of a sample label.The results are comparable to the state-of-the-art techniques.The method can be easily integrated in massive consume devices. Evaluation of driving performance is of utmost importance in order to reduce road accident rate. Since driving ability includes visual-spatial and operational attention, among others, head pose estimation of the driver is a crucial indicator of driving performance. This paper proposes a new automatic method for coarse and fine head's yaw angle estimation of the driver. We rely on a set of geometric features computed from just three representative facial keypoints, namely the center of the eyes and the nose tip. With these geometric features, our method combines two manifold embedding methods and a linear regression one. In addition, the method has a confidence mechanism to decide if the classification of a sample is not reliable. The approach has been tested using the CMU-PIE dataset and our own driver dataset. Despite the very few facial keypoints required, the results are comparable to the state-of-the-art techniques. The low computational cost of the method and its robustness makes feasible to integrate it in massive consume devices as a real time application.

Background and Purpose— Since demonstration of the benefit of endovascular treatment (EVT) in acute ischemic stroke patients with proximal arterial occlusion, stroke care systems need to be reorganized to deliver EVT in a timely and equitable way. We analyzed differences in the access to EVT by geographical areas in Catalonia, a territory with a highly decentralized stroke model. Methods— We studied 965 patients treated with EVT from a prospective multicenter population-based registry of stroke patients treated with reperfusion therapies in Catalonia, Spain (SONIIA). Three different areas were defined: (A) health areas primarily covered by Comprehensive Stroke Centers, (B) areas primarily covered by local stroke centers located less than hour away from a Comprehensive Stroke Center, and (C) areas primarily covered by local stroke centers located more than hour away from a Comprehensive Stroke Center. We compared the number of EVT×100 000 inhabitants/year and time from stroke onset to groin puncture between groups. Results— Baseline characteristics were similar between groups. Throughout the study period, there were significant differences in the population rates of EVT across geographical areas. EVT rates by 100 000 in 2015 were 10.5 in A area, 3.7 in B, and 2.7 in C. Time from symptom onset to groin puncture was 82 minutes longer in group B (312 minutes [245–435]) and 120 minutes longer in group C (350 minutes [284–408]) compared with group A (230 minutes [160–407]; P Conclusions— Accessibility to EVT from remote areas is hampered by lower rate and longer time to treatment compared with areas covered directly by Comprehensive Stroke Centers.

Falls in the elderly are a major health and economic problem. Hence, reliable fall detection is of great importance. Many fall detection systems have been developed but this topic is still under reach. In this work, a smart sensor based on a tri-axial digital accelerometer and a microcontroller Wi-Fi module are used to detect changes in acceleration that occur when a human being is falling. A novel real-time algorithm to distinguish between falls and daily activities is proposed. The prototype can easily connect to the internet of things thus expanding its applications. The experimental results have demonstrated high sensitivity and specificity on human fall detection.

Autonomous driving has harsh requirements of small model size and energy efficiency, in order to enable the embedded system to achieve real-time on-board object detection. Recent deep convolutional neural network based object detectors have achieved state-of-the-art accuracy. However, such models are trained with numerous parameters and their high computational costs and large storage prohibit the deployment to memory and computation resource limited systems. Low-precision neural networks are popular techniques for reducing the computation requirements and memory footprint. Among them, binary weight neural network (BWN) is the extreme case which quantizes the float-point into just $1$ bit. BWNs are difficult to train and suffer from accuracy deprecation due to the extreme low-bit representation. To address this problem, we propose a knowledge transfer (KT) method to aid the training of BWN using a full-precision teacher network. We built DarkNet- and MobileNet-based binary weight YOLO-v2 detectors and conduct experiments on KITTI benchmark for car, pedestrian and cyclist detection. The experimental results show that the proposed method maintains high detection accuracy while reducing the model size of DarkNet-YOLO from 257 MB to 8.8 MB and MobileNet-YOLO from 193 MB to 7.9 MB., Accepted by ICRA 2019

