DCMTrack: Rethinking the Motion for Vehicle Multi-Object Tracking

Peng Li, Jun Ni, Dapeng Tao
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:1-7, 2025.

Abstract

Vehicle multi-object tracking has significant applications in many fields. Existing methods struggle to address the challenges of nonlinear motion and prolonged occlusion in vehicle tracking. In this paper, an advanced tracker featuring a Nonlinear Noise Adaptive Unscented Kalman Filter, namely DCMTrack, is designed to finely adjust measurement noise and significantly enhance the accuracy of target motion state predictions. An Adaptive Direction and Confidence Cost Matrix is designed to more precisely calculate trajectory direction and confidence, enhancing the accuracy of target association. Ultimately, a Category-Aware Initialization Mechanism that integrates target category and environmental information is proposed to minimize false trajectories and optimize the overall trajectory management process. We conducted extensive experiments on the VehiclesMOT dataset, validating its effectiveness.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-li25a, title = {DCMTrack: Rethinking the Motion for Vehicle Multi-Object Tracking}, author = {Li, Peng and Ni, Jun and Tao, Dapeng}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {1--7}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/li25a/li25a.pdf}, url = {https://proceedings.mlr.press/v278/li25a.html}, abstract = {Vehicle multi-object tracking has significant applications in many fields. Existing methods struggle to address the challenges of nonlinear motion and prolonged occlusion in vehicle tracking. In this paper, an advanced tracker featuring a Nonlinear Noise Adaptive Unscented Kalman Filter, namely DCMTrack, is designed to finely adjust measurement noise and significantly enhance the accuracy of target motion state predictions. An Adaptive Direction and Confidence Cost Matrix is designed to more precisely calculate trajectory direction and confidence, enhancing the accuracy of target association. Ultimately, a Category-Aware Initialization Mechanism that integrates target category and environmental information is proposed to minimize false trajectories and optimize the overall trajectory management process. We conducted extensive experiments on the VehiclesMOT dataset, validating its effectiveness.} }
Endnote
%0 Conference Paper %T DCMTrack: Rethinking the Motion for Vehicle Multi-Object Tracking %A Peng Li %A Jun Ni %A Dapeng Tao %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-li25a %I PMLR %P 1--7 %U https://proceedings.mlr.press/v278/li25a.html %V 278 %X Vehicle multi-object tracking has significant applications in many fields. Existing methods struggle to address the challenges of nonlinear motion and prolonged occlusion in vehicle tracking. In this paper, an advanced tracker featuring a Nonlinear Noise Adaptive Unscented Kalman Filter, namely DCMTrack, is designed to finely adjust measurement noise and significantly enhance the accuracy of target motion state predictions. An Adaptive Direction and Confidence Cost Matrix is designed to more precisely calculate trajectory direction and confidence, enhancing the accuracy of target association. Ultimately, a Category-Aware Initialization Mechanism that integrates target category and environmental information is proposed to minimize false trajectories and optimize the overall trajectory management process. We conducted extensive experiments on the VehiclesMOT dataset, validating its effectiveness.
APA
Li, P., Ni, J. & Tao, D.. (2025). DCMTrack: Rethinking the Motion for Vehicle Multi-Object Tracking. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:1-7 Available from https://proceedings.mlr.press/v278/li25a.html.

Related Material