Scholarlink

<p>Multi-pedestrian tracking (MPT) has been widely applied in the fields of computer vision and artificial intelligence. Tracking-by-detection is a classic and efficient genre in MPT, which relies on the performance of detectors. In this paper, we proposed a multi-pedestrian tracker focused on effective detection results in complex scene. In detection algorithm, combination-of-parts (CoP) detection is operated over three iterations to suppress false negatives (FN), meanwhile, the overlap ratios of gray-scale histograms for head detection boxes are used to reduce false positives (FP), as the head detection boxes with low overlap ratios are filtered out. In tracking algorithm, the Kullback-Leibler (KL) divergence product of HSV histograms is introduced to reduce the re-initialization of pedestrian identity label, and the scale information of local observable regions for occluded or incomplete pedestrians is extracted, so as to re-associate the lost tracklets. Evaluation experiments tested on four datasets with various challenges demonstrate that the proposed method shows the superiority compared with several state-of-the-art MPT methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-pedestrian tracking based on iterative filtering and observation discrimination

  • 期刊论文

  • 发表日期：

    2025年7月30日

查看全文
  • Chen Yang,
  • Aihua Xu,
  • Yuting Li,
  • Xinping Wu

摘要

Multi-pedestrian tracking (MPT) has been widely applied in the fields of computer vision and artificial intelligence. Tracking-by-detection is a classic and efficient genre in MPT, which relies on the performance of detectors. In this paper, we proposed a multi-pedestrian tracker focused on effective detection results in complex scene. In detection algorithm, combination-of-parts (CoP) detection is operated over three iterations to suppress false negatives (FN), meanwhile, the overlap ratios of gray-scale histograms for head detection boxes are used to reduce false positives (FP), as the head detection boxes with low overlap ratios are filtered out. In tracking algorithm, the Kullback-Leibler (KL) divergence product of HSV histograms is introduced to reduce the re-initialization of pedestrian identity label, and the scale information of local observable regions for occluded or incomplete pedestrians is extracted, so as to re-associate the lost tracklets. Evaluation experiments tested on four datasets with various challenges demonstrate that the proposed method shows the superiority compared with several state-of-the-art MPT methods.