High-speed tracking for overlapped vehicles using Instance Segmentation and contour deformation

Hiromichi Kawahara, Taku Senoo, Idaku Ishii, Masahiro Hirano, Norimasa Kishi, Masatoshi Ishikawa

Abstract: In this paper, we propose a method for tracking with identification and separation using high-speed vision. This method is based on template matching and updating of template images by instance segmentation. We used an image inside the contour of the tracking object as the template image for template matching, and reduced the tracking noise. To validate the proposed method, experiments were conducted on images acquired by a camera mounted on an actual vehicle. Tracking based on contour information realized more robust high-speed tracking than that achieved in recent research in case of overlapping of vehicles.