Search results

The conventional pedestrian detection algorithms lack in scale sensitivity. The purpose of this paper is to propose a novel algorithm of self-adaptive scale pedestrian detection, based on deep residual network (DRN), to address such lacks.

First, the “Edge boxes” algorithm is introduced to extract region of interests from pedestrian images. Then, the extracted bounding boxes are incorporated to different DRNs, one is a large-scale DRN and the other one is the small-scale DRN. The height of the bounding boxes is used to classify the results of pedestrians and to regress the bounding boxes to the entity of the pedestrian. At last, a weighted self-adaptive scale function, which combines the large-scale results and small-scale results, is designed for the final pedestrian detection.

To validate the effectiveness and feasibility of the proposed algorithm, some comparison experiments have been done on the common pedestrian detection data sets: Caltech, INRIA, ETH and KITTI. Experimental results show that the proposed algorithm is adapted for the various scales of the pedestrians. For the hard detected small-scale pedestrians, the proposed algorithm has improved the accuracy and robustness of detections.

By applying different models to deal with different scales of pedestrians, the proposed algorithm with the weighted calculation function has improved the accuracy and robustness for different scales of pedestrians.

This paper aims to present a system that is fully capable of addressing the issue of detection, tracking and following pedestrians, which is a very challenging task, especially when it is considered for using in large outdoors infrastructures. Three modules, detection, tracking and following, are integrated and tested over long distances in semi-structured scenarios, where static or dynamic obstacles, including other pedestrians, can be found.

The detection is based on the probabilistic fusion of a laser scanner and a camera. The tracking module pairs observations with previously detected targets by using Kalman Filters and a Mahalanobis-distance. The following module allows to safely pursue the target by using a well-defined navigation scheme.

The system can track pedestrians from static position to 3.46 m/s (running). It handles occlusions, crossings or miss-detections, keeping track of the position even if the pedestrian is only detected in 55/per cent of the observations. Moreover, it autonomously selects and follows a target at a maximum speed of 1.46 m/s.

The main novelty of this study is the integration of the three algorithms in a fully operational system, tested in real outdoor scenarios. Furthermore, the addition of labelling to the detection algorithm allows using the full range of a single sensor while preserving the high performance of a combined detection. False-positives’ rate is reduced by handling the uncertainty level when pairing observations. The inclusion of pedestrian speed in the model speeds up and simplifies tracking process. Finally, the most suitable target is automatically selected by a scoring system.

The purpose of the study is to address the problems of low accuracy and missed detection of occluded pedestrians and small target pedestrians when using the YOLOX general object detection algorithm for pedestrian detection. This study proposes a multi-level fine-grained YOLOX pedestrian detection algorithm.

First, to address the problem of the original YOLOX algorithm in obtaining a single perceptual field for the feature map before feature fusion, this study improves the PAFPN structure by adding the ResCoT module to increase the diversity of the perceptual field of the feature map and divides the pedestrian multi-scale features into finer granularity. Second, for the CSPLayer of the PAFPN, a weight gain-based normalization-based attention module (NAM) is proposed to make the model pay more attention to the context information when extracting pedestrian features and highlight the salient features of pedestrians. Finally, the authors experimentally determined the optimal values for the confidence loss function.

The experimental results show that, compared with the original YOLOX algorithm, the AP of the improved algorithm increased by 2.90%, the Recall increased by 3.57%, and F1 increased by 2% on the pedestrian dataset.

The multi-level fine-grained YOLOX pedestrian detection algorithm can effectively improve the detection of occluded pedestrians and small target pedestrians.

The authors introduce a multi-level fine-grained ResCoT module and a weight gain-based NAM attention module.

The purpose of this paper is to relate to the real-time navigation and tracking of pedestrians in a closed environment. To restrain accumulated error of low-cost microelectromechanical system inertial navigation system and adapt to the real-time navigation of pedestrians at different speeds, the authors proposed an improved inertial navigation system (INS)/pedestrian dead reckoning (PDR)/ultra wideband (UWB) integrated positioning method for indoor foot-mounted pedestrians.

This paper proposes a self-adaptive integrated positioning algorithm that can recognize multi-gait and realize a high accurate pedestrian multi-gait indoor positioning. First, the corresponding gait method is used to detect different gaits of pedestrians at different velocities; second, the INS/PDR/UWB integrated system is used to get the positioning information. Thus, the INS/UWB integrated system is used when the pedestrian moves at normal speed; the PDR/UWB integrated system is used when the pedestrian moves at rapid speed. Finally, the adaptive Kalman filter correction method is adopted to modify system errors and improve the positioning performance of integrated system.

The algorithm presented in this paper improves performance of indoor pedestrian integrated positioning system from three aspects: in the view of different pedestrian gaits at different speeds, the zero velocity detection and stride frequency detection are adopted on the integrated positioning system. Further, the accuracy of inertial positioning systems can be improved; the attitude fusion filter is used to obtain the optimal quaternion and improve the accuracy of INS positioning system and PDR positioning system; because of the errors of adaptive integrated positioning system, the adaptive filter is proposed to correct errors and improve integrated positioning accuracy and stability. The adaptive filtering algorithm can effectively restrain the divergence problem caused by outliers. Compared to the KF algorithm, AKF algorithm can better improve the fault tolerance and precision of integrated positioning system.

The INS/PDR/UWB integrated system is built to track pedestrian position and attitude. Finally, an adaptive Kalman filter is used to improve the accuracy and stability of integrated positioning system.

Vehicles estimation can be used in evaluating traffic conditions and facilitating traffic control, which is an important task in intelligent transportation system. The paper aims to propose a vehicle-counting method based on the analysis of surveillance videos.

