Search

The utilization of inertial measurement units as wearable sensors is proliferating across various domains, such as health care, sports, and rehabilitation. This expansion has produced a market of devices tailored to accommodate very specific ranges of operational demands. Simultaneously, this growth is creating opportunities for the development of a new class of devices more oriented towards general-purpose use and capable of capturing both high-frequency signals for short-term, event-driven motion analysis and low-frequency signals for extended monitoring. For such a design, which combines flexibility and low cost, a rigorous evaluation of the device in terms of deviation, noise levels, and precision is essential. This evaluation is crucial for identifying potential improvements and refining the design accordingly, yet it is rarely addressed in the literature. This paper presents the development process of such a device. The results of the design process demonstrate acceptable performance in optimizing energy consumption and storage capacity while highlighting the most critical optimizations needed to advance the device towards the goal of a smart, general-purpose unit for human motion monitoring. [ABSTRACT FROM AUTHOR]

This study presents a novel method based on ray tracing for analyzing wave propagation in composites specifically tailored for structural health monitoring applications. This method offers distinct advantages over the commonly used finite element method mainly in computational resource utilization, which has become a limiting factor for these kinds of analyses. The ray tracing method is evaluated against a number of example cases representing structural details such as thickness changes, stringers, or simulated damage, and the significance of ray tracing to study wave propagation under these conditions and how it can serve as a valuable tool for structural health monitoring are highlighted. This model has been developed as part of a complete SHM framework with the intention of being an efficient and simple way to calculate wave propagation and therefore it could be used as a way to determine relevant damage indicators or train an artificial intelligence model. [ABSTRACT FROM AUTHOR]

Despite several existing techniques for distributed sensing (temperature and strain) using standard Single-Mode optical Fiber (SMF), compensating or decoupling both effects is mandatory for many applications. Currently, most decoupling techniques require special optical fibers and are difficult to implement with high-spatial-resolution distributed techniques, such as OFDR. Therefore, this work's objective is to study the feasibility of decoupling temperature and strain out of the readouts of a phase and polarization analyzer OFDR (ϕ -PA-OFDR) taken over an SMF. For this purpose, the readouts will be subjected to a study using several machine learning algorithms, among them Deep Neural Networks. The motivation that underlies this target is the current blockage in the widespread use of Fiber Optic Sensors in situations where both strain and temperature change, due to the coupled dependence of currently developed sensing methods. Instead of using other types of sensors or even other interrogation methods, the objective of this work is to analyze the available information in order to develop a sensing method capable of providing information about strain and temperature simultaneously. [ABSTRACT FROM AUTHOR]

Ground contact time (GCT) is one of the most relevant factors when assessing running performance in sports practice. In recent years, inertial measurement units (IMUs) have been widely used to automatically evaluate GCT, since they can be used in field conditions and are friendly and easy to wear devices. In this paper we describe the results of a systematic search, using the Web of Science, to assess what reliable options are available to GCT estimation using inertial sensors. Our analysis reveals that estimation of GCT from the upper body (upper back and upper arm) has rarely been addressed. Proper estimation of GCT from these locations could permit an extension of the analysis of running performance to the public, where users, especially vocational runners, usually wear pockets that are ideal to hold sensing devices fitted with inertial sensors (or even using their own cell phones for that purpose). Therefore, in the second part of the paper, an experimental study is described. Six subjects, both amateur and semi-elite runners, were recruited for the experiments, and ran on a treadmill at different paces to estimate GCT from inertial sensors placed at the foot (for validation purposes), the upper arm, and upper back. Initial and final foot contact events were identified in these signals to estimate the GCT per step, and compared to times estimated from an optical MOCAP (Optitrack), used as the ground truth. We found an average error in GCT estimation of 0.01 s in absolute value using the foot and the upper back IMU, and of 0.05 s using the upper arm IMU. Limits of agreement (LoA, 1.96 times the standard deviation) were [−0.01 s, 0.04 s], [−0.04 s, 0.02 s], and [0.0 s, 0.1 s] using the sensors on the foot, the upper back, and the upper arm, respectively. [ABSTRACT FROM AUTHOR]

