A Unified Multi-view Multi-person Tracking Framework

Although there is a significant development in 3D Multi-view Multi-personTracking (3D MM-Tracking), current 3D MM-Tracking frameworks are designedseparately for footprint and pose tracking. Specifically, frameworks designedfor footprint tracking cannot be utilized in 3D pose tracking, because theydirectly obtain 3D positions on the ground plane with a homography projection,which is inapplicable to 3D poses above the ground. In contrast, frameworksdesigned for pose tracking generally isolate multi-view and multi-frameassociations and may not be robust to footprint tracking, since footprinttracking utilizes fewer key points than pose tracking, which weakens multi-viewassociation cues in a single frame. This study presents a Unified Multi-viewMulti-person Tracking framework to bridge the gap between footprint trackingand pose tracking. Without additional modifications, the framework can adoptmonocular 2D bounding boxes and 2D poses as the input to produce robust 3Dtrajectories for multiple persons. Importantly, multi-frame and multi-viewinformation are jointly employed to improve the performance of association andtriangulation. The effectiveness of our framework is verified by accomplishingstate-of-the-art performance on the Campus and Shelf datasets for 3D posetracking, and by comparable results on the WILDTRACK and MMPTRACK datasets for3D footprint tracking.