Car tracking in rear view based on bicycle specific motions in vertical vibration and angular variation via prediction and likelihood models with particle filter for rear confirmation support

Aiming at rear confirmation support of a bicycle, our objective is to track of a car in rear view captured by a camera settled under the saddle. Where specific motions of bicycle such as angular variation and vertical vibration are necessary to cope with. We propose a novel state space model coping with the two specific motions and utilize particle filter for state estimation. For angular variation, an elaborated system noise with variable mean having larger pulling force for larger angle. For likelihood model to cope with vertical vibration, we propose an elaborated likelihood evaluation having more sensitive feature for horizontal while less sensitive for vertical motion. Experimental result with real scene videos achieves 74.24 percent precision of the tracking.