Performance evaluation of imaging seeker tracking algorithm based on multi-features

Abstract

The paper presents a new efficient method for performance evaluation of imaging seeker tracking algorithm. The method utilizes multi features which associate with tracking point of each video frame, gets local score(LS) for every feature, and achieves global score(GS) for given tracking algorithm according to the combined strategy. The method can be divided into three steps. In a first step, it extracts evaluation feature from neighbor zone of each tracking point. The feature may include tracking error, shape of target, area of target, tracking path, and so on. Then, as to each feature, a local score can be got rely on the number of target which tracked successfully. It uses similarity measurement and experiential threshold between neighbor zone of tracking point and target template to define tracking successful or not. Of course, the number should be 0 or 1 for single target tracking. Finally, it assigns weight for each feature according to the validity grade for the performance. The weights multiply by local scores and normalized between 0 and 1, this gets global score of certain tracking algorithm. By compare the global score of each tracking algorithm as to certain type of scene, it can evaluate the performance of tracking algorithm quantificational. The proposed method nearly covers all tracking error factors which can be introduced into the process of target tracking, so the evaluation result has a higher reliability. Experimental results, obtained with flying video of infrared imaging seeker, and also included several target tracking algorithms, illustrate the performance of target tracking, demonstrate the effectiveness and robustness of the proposed method.