Visual Tracking based on Moving Monocular Camera

This paper investigates the problem of moving target tracking in the case of a monocular mobile camera. Firstly, a scene space model was established under a moving camera based on the constraint conditions of ground plane. Thus, a three-dimensional (3D) motion model of target could be established. The scene space model and the target 3D motion model could both be compensated by the posture of the camera, which can be calculated according to the vanishing point in the image or the motion parameters of the camera acquired by sensors. Secondly, an interception function was established to extract the target motion range according to the target 3D motion model. Both the influence of background factors and the computational complexity of the algorithm were reduced. Then, a dual-kernels tracking method was used to track the target under a situation of non-obvious color characteristics and partial shielding. Finally, a public data set was used to conduct comparison experiments with three other algorithms to verify the effectiveness and robustness of the proposed algorithm.