The paper proposes a novel two-step method using low-rank representation (LRR) detection and locality-constrained linear coding (LLC) classification to count the number of vehicles in traffic video sequences automatically. The proposed method is based on an offline training to understand an LLC-based classifier with extracted features for vehicle and pedestrian classification, followed by an online counting algorithm to count the number of vehicles detected from the image sequence.

The proposed method allows delivery estimation (counting the number of vehicles at each frame only) and total number estimation of vehicles shown in the scene. The paper compares the proposed method with other similar methods on three public data sets. The experimental results show that the proposed method is competitive and effective in terms of computational speed and evaluation accuracy.

The proposed method does not consider illumination. Hence, the results might be unsatisfactory under low-lighting condition. Therefore, researchers are encouraged to add a term that controls the illumination changes into the energy function of vehicle detection in future work.

The paper bridges the gap between LRR detection and vehicle counting by taking advantage of existing LLC classification algorithm to distinguish different moving objects.

This study aims to design an optimized algorithm for low-cost pedestrian navigation system (PNS) to correct the heading drift and altitude error, thus achieving high-precise pedestrian location in both two-dimensional (2-D) and three-dimensional (3-D) space.

A novel heading correction algorithm based on smoothing filter at the terminal of zero velocity interval (ZVI) is proposed in the paper. This algorithm adopts the magnetic sensor to calculate all the heading angles in the ZVI and then applies a smoothing filter to obtain the optimal heading angle. Furthermore, heading correction is executed at the terminal moment of ZVI. Meanwhile, an altitude correction algorithm based on step height constraint is proposed to suppress the altitude channel divergence of strapdown inertial navigation system by using the step height as the measurement of the Kalman filter.

The verification experiments were carried out in 2-D and 3-D space to evaluate the performance of the proposed pedestrian navigation algorithm. The results show that the heading drift and altitude error were well corrected. Meanwhile, the path calculated by the novel algorithm has a higher match degree with the reference trajectory, and the positioning errors of the 2-D and 3-D trajectories are both less than 0.5 per cent.

Besides zero velocity update, another two problems, namely, heading drift and altitude error in the PNS, are solved, which ensures the high positioning precision of pedestrian in indoor and outdoor environments.

The purpose of this paper is to summarize the existing point cloud target detection algorithms based on deep learning, and provide reference for researchers in related fields. In recent years, with its outstanding performance in target detection of 2D images, deep learning technology has been applied in light detection and ranging (LiDAR) point cloud data to improve the automation and intelligence level of target detection. However, there are still some difficulties and room for improvement in target detection from the 3D point cloud. In this paper, the vehicle LiDAR target detection method is chosen as the research subject.

Firstly, the challenges of applying deep learning to point cloud target detection are described; secondly, solutions in relevant research are combed in response to the above challenges. The currently popular target detection methods are classified, among which some are compared with illustrate advantages and disadvantages. Moreover, approaches to improve the accuracy of network target detection are introduced.

Finally, this paper also summarizes the shortcomings of existing methods and signals the prospective development trend.

This paper introduces some existing point cloud target detection methods based on deep learning, which can be applied to a driverless, digital map, traffic monitoring and other fields, and provides a reference for researchers in related fields.

One of the most neglected sources of energy loss is streetlights that generate too much light in areas where it is not required. Energy waste has enormous economic and environmental effects. In addition, due to the conventional manual nature of operation, streetlights are frequently seen being turned ‘ON’ during the day and ‘OFF’ in the evening, which is regrettable even in the twenty-first century. These issues require automated streetlight control in order to be resolved. This study aims to develop a novel streetlight controlling method by combining a smart transport monitoring system powered by computer vision technology with a closed circuit television (CCTV) camera that allows the light-emitting diode (LED) streetlight to automatically light up with the appropriate brightness by detecting the presence of pedestrians or vehicles and dimming the streetlight in their absence using semantic image segmentation from the CCTV video streaming. Consequently, our model distinguishes daylight and nighttime, which made it feasible to automate the process of turning the streetlight ‘ON’ and ‘OFF’ to save energy consumption costs. According to the aforementioned approach, geo-location sensor data could be utilised to make more informed streetlight management decisions. To complete the tasks, we consider training the U-net model with ResNet-34 as its backbone. Validity of the models is guaranteed with the use of assessment matrices. The suggested concept is straightforward, economical, energy-efficient, long-lasting and more resilient than conventional alternatives.

To improve insufficient management by artificial management, especially for traffic accidents that occur at crossroads, the purpose of this paper is to develop a pro-active warning system for crossroads at construction sites. Although prior studies have made efforts to develop warning systems for construction sites, most of them paid attention to the construction process, while the accidents that occur at crossroads were probably overlooked.

By summarizing the main reasons resulting for those accidents occurring at crossroads, a pro-active warning system that could provide six functions for countermeasures was designed. Several approaches relating to computer vision and a prediction algorithm were applied and proposed to realize the setting functions.

One 12-hour video that films a crossroad at a construction site was selected as the original data. The test results show that all designed functions could operate normally, several predicted dangerous situations could be detected and corresponding proper warnings could be given. To validate the applicability of this system, another 36-hour video data were chosen for a performance test, and the findings indicate that all applied algorithms show a significant fitness of the data.

Computer vision algorithms have been widely used in previous studies to address video data or monitoring information; however, few of them have demonstrated the high applicability of identification and classification of the different participants at construction sites. In addition, none of these studies attempted to use a dynamic prediction algorithm to predict risky events, which could provide significant information for relevant active warnings.