On-Line Selection of Discriminative Tracking Features

On-Line Selection of Discriminative Tracking Features

CMU-RI-TR-03-12 | Robert T. Collins and Yanