A rotating adaptive model for human tracking in Thermal Catadioptric Omnidirectional Vision

Previously, most human tracking system mainly focuses on conventional imaging system that depends on the illumination and has a limited field of view. In this paper, we propose to introduce a novel surveillance system that is Thermal Catadioptric Omnidirectional (TCO) Vision System. The proposed system is able to realize the surveillance in all-weather and big field of view conditions. For human tracking in TCO vision, the most of existing contour based feature cannot be used directly unless unwarp the distorted omnidirectional image into the traditional rectangle panoramic image for rectification. In this paper, a contour coding based rotating adaptive model is proposed, which is developed based on the characteristic of TCO vision. The proposed model not requires unwarping but only use relative angle based on location of target in the omnidirectional image to change the sequence of sampling from the proposed model. Based on this model, contour feature is extracted, which contains geometric location (r, θ) and gradient information g(r, θ) of the target. The extracted feature is fed into Support Vector Machine (SVM) for classification. For tracking purpose, we integrate the obtained classification posterior probability of SVM with the likelihood observation of particle filter to realize tracking in TCO vision. Finally, a series comparative experiments and quantitative analysis are presented to verify the feasibility and performance of proposed tracking method.