RLM-Tracking: Online Multi-Pedestrian Tracking Supported by Relative Location Mapping

• DOI: 10.48550/arxiv.2210.10477
• Publication Date: 2022-01-01
• Authors: Kai Ren; Chuanping Hu
• Keywords: Tracking (education); Position (finance); Tracking system

The problem of multi-object tracking is a fundamental computer vision research focus, widely used in public safety, transport, autonomous vehicles, robotics, and other regions involving artificial intelligence. Because the complexity natural scenes, object occlusion semi-occlusion usually occur tasks. These can easily lead to ID switching, loss, detect errors, misaligned limitation boxes. conditions have significant impact on precision tracking. In this paper, we design new tracker for above issues that contains an \textbf{Relative Location Mapping} (RLM) model \textbf{Target Region Density} (TRD) model. more sensitive differences position relationships between objects. It introduce low-score detection frames into different real-time according density video. This improves accuracy without consuming extensive arithmetic resources. Our study shows proposed has considerably enhanced HOTA DF1 measurements MOT17 MOT20 data sets when applied advanced MOT method.