Depth estimation provides essential information to perform autonomous driving and driver assistance. Especially, Monocular Depth Estimation is interesting from a practical point of view, since using a single camera is cheaper than many other options and avoids the need for continuous calibration strategies as required by stereo-vision approaches. State-of-the-art methods for Monocular Depth Estimation are based on Convolutional Neural Networks (CNNs). A promising line of work consists of introducing additional semantic information about the traffic scene when training CNNs for depth estimation. In practice, this means that the depth data used for CNN training is complemented with images having pixel-wise semantic labels, which usually are difficult to annotate (e.g. crowded urban images). Moreover, so far it is common practice to assume that the same raw training data is associated with both types of ground truth, i.e., depth and semantic labels. The main contribution of this paper is to show that this hard constraint can be circumvented, i.e., that we can train CNNs for depth estimation by leveraging the depth and semantic information coming from heterogeneous datasets. In order to illustrate the benefits of our approach, we combine KITTI depth and Cityscapes semantic segmentation datasets, outperforming state-of-the-art results on Monocular Depth Estimation., Accepted in IEEE-Intelligent Vehicles Symposium, IV'2018

Autonomous driving models should ideally be evaluated by deploying them on a fleet of physical vehicles in the real world. Unfortunately, this approach is not practical for the vast majority of researchers. An attractive alternative is to evaluate models offline, on a pre-collected validation dataset with ground truth annotation. In this paper, we investigate the relation between various online and offline metrics for evaluation of autonomous driving models. We find that offline prediction error is not necessarily correlated with driving quality, and two models with identical prediction error can differ dramatically in their driving performance. We show that the correlation of offline evaluation with driving quality can be significantly improved by selecting an appropriate validation dataset and suitable offline metrics.

Environmental perception represents, because of its complexity, a challenge for Intelligent Transport Systems due to the great variety of situations and different elements that can happen in road environments and that must be faced by these systems. In connection with this, so far there are a variety of solutions as regards sensors and methods, so the results of precision, complexity, cost, or computational load obtained by these works are different. In this chapter some systems based on computer vision and laser techniques are presented. Fusion methods are also introduced in order to provide advanced and reliable perception systems.

In human motion science, accelerometers are used as linear distance sensors by attaching them to moving body parts, with their measurement axes its measurement axis aligned in the direction of motion. When double integrating the raw sensor data, multiple error sources are also integrated integrated as well, producing inaccuracies in the final position estimation which increases fast with the integration time. In this paper, we make a systematic and experimental comparison of different methods for position estimation, with different sensors and in different motion conditions. The objective is to correlate practical factors that appear in real applications, such as motion mean velocity, path length, calibration method, or accelerometer noise level, with the quality of the estimation. The results confirm that it is possible to use accelerometers to estimate short linear displacements of the body with a typical error of around 4.5% in the general conditions tested in this study. However, they also show that the motion kinematic conditions can be a key factor in the performance of this estimation, as the dynamic response of the accelerometer can affect the final results. The study lays out the basis for a better design of distance estimations, which are useful in a wide range of ambulatory human motion monitoring applications.

Action recognition in still images is a challenging problem in computer vision. To facilitate comparative evaluation independently of person detection, the standard evaluation protocol for action recognition uses an oracle person detector to obtain perfect bounding box information at both training and test time. The assumption is that, in practice, a general person detector will provide candidate bounding boxes for action recognition. In this paper, we argue that this paradigm is suboptimal and that action class labels should already be considered during the detection stage. Motivated by the observation that body pose is strongly conditioned on action class, we show that: 1) the existing state-of-the-art generic person detectors are not adequate for proposing candidate bounding boxes for action classification; 2) due to limited training examples, the direct training of action-specific person detectors is also inadequate; and 3) using only a small number of labeled action examples, the transfer learning is able to adapt an existing detector to propose higher quality bounding boxes for subsequent action classification. To the best of our knowledge, we are the first to investigate transfer learning for the task of action-specific person detection in still images. We perform extensive experiments on two benchmark data sets: 1) Stanford-40 and 2) PASCAL VOC 2012. For the action detection task (i.e., both person localization and classification of the action performed), our approach outperforms methods based on general person detection by 5.7% mean average precision (MAP) on Stanford-40 and 2.1% MAP on PASCAL VOC 2012. Our approach also significantly outperforms the state of the art with a MAP of 45.4% on Stanford-40 and 31.4% on PASCAL VOC 2012. We also evaluate our action detection approach for the task of action classification (i.e., recognizing actions without localizing them). For this task, our approach, without using any ground-truth person localization at test time, outperforms on both data sets state-of-the-art methods, which do use person locations.

Financial and political factors are widely considered to be an explanatory factor of the privatisation of public services. However, the empirical evidence in this respect is not convincing. This paper considers elements of long-term financial condition and ideological and political strength, together with other socio-economic factors and the effects of the current global financial crisis, to address the financial stress–privatisation relationship, with respect to urban water services, observed during the period 2002–2012 in Spanish municipalities. A discrete time survival model is applied, and the results obtained show that certain elements of financial condition, together with political and ideological factors, are decisive influences in decisions to privatise urban water services. We also highlight the major impact of the Great Recession and its effect on the financial stress–privatisation relationship.

Thermoelectric modules are currently used both in Peltier cooling and in Seebeck mode for electricity generation. The developments experienced in both cases depend essentially on two factors: the thermoelectric properties of the materials that form these elements (mainly semiconductors), and the external structure of the semiconductors. Figure of Merit Z is currently the best way of measuring the efficiency of semiconductors, as it relates to the intrinsic parameters of the semiconductor: Seebeck coefficient, thermal resistance, and thermal conductivity. When it comes to evaluating the complete structure, the Coefficient of Performance (COP) is used, relating the electrical power to the thermal power of the module. This paper develops a Thermoelectric Generator (TEG) structure which allows minimising the response time of the thermoelectric device, obtaining short working cycles and, therefore, a higher working frequency.

According to the conventional theoretical framework, fiscal stress is an explanatory factor of privatisation, since the latter can provide economies of scale and cost savings, as indicated by the theory of public choice. However, urban planning theories suggest that public choice does not take into account the collective needs of those receiving public services. The present study aims to clarify two major issues. Firstly, in the presence of fiscal stress, do public managers really privatise? And secondly, does privatisation harm the public interest? To investigate these questions, we constructed and analysed a discrete-time survival model, which was applied to the period 2000–2010, to reflect the effects of the current global financial crisis on the fiscal stress–privatisation relationship with respect to urban waste collection services. The results obtained indicate that when certain aspects of financial condition worsen, the likelihood of public services being privatised increases. This was the case during the Great Recession of 2008–2010, especially when a neighbouring municipality had previously privatised its services. After privatisation, service quality did not vary significantly, and so we conclude that public choice does not prejudice the public interest.

Background and Purpose— Computed tomographic (CT) high attenuation (HA) areas after endovascular therapy for acute ischemic stroke are a common finding indicative of blood–brain barrier disruption. Dual-energy CT allows an accurate differentiation between HA areas related to contrast staining (CS) or to brain hemorrhage (BH). We sought to evaluate the prognostic significance of the presence of CS and BH after endovascular therapy. Methods— A prospective cohort of 132 patients treated with endovascular therapy was analyzed. According to dual-energy CT findings, patients were classified into 3 groups: no HA areas (n=53), CS (n=32), and BH (n=47). The rate of new hemorrhagic transformations was recorded at follow-up neuroimaging. Clinical outcome was evaluated at 90 days with the modified Rankin Scale (poor outcome, 3–6). Results— Poor outcome was associated with the presence of CS (odds ratio [OR], 11.3; 95% confidence interval, 3.34–38.95) and BH (OR, 10.4; 95% confidence interval, 3.42–31.68). The rate of poor outcome despite complete recanalization was also significantly higher in CS (OR, 9.7; 95% confidence interval, 2.55–37.18) and BH (OR, 15.1; 95% confidence interval, 3.85–59.35) groups, compared with the no-HA group. Patients with CS disclosed a higher incidence of delayed hemorrhagic transformation at follow-up (OR, 4.5; 95% confidence interval, 1.22–16.37) compared with no-HA patients. Conclusions— Blood–brain barrier disruption, defined as CS and BH on dual-energy CT, was associated with poor clinical outcomes in patients with stroke treated with endovascular therapies. Moreover, isolated CS was associated with delayed hemorrhagic transformation. These results support the clinical relevance of blood–brain barrier disruption in acute stroke.

Deep networks trained on demonstrations of human driving have learned to follow roads and avoid obstacles. However, driving policies trained via imitation learning cannot be controlled at test time. A vehicle trained end-to-end to imitate an expert cannot be guided to take a specific turn at an upcoming intersection. This limits the utility of such systems. We propose to condition imitation learning on high-level command input. At test time, the learned driving policy functions as a chauffeur that handles sensorimotor coordination but continues to respond to navigational commands. We evaluate different architectures for conditional imitation learning in vision-based driving. We conduct experiments in realistic three-dimensional simulations of urban driving and on a 1/5 scale robotic truck that is trained to drive in a residential area. Both systems drive based on visual input yet remain responsive to high-level navigational commands. The supplementary video can be viewed at https://youtu.be/cFtnflNe5fM, Published at the International Conference on Robotics and Automation (ICRA), 2018

The implementation of e-Government could reduce administrative costs and the time devoted to repetitive tasks by civil servants, offering greater transparency to public administration, improving the current performance of public sector services, and expanding access to services. However, previous research indicates that there are significant barriers for citizens, assuming major obstacles. In this sense, the EU and, in particular, the Spanish government has formulated policies and legal frameworks to introduce in the field of the provision of public sector services in order to customize and to access these services. Therefore, the purpose of this chapter is to obtain a vision of government strategies adopted by Spanish regional governments to reduce the digital divide.

In this work we present a novel compact scene representation based on Stixels that infers geometric and semantic information. Our approach overcomes the previous rather restrictive geometric assumptions for Stixels by introducing a novel depth model to account for non-flat roads and slanted objects. Both semantic and depth cues are used jointly to infer the scene representation in a sound global energy minimization formulation. Furthermore, a novel approximation scheme is introduced that uses an extremely efficient over-segmentation. In doing so, the computational complexity of the Stixel inference algorithm is reduced significantly, achieving real-time computation capabilities with only a slight drop in accuracy. We evaluate the proposed approach in terms of semantic and geometric accuracy as well as run-time on four publicly available benchmark datasets. Our approach maintains accuracy on flat road scene datasets while improving substantially on a novel non-flat road dataset., Accepted to BMVC 2017 as oral presentation

Despite recent significant advances, object detection continues to be an extremely challenging problem in real scenarios. In order to develop a detector that successfully operates under these conditions, it becomes critical to leverage upon multiple cues, multiple imaging modalities, and a strong multiview (MV) classifier that accounts for different object views and poses. In this paper, we provide an extensive evaluation that gives insight into how each of these aspects (multicue, multimodality, and strong MV classifier) affect accuracy both individually and when integrated together. In the multimodality component, we explore the fusion of RGB and depth maps obtained by high-definition light detection and ranging, a type of modality that is starting to receive increasing attention. As our analysis reveals, although all the aforementioned aspects significantly help in improving the accuracy, the fusion of visible spectrum and depth information allows to boost the accuracy by a much larger margin. The resulting detector not only ranks among the top best performers in the challenging KITTI benchmark, but it is built upon very simple blocks that are easy to implement and computationally efficient.