Search

In this study, we propose a new appointment window scheduling (AWS) approach of informing customers of an admission window (AW) rather than the traditional appointment time. We provide a formal des...

With the development of deep learning, the performance of many computer vision tasks has been greatly improved. For visual tracking, deep learning methods mainly focus on extracting better features or designing end-to-end trackers. However, during tracking specific targets most of the existing trackers based on deep learning are less discriminative and time-consuming. In this paper, a cascade based tracking algorithm is proposed to promote the robustness of the tracker and reduce time consumption. First, we propose a novel deep network for feature extraction. Since some pruning strategies are applied, the speed of the feature extraction stage can be more than 50 frames per second. Then, a cascade tracker named DCCT is presented to improve the performance and enhance the robustness by utilizing both texture and semantic features. Similar to the cascade classifier, the proposed DCCT tracker consists of several weaker trackers. Each weak tracker rejects some false candidates of the tracked object, and the final tracking results are obtained by synthesizing these weak trackers. Intensive experiments are conducted in some public datasets and the results have demonstrated the effectiveness of the proposed framework.

Forward Vehicle Collision Warning (FCW) is one of the most important functions for the Advanced Driver Assistance System (ADAS). In this procedure, vehicle detection and distance measurement are core components, requiring accurate localization and estimation. In this paper, we propose a simple but efficient forward vehicle collision warning framework by aggregating monocular distance measurement and precise vehicle detection. In order to obtain forward vehicle distance, a quick camera calibration method which only needs three physical points to calibrate related camera parameters is utilized. As for the forward vehicle detection, a multi-scale detection algorithm that regards the result of calibration as distance prior is proposed to improve the precision. What’s more, traditional deterministic FCW approaches cannot be personalized for different drivers, which will lead to false warnings when drivers are in diverse driving status. Therefore, abnormal driver behaviors are introduced to make FCW adaptive. Specifically, the proposed adaptive FCW generates warnings by considering the different behaviors of the driver. Intensive experiments are conducted in our established real scene dataset and the results have demonstrated the effectiveness of the proposed framework.

Knowledge map is a large-scale semantic network that represents knowledge in the form of triples (entity, relationship, entity). It is widely used in information search, question answering system, e-commerce, recommendation system and other fields. In order to alleviate the problem of data sparsity in knowledge map with large amount of data, knowledge map representation learning represents entities and relationships as low dimensional real valued vectors, which effectively reduces the computational complexity and improves the efficiency of knowledge reasoning and knowledge map completion. In order to solve the problem that DKRL does not consider the relationship between sentences, which leads to the serious loss of entity description information, this paper proposes a new model T-CGRU. The model takes advantage of recurrent neural network, which can capture the context information of entity description. Convolution and maximum pooling are used to obtain the feature of entity description, and then the context word order of entity description information is extracted by cyclic operation, and finally the semantic representation of entity description information is obtained.

The real-time process fault detection in the multi-station assembly process is always a challenging problem for auto body manufactures. Traditionally, the fault diagnosis approaches for variation source identification are divided into two categories, i.e. the pattern matching methods and model-based estimation ones based on the collected data set. The measurements provide effective process monitoring, but the real-time process fault diagnosis in the assembly process is still difficult with the traditional diagnosis techniques, and always depends on the engineering experience in practice. Based on the assembly process knowledge, including multi-station assembly hierarchy, fixture scheme, measurement characteristics and tolerances etc. in the multi-station, a knowledge-based diagnostic methodology and procedures are proposed with the measurements of each body in white for part/component defections and faulty assembly station identification. For the station involved with defective parts/components, the sub-coordinate system of the part/component is established reflecting its position and pose in the space, and then the relative pose matrix to the “normally build” pose is calculated based on the deviations of sub-coordinates of the parts in this station. Finally, the assembly process malfunctions are determined by a proposed rule-based strategy with the relative pose matrix in real time. A simple 3 stations assembly process with 5 sheet metal parts was analyzed and compared with the traditional diagnostic method to verify the effectiveness and stability of the proposed method.

In this paper, a novel multi-frame image super-resolution algorithm for small amount of data is proposed. Our method solve the problem that the spatial resolution of the reconstructed image is low and the visual quality of it is poor when the number of input low-resolution images is small. In order to improve the quality of the initial estimation, we construct the initial estimation with multi-frame low-resolution images according to the registration parameter and interpolate the missing pixels by directional Gaussian-like filtering. In order to solve the problem of fuzzy initial estimation, the enhancement method is used to highlight the image details. A large number of qualitative and quantitative evaluation results show that our method has strong reconstruction performance for various types of low-resolution images under different amount of data.

Infrared maritime target detection is the key technology of maritime target search systems. However, infrared images generally have the defects of low signal-to-noise ratio and low resolution. At the same time, the maritime environment is complicated and changeable. Under the interference of islands, waves and other disturbances, the brightness of small dim targets is easily obscured, which makes them difficult to distinguish. This is difficult for traditional target detection algorithms to deal with. In order to solve these problems, through the analysis of infrared maritime images under a variety of sea conditions including small dim targets, this paper concludes that in infrared maritime images, small targets occupy very few pixels, often do not have any edge contour information, and the gray value and contrast values are very low. The background such as island and strong sea wave occupies a large number of pixels, with obvious texture features, and often has a high gray value. By deeply analyzing the difference between the target and the background, this paper proposes a detection algorithm (SRGM) for infrared small dim targets under different maritime background. Firstly, this algorithm proposes an efficient maritime background filter for the common background in the infrared maritime image. Firstly, the median filter based on the sensitive region selection is used to extract the image background accurately, and then the background is eliminated by image difference with the original image. In addition, this article analyzes the differences in gradient features between strong interference caused by the background and targets, proposes a small dim target extraction operator with two analysis factors that fit the target features perfectly and combines the adaptive threshold segmentation to realize the accurate extraction of the small dim target. The experimental results show that compared with the current popular small dim target detection algorithms, this paper has better performance for target detection in various maritime environments.

This paper provides a formal framework for optimal dynamic appointment scheduling with wait-dependent abandonment in order to reduce the blind wait (wait without queue information) and useless wait (wait but abandon). A discrete-time Markov decision process model is proposed to best balance server utility and experiences of customers waiting for admission. The properties of the optimal policy are investigated both theoretically and numerically. Efficient policies are implemented, the benefit from instituting the appointment system is checked numerically and managerial insights derived. In particular, appointment scheduling is shown to significantly reduce blind and useless wait without sacrificing the server utility when the system is overloaded with long service time and customers are more sensitive to blind wait than known wait (wait with a target). A revenue-oriented system tends to inform earlier appointment times and hence higher effective admission rate than a system taking into account customer experience of wait.

Forward Vehicle Collision Warning (FCW) is one of the most important functions for autonomous vehicles. In this procedure, vehicle detection and distance measurement are core components, requiring accurate localization and estimation. In this paper, we propose a simple but efficient forward vehicle collision warning framework by aggregating monocular distance measurement and precise vehicle detection. In order to obtain forward vehicle distance, a quick camera calibration method which only needs three physical points to calibrate related camera parameters is utilized. As for the forward vehicle detection, a multi-scale detection algorithm that regards the result of calibration as distance priori is proposed to improve the precision. Intensive experiments are conducted in our established real scene dataset and the results have demonstrated the effectiveness of the proposed framework.

Video-based vehicle detection and tracking is one of the most important components for Intelligent Transportation Systems (ITS). When it comes to road junctions, the problem becomes even more difficult due to the occlusions and complex interactions among vehicles. In order to get a precise detection and tracking result, in this work we propose a novel tracking-by-detection framework. In the detection stage, we present a sequential detection model to deal with serious occlusions. In the tracking stage, we model group behavior to treat complex interactions with overlaps and ambiguities. The main contributions of this paper are twofold: 1) Shape prior is exploited in the sequential detection model to tackle occlusions in crowded scene. 2) Traffic force is defined in the traffic scene to model group behavior, and it can assist to handle complex interactions among vehicles. We evaluate the proposed approach on real surveillance videos at road junctions and the performance has demonstrated the effectiveness of our method.

Identification of multicomponent in traditional Chinese medicine (TCM) is complex and time-consuming. The inspection of the full-scan mass chromatograms was usually performed manually, which is labor-intensive. It is difficult to distinguish low response signals from complex chemical background. Furthermore, this process is typically based on earlier knowledge of the chemical composition of TCM, and those molecules that have not been characterized earlier were thus ignored. In this paper, a strategy using UPLC-MS combined with pattern recognition analysis was developed to simplify and quicken the identification of multicomponent in Abelmoschus manihot (L.) Medik. First, complex signals obtained by UPLC-MS were processed using automated data mining algorithm and further processed with multivariate chemometric methods. Multicomponent in Abelmoschus manihot L. can be clearly displayed in S- and VIP-plot. Using this method, 320 peaks which present in Abelmoschus manihot L. were detected. In the next step, accurate mass spectra of the characteristic markers acquired by QTOF MS were used to estimate their elemental formulae and enable structure identification. By searching in METLIN database, 41 components were tentatively identified in Abelmoschus manihot L. Our results showed that UPLC-MS based-pattern recognition analysis approach can be used to quickly identify TCM multicomponent and for standardization of herbal preparations.

To develop a non-polluting and sustainable city, urban administrators encourage logistics companies to use electric vehicles instead of conventional (i.e., fuel-based) vehicles for transportation services. However, electric energy-based limitations pose a new challenge in designing reasonable visiting routes that are essential for the daily operations of companies. Therefore, this paper investigates a real-world electric vehicle routing problem (VRP) raised by a logistics company. The problem combines the features of the capacitated VRP, the VRP with time windows, the heterogeneous fleet VRP, the multi-trip VRP, and the electric VRP with charging stations. To solve such a complicated problem, a heuristic approach based on the adaptive large neighborhood search (ALNS) and integer programming is proposed in this paper. Specifically, a charging station adjustment heuristic and a departure time adjustment heuristic are devised to decrease the total operational cost. Furthermore, the best solution obtained by the ALNS is improved by integer programming. Twenty instances generated from real-world data were used to validate the effectiveness of the proposed algorithm. The results demonstrate that using our algorithm can save 7.52% of operational cost.

International audience; This paper investigates the improvement of the hospitalisation admission process of elective inpatients by delay announcement to free patients from blind and long wait, a phenomenon often observed in top Chinese hospitals. We propose a formal simulation model in which relevant resource requirements during the hospitalisation of a patient are represented by an original ‘multistage treatment model’. Other key components of the formal model include impatience of patients, resource capacities and key performance indicators such as the original concept of waiting time visibility. Through sensitivity analysis of the simulation model, we demonstrate the benefits of delay announcement and the value of our ‘multistage treatment model’.

Spectral fault localization (SFL) is one automatic fault-localization technique, which uses ranking metric to rank the risk of fault existence in each program entity after dynamically collecting the testing information. The effectiveness evaluation and comparison of ranking metrics are two important research problems. In this paper, we provide a uniformly theoretical investigation framework on longitudinally evaluating ranking metrics and horizontally comparing them for SFL techniques under any single fault scenario. We propose a generic vector table model as a novel device of thoroughly understanding various SFL techniques. By investigating rankings' mathematical formula of statements in the vector table model, the performance of different SFL techniques could be systematically analysed and compared. Under table model-driven evaluation framework, seven typical metrics as examples are explored, the existing equivalent group is extended, and the new relation of two equivalent groups is found. Our framework overcomes limitations of current empirical and theoretical approaches, and can theoretically evaluate the advantage and disadvantage of a SFL technique and compare different SFL techniques.

This paper investigates the improvement of the hospitalization admission process of elective inpatients by offering delay announcement to free patients from blind and long wait. We propose a Markov Decision Process (MDP) model for optimization of delay announcement strategies by taking into account patients' behavior with respect to waiting for hospitalization. Numerical experiments are conducted to study the hospital dominant case in which the delay announcement only tries to maximize the revenue of the hospital. Properties of the optimal delay announcement strategies were observed and conjectured.