An adaptive compressive tracking algorithm for amphibious spherical robots

Abstract

As a critical important function for autonomous mobile robots, visual tracking is a challenge work in the field of computer vision, for the reason that factors like illumination variance, partial occlusions and target appearance changes shall be carefully considered. Focus on applications of our amphibious spherical robots, an adaptive visual tracking algorithm was proposed on the basis of compressive tracking. A feature selection method was designed to choose random Haar-like feature templates in various scales by calculating Fisher's criterion functions of features. On this basis, a random feature pool, which tried to preserve discriminative features at different frames, were constructed and then maintained on-line to provide candidate appearance model of the target. Moreover, an adaptive update mechanism was adopted for selectively updating feature templates and classifier parameters of the improved compressive tracking algorithm, which alleviated the drift problem. Experimental results with various image sequences demonstrated the effectiveness and robustness of the proposed tracking algorithm, which can meet practical application requirements of the amphibious spherical robots.