A performance evaluation method for infrared tracker
Haichao Zheng, Jie Yang, Jianjun Chen · 2017
In recent years, significant progress has been made in the evaluation of tracking performance. However, there is still one issue that remains unsolved. All existing works evaluate tracking performance based on empirical comparison using limited video sequences. Performance evaluation based on empirical comparison leads to the following problems: (1) we don't know that how robust a tracker is for all possible scenarios; (2) we will never know what the performance of a tracker is for untested video sequences until we execute it on those sequences. To address these problems, we propose a performance evaluation method for infrared tracker, which can predict the tracking performance of a tracker for untested infrared video sequences. In the proposed method, an image sequence metric is introduced first to quantify tracking difficulty of the infrared video sequence. Afterwards, we establish a tracking dataset including infrared video sequences with gradually increasing tracking difficulties. Then, the tracker need to be evaluated is executed on the tracking dataset. Meanwhile, the tracking performance is quantitatively evaluated by a measure. Thereafter, we identify the relationship between tracking difficulties of infrared video sequences and corresponding quantitative tracking results of that evaluated tracker. Based on that relationship, when tracking difficulty of an untested video sequence is determined, we can predict the tracking performance of that evaluated tracker for that untested infrared video sequence. Experimental results prove that the proposed performance evaluation method can effectively predict the tracking performance of a tracker for untested infrared video sequences.