Particle filter planar target tracking with a monocular camera for mobile robots

摘要

This paper presents an effective and simple target tracking approach called PSIPT (Particle filter Single Image based Planar Target tracking). Compared with other works, the uniqueness of PSIPT includes: (1) only a single color camera provides the images to be processed; (2) the particle filter does not perform data fusion calculations; (3) the distance evaluation carried out in the particle filter does not need the camera's intrinsic and extrinsic parameters. Meanwhile, this paper also reveals that, under different target shapes and cameras, a high degree of negative linear dependence remains between: (1) a target's pixel height and vertical distance; (2) a target's vertical distance and PWHD (Pixel-Width-to-Horizontal Distance) ratio. According to the experimental results, PSIPT performs better than its Kalman filter variant in both the L-shape and S-shape tracking experiments. In addition, PSIPT has moderate performance in the target missing surveillance experiment. Moreover, a hybrid and enhanced version of PSIPT, which is equipped with an AdaBoost classifier in this study, leads to good surveillance performance in the target missing experiment.