Article
Computer Science, Artificial Intelligence
Fugang Liu, Songnan Duan, Wang Juan
Summary: In this study, a deep learning-based method for pedestrian trajectory prediction is proposed. The method combines YOLOv7, StrongSORT, and improved LSTM algorithm to solve the problems of target switch and jump, and improves the prediction performance.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Jieming Yang, Hongwei Ge, Jinlong Yang, Yubing Tong, Shuzhi Su
Summary: This study proposes a novel online pedestrian multiple object tracking method for targets with severe occlusion. By refining the predicted position of the target using a regression network and categorizing heavily occluded targets into different types, the proposed method achieves state-of-the-art performance.
NEURAL PROCESSING LETTERS
(2022)
Article
Computer Science, Theory & Methods
Azzedine Boukerche, Mingzhi Sha
Summary: This article presents design guidelines on deep learning-based pedestrian detection methods to support autonomous vehicles. It covers classic backbone models and frameworks, inherent attributes of pedestrian detection, representative pedestrian detectors in aspects such as occlusion handling, multi-scale feature extraction, multi-perspective data utilization, and hard negative handling. The developments, trends and challenges in this area are also discussed.
ACM COMPUTING SURVEYS
(2021)
Article
Computer Science, Information Systems
Hyejin Lee, Haechan Cho, Byeongjoon Noh, Hwasoo Yeo
Summary: This study introduces NAVIBox, a novel system that uses vision sensors in edge computing devices to proactively identify vehicle-pedestrian risks. The system combines motioned-video capture, object detection and tracking, trajectory refinement, and predictive risk recognition and warning decision to accurately identify and address risks. Real-world testing in Sejong City, South Korea validates the feasibility and practicality of the proposed system. It provides a comprehensive solution to enhance safety and efficiency in urban environments.
Article
Multidisciplinary Sciences
Imran N. Junejo, Naveed Ahmed, Mohammad Lataifeh
Summary: This paper addresses the issue of pedestrian attribute recognition in surveillance scenarios, proposing a novel method that integrates trainable Gabor wavelet layers within a CNN, achieving better recognition performance.
Article
Robotics
Yu Yao, Ella Atkins, Matthew Johnson-Roberson, Ram Vasudevan, Xiaoxiao Du
Summary: This letter introduces BiTraP, a goal-conditioned bidirectional multi-modal trajectory prediction method based on CVAE, which accurately predicts pedestrian trajectory goals and improves long-term prediction accuracy. Extensive experiments show BiTraP's superior performance across different scenarios, outperforming state-of-the-art methods by 10-50% in accuracy.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2021)
Article
Computer Science, Information Systems
Yueyan Zhu, Hai Huang, Huayan Yu, Aoran Chen, Guanliang Zhao
Summary: This paper proposes an anchor-free pedestrian detector called CAPNet to address the challenges in pedestrian detection. It introduces a feature extraction module, a global feature mining and aggregation network, and an attribute-guided multiple receptive field module to enhance the detection performance. Experimental results show that the context and attribute perception greatly improves the detection, and CAPNet achieves new state-of-the-art performance on Caltech and CityPersons datasets.
Article
Computer Science, Information Systems
Imran N. Junejo
Summary: The paper addresses the problem of pedestrian attribute recognition, proposing a method using trainable Gabor wavelet layers combined with convolution neural network. A multi-branch neural network is designed for this purpose. Testing on challenging datasets and comparing with state of the art validates the effectiveness of the proposed method.
Article
Chemistry, Multidisciplinary
Han Xie, Wenqi Zheng, Hyunchul Shin
Summary: A novel pedestrian detector using deformable attention-guided network (DAGN) was developed in this research, featuring a deformable convolution with an attention module (DCAM) and optimized loss function. Extensive evaluations on multiple datasets showed promising detection performance compared to state-of-the-art methods.
APPLIED SCIENCES-BASEL
(2021)
Article
Computer Science, Artificial Intelligence
Fei Gao, Changxin Cai, Ruohui Jia, Xinzhong Hu
Summary: In recent years, significant progress has been made in object detection in computer vision. However, crowded pedestrian detection in one-stage detectors remains challenging, with few improved solutions available. This paper introduces YOLO-CPD, a novel method for crowded pedestrian detection that outperforms other one-stage models in crowded environments. YOLO-CPD enhances the one-stage detector's ability to detect multiple overlapping objects in a single area through an optimized score module and adjustment of the IoU value in non-maximum suppression. Experimental results demonstrate the superior performance of YOLO-CPD on the CrowdHuman and WiderPerson datasets.
JOURNAL OF REAL-TIME IMAGE PROCESSING
(2023)
Article
Computer Science, Interdisciplinary Applications
Arianna Salazar Miranda, Zhuangyuan Fan, Fabio Duarte, Carlo Ratti
Summary: The study uses pedestrian trajectories and built environment analysis to show that desirable streets provide better access to public amenities and have characteristics such as sinuosity and visual enclosure.
COMPUTERS ENVIRONMENT AND URBAN SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Narina Thakur, Preeti Nagrath, Rachna Jain, Dharmender Saini, Nitika Sharma, D. Jude Hemanth
Summary: Pedestrian detection is crucial for crowd surveillance applications and cyber-physical systems. In this paper, the hyperparameters of the Yolov5 object detection algorithm have been tuned to develop a new efficient and reliable object detector. The modified Yolov5 model outperforms existing algorithms and achieves better object detection accuracy.
Article
Computer Science, Artificial Intelligence
Shuangquan Zuo, Yun Xiao, Xiaojun Chang, Xuanhong Wang
Summary: Transformers have shown impressive capabilities in computer vision, especially in dense prediction tasks. Their inherent properties allow for stable and high-resolution feature processing, satisfying the demands of fine-grained and globally coherent predictions. Compared to convolutional networks, transformers require minimal inductive bias and allow for long-range information interaction. This survey provides a comprehensive overview of transformer models, focusing on dense prediction, and explores optimization strategies and development directions.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Yannis Panagakis, Jean Kossaifi, Grigorios G. Chrysos, James Oldfield, Mihalis A. Nicolaou, Anima Anandkumar, Stefanos Zafeiriou
Summary: Tensors are data structures that can naturally represent multi-dimensional visual data and have a long history of applications in various computer vision problems. With the rise of deep learning in computer vision, tensors have become fundamental and play crucial roles in modern deep learning architectures. Tensor methods are increasingly used in deep learning for designing efficient network architectures, enhancing robustness, and aiding in theoretical understanding of deep networks.
PROCEEDINGS OF THE IEEE
(2021)
Review
Multidisciplinary Sciences
Duidi Wu, Haiqing Huang, Qianyou Zhao, Shuo Zhang, Jin Qi, Jie Hu
Summary: This article provides an overview of the importance of pedestrian attribute recognition (PAR) and re-identification (ReID) in the field of computer vision. It focuses on ReID based on deep learning and analyzes the associations between PAR and ReID. The article summarizes the major ideas and methods of Attribute-Assisted ReID, and provides solutions for addressing challenges in ReID. It concludes the performance and characteristics of state-of-the-art methods, presents future research directions, and demonstrates the effectiveness of Attribute-Assisted ReID.