Research Article
Multiple People Tracking Based on Improved SiameseFC Combined with Lightweight YOLO-V4
@INPROCEEDINGS{10.1007/978-3-031-65123-6_21, author={Lu Shen and Zhiwen Chen and Boliang Zhang and Su-Kit Tang and Silvia Mirri}, title={Multiple People Tracking Based on Improved SiameseFC Combined with Lightweight YOLO-V4}, proceedings={Quality, Reliability, Security and Robustness in Heterogeneous Systems. 19th EAI International Conference, QShine 2023, Shenzhen, China, October 8 -- 9, 2023, Proceedings, Part II}, proceedings_a={QSHINE PART 2}, year={2024}, month={8}, keywords={Multi-object tracking Object detection Object tracking}, doi={10.1007/978-3-031-65123-6_21} }- Lu Shen
Zhiwen Chen
Boliang Zhang
Su-Kit Tang
Silvia Mirri
Year: 2024
Multiple People Tracking Based on Improved SiameseFC Combined with Lightweight YOLO-V4
QSHINE PART 2
Springer
DOI: 10.1007/978-3-031-65123-6_21
Abstract
Multi-object tracking (MOT) is an active area of research in computer vision that is extensively applied in various domains, including but not limited to video surveillance, security, and intelligent transportation. There are two types of tracking algorithms: standard visual tracking techniques and deep learning tracking methods. Deep learning methods are becoming more common, but current tracking algorithms still need to overcome the challenge of false detection due to occlusion, similar backgrounds, and also the problem of slow speed. In response to the existing difficulties in multi-object tracking, this paper improves the fully convolutional Siamese (SiameseFC) network and integrates the Kalman filter to enhance the performance of the tracker. The lightweight network is used to improve the YOLO-V4 structure. The multi-people tracking network designed in this paper combines both networks, enabling objects to be detected and re-tracked after they reappear. By comparing with the performance of the network before improvement and other high-performance multi-object tracking algorithms, our proposed method can improve the processing speed of images while almost not losing too much precision, significantly reducing the model size.
