Xu, Wei (Gustave Eiffel University), Gruyer, Dominique (Université Gustave Eiffel), Ieng, Sio-Song (Université Gustave Eiffel)

BeliefTrack: A New Framework for Improving SORT-Like Tracking Algorithms with Multi-Feature Association and Confidence Management

Scheduled for presentation during the Regular Session "Sensing, Vision, and Perception III" (ThAT5), Thursday, September 26, 2024, 12:10−12:30, Salon 13

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Abstract

DeepSORT, a widely recognized Kalman Filter-based tracking-by-detection (KFTBD)-based algorithm, has inspired various derivative versions, named SORT-like algorithms in this paper. Building on our previous work and SORT-like algorithm, this paper introduces an innovative framework that integrates a multi-feature association based on belief theory, and several enhanced components based on confidence, such as the Kalman filter introducing associations confidence, which aims to improve tracking performance in various complex environments. The proposed method is designed to be generic and can be integrated into a SORT-like tracker for improvement. The performance and compatibility of the new framework, namely BeliefTrack, have been evaluated and validated by 1) applying it to several datasets containing various and complex environments, 2) comparing it to DeepSORT as the baseline and several variants versions from StrongSORT, and 3) adapting it to different detectors. In all cases, our BeliefTrack demonstrates improved results, sometimes significantly.