Long-term single-target pedestrian tracking combined with reliability discrimination and re-detection
Jie Zhang, Tao Tao · 2023
Long-term pedestrian tracking is susceptible to reduced accuracy or even failure due to challenging factors such as occlusion and out-of-field events. To address this problem, we propose a novel long-term pedestrian tracking algorithm that combines reliability discrimination and re-detection techniques. The proposed algorithm assesses the reliability of tracking results by analyzing the response peak of correlation filtering and average peak correlation energy, which facilitates adaptive updating of the model. Additionally, we use the YOLO algorithm to detect the location of candidate pedestrian targets and utilize their response peak to determine whether to update the tracking location. In comparison to an advanced correlation filtering algorithm using the pedestrian sequences in the OTB-100 dataset, our algorithm achieves accuracy and success rates of 0.974 and 0.709, respectively. Moreover, it exhibits strong robustness and real-time performance in complex environments such as target occlusion and out-of-field challenges.