SmartTrack: Sparse multiple objects association with selective re-identification tracking

Tracking multiple objects in traffic scenes is vital for the realization of autonomous driving (AD) and intelligent transportation systems (ITS), and doing so in tunnel scenes introduces notable difficulties. Due to the limitations of tunnel environments, congested scenarios and frequent occlusion issues are unavoidable. To address these challenges, this paper proposes a novel tracker called SmartTrack, which significantly improves tracking performance through the Selective Comprehension Module (SCM) and Adaptive Depth Cascaded Matching (ADCM). The SCM, based on Siamese networks, generates more discriminative appearance features by integrating the Fused Feature Comprehension (FFC) module and the Selective Visual Feature (SVF) module. The ADCM adaptively selects the depth level according to the number of detected objects in the 2D image, transforming the dense object set into multiple sparse subsets. Then, the tracker effectively associates objects across frames by combining appearance and motion information. We conducted comprehensive experiments on the Traffic MOT datasets and Tunnel MOT datasets, and the results demonstrate that our method achieves competitive performance compared to state-of-the-art (SOTA) trackers. The code and datasets are available at: https://github.com/klaygky/SmartTrack.