Motion-Based Kernel Particle Filter for Visually Tracking Antenna Deployment

For reliable and precise deployment of a deployable antenna, this paper presents a vision system for tracking the joint of antenna by a motion-based kernel particle filter (M-KPF) algorithm. M-KPF improves the performance of tracking relied on the motion-based proposal density which takes into a precise dynamic model. For obtaining the motion-based proposal density, firstly, the dynamic model of the deployable antenna is built by Moore-Penrose generalized inverse matrix method; secondly, the motion estimation error is corrected through initiative correction method. Additionally, in order to achieve more robust performance, a target representation based on multiple kernel histograms calculated over interesting and uninteresting regions is utilized to model appearance of the joint to eliminate the background clutter. By analyzing the video of the deploying process, compared with the KPF, the SIR particle filter and the mean shift algorithm, the proposed tracking scheme is proved superior in its robust performance, high accuracy and agility, as well as the capability of adapting to the visual analysis of the deployable antenna.