Semantic image segmentation is a core task for autonomous driving, which is performed by deep models. Since training these models draws to a curse of human-based image labeling, the use of synthetic images with automatically generated labels together with unlabeled real-world images is a promising alternative. This implies addressing an unsupervised domain adaptation (UDA) problem. In this paper, we propose a new co-training procedure for synth-to-real UDA of semantic segmentation models. It performs iterations where the (unlabeled) real-world training images are labeled by intermediate deep models trained with both the (labeled) synthetic images and the real-world ones labeled in previous iterations. More specifically, a self-training stage provides two domain-adapted models and a model collaboration loop allows the mutual improvement of these two models. The final semantic segmentation labels (pseudo-labels) for the real-world images are provided by these two models. The overall procedure treats the deep models as black boxes and drives their collaboration at the level of pseudo-labeled target images, i.e., neither modifying loss functions is required, nor explicit feature alignment. We test our proposal on standard synthetic and real-world datasets for onboard semantic segmentation. Our procedure shows improvements ranging from approximately 13 to 31 mIoU points over baselines. [ABSTRACT FROM AUTHOR]

Non-contact vibration measurements are relevant for non-invasively characterizing the mechanical behavior of structures. This paper presents a novel methodology for full-field vibrational analysis at high frequencies using the three-dimensional digital image correlation technique combined with the projection of a speckle pattern. The method includes stereo calibration and image processing routines for accurate three-dimensional data acquisition. Quantitative analysis allows the extraction of several deformation parameters, such as the cross-correlation coefficients, shape and intensity, as well as the out-of-plane displacement fields and mode shapes. The potential of the methodology is demonstrated on an Unmanned Aerial Vehicle wing made of composite material, followed by experimental validation with reference accelerometers. The results obtained with the projected three-dimensional digital image correlation show a percentage of error below 5% compared with the measures of accelerometers, achieving, therefore, high sensitivity to detect the dynamic modes in structures made of composite material. [ABSTRACT FROM AUTHOR]

The short-term prediction of a person's trajectory during normal walking becomes necessary in many environments shared by humans and robots. Physics-based approaches based on Newton's laws of motion seem best suited for short-term predictions, but the intrinsic properties of human walking conflict with the foundations of the basic kinematical models compromising their performance. In this paper, we propose a short-time prediction method based on gait biomechanics for real-time applications. This method relays on a single biomechanical variable, and it has a low computational burden, turning it into a feasible solution to implement in low-cost portable devices. We evaluate its performance from an experimental benchmark where several subjects walked steadily over straight and curved paths. With this approach, the results indicate a performance good enough to be applicable to a wide range of human–robot interaction applications. [ABSTRACT FROM AUTHOR]

Fiber-optic sensors cannot measure damage; to get information about damage from strain measurements, additional strategies are needed, and several alternatives are available in the existing literature. This paper discusses two independent procedures. The first is based on detecting new strains appearing around a damage spot. The structure does not need to be under loads, the technique is very robust, and damage detectability is high, but it requires sensors to be located very close to the damage, so it is a local technique. The second approach offers wider coverage of the structure; it is based on identifying the changes caused by damage on the strain field in the whole structure for similar external loads. Damage location does not need to be known a priori, and detectability is dependent upon the sensor’s network density, the damage size, and the external loads. Examples of application to real structures are given. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]

Despite all the significant advances in pedestrian detection brought by computer vision for driving assistance, it is still a challenging problem. One reason is the extremely varying lighting conditions under which such a detector should operate, namely day and nighttime. Recent research has shown that the combination of visible and non-visible imaging modalities may increase detection accuracy, where the infrared spectrum plays a critical role. The goal of this paper is to assess the accuracy gain of different pedestrian models (holistic, part-based, patch-based) when training with images in the far infrared spectrum. Specifically, we want to compare detection accuracy on test images recorded at day and nighttime if trained (and tested) using (a) plain color images; (b) just infrared images; and (c) both of them. In order to obtain results for the last item, we propose an early fusion approach to combine features from both modalities. We base the evaluation on a new dataset that we have built for this purpose as well as on the publicly available KAIST multispectral dataset. [ABSTRACT FROM AUTHOR]

In the context of human–robot collaborative shared environments, there has been an increase in the use of optical motion capture (OMC) systems for human motion tracking. The accuracy and precision of OMC technology need to be assessed in order to ensure safe human–robot interactions, but the accuracy specifications provided by manufacturers are easily influenced by various factors affecting the measurements. This article describes a new methodology for the metrological evaluation of a human–robot collaborative environment based on optical motion capture (OMC) systems. Inspired by the ASTM E3064 test guide, and taking advantage of an existing industrial robot in the production cell, the system is evaluated for mean error, error spread, and repeatability. A detailed statistical study of the error distribution across the capture area is carried out, supported by a Mann–Whitney U-test for median comparisons. Based on the results, optimal capture areas for the use of the capture system are suggested. The results of the proposed method show that the metrological characteristics obtained are compatible and comparable in quality to other methods that do not require the intervention of an industrial robot. [ABSTRACT FROM AUTHOR]

Top-performing computer vision models are powered by convolutional neural networks (CNNs). Training an accurate CNN highly depends on both the raw sensor data and their associated ground truth (GT). Collecting such GT is usually done through human labeling, which is time-consuming and does not scale as we wish. This data-labeling bottleneck may be intensified due to domain shifts among image sensors, which could force per-sensor data labeling. In this paper, we focus on the use of co-training, a semi-supervised learning (SSL) method, for obtaining self-labeled object bounding boxes (BBs), i.e., the GT to train deep object detectors. In particular, we assess the goodness of multi-modal co-training by relying on two different views of an image, namely, appearance (RGB) and estimated depth (D). Moreover, we compare appearance-based single-modal co-training with multi-modal. Our results suggest that in a standard SSL setting (no domain shift, a few human-labeled data) and under virtual-to-real domain shift (many virtual-world labeled data, no human-labeled data) multi-modal co-training outperforms single-modal. In the latter case, by performing GAN-based domain translation both co-training modalities are on par, at least when using an off-the-shelf depth estimation model not specifically trained on the translated images. [ABSTRACT FROM AUTHOR]

The development of deep learning has achieved great success in object detection, but small object detection is still a difficult and challenging task in computer vision. To address the problem, we propose an improved single-shot multibox detector (SSD) using enhanced feature map blocks (SSD-EMB). The enhanced feature map block (EMB) consists of attention stream and feature map concatenation stream. The attention stream allows the proposed model to focus on the object regions rather than background owing to channel averaging and the effectiveness of the normalization. The feature map concatenation stream provides additional semantic information to the model without degrading the detection speed. By combining the output of these two streams, the enhanced feature map, which improves the detection of a small object, is generated. Experimental results show that the proposed model has high accuracy in small object detection. The proposed model not only achieves good detection accuracy, but also has a good detection speed. The SSD-EMB achieved a mean average precision (mAP) of 80.4% on the PASCAL VOC 2007 dataset at 30 frames per second on an RTX 2080Ti graphics processing unit, an mAP of 79.9% on the VOC 2012 dataset, and an mAP of 26.6% on the MS COCO dataset. [ABSTRACT FROM AUTHOR]

Accurate segmentation of entity categories is the critical step for 3D scene understanding. This paper presents a fast deep neural network model with Dense Conditional Random Field (DCRF) as a post-processing method, which can perform accurate semantic segmentation for 3D point cloud scene. On this basis, a compact but flexible framework is introduced for performing segmentation to the semantics of point clouds concurrently, contribute to more precise segmentation. Moreover, based on semantics labels, a novel DCRF model is elaborated to refine the result of segmentation. Besides, without any sacrifice to accuracy, we apply optimization to the original data of the point cloud, allowing the network to handle fewer data. In the experiment, our proposed method is conducted comprehensively through four evaluation indicators, proving the superiority of our method. [ABSTRACT FROM AUTHOR]

Particle size is the most important index to reflect the crushing quality of ores, and the accuracy of particle size statistics directly affects the subsequent operation of mines. Accurate ore image segmentation is an important prerequisite to ensure the reliability of particle size statistics. However, given the diversity of the size and shape of ores, the influence of dust and light, the complex texture and shadows on the ore surface, and especially the adhesion between ores, it is difficult to segment ore images accurately, and under-segmentation can be a serious problem. The construction of a large, labeled dataset for complex and unclear conveyor belt ore images is also difficult. In response to these challenges, we propose a novel, multi-task learning network based on U-Net for ore image segmentation. To solve the problem of limited available training datasets and to improve the feature extraction ability of the model, an improved encoder based on Resnet18 is proposed. Different from the original U-Net, our model decoder includes a boundary subnetwork for boundary detection and a mask subnetwork for mask segmentation, and information of the two subnetworks is fused in a boundary mask fusion block (BMFB). The experimental results showed that the pixel accuracy, Intersection over Union (IOU) for the ore mask (IOU_M), IOU for the ore boundary (IOU_B), and error of the average statistical ore particle size (ASE) rate of our proposed model on the testing dataset were 92.07%, 86.95%, 52.32%, and 20.38%, respectively. Compared to the benchmark U-Net, the improvements were 0.65%, 1.01%, 5.78%, and 12.11% (down), respectively. [ABSTRACT FROM AUTHOR]

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 ∼ 1 / 4 for new/known classes; even for more challenging ratios such as ∼ 4 / 1 , the results are also very positive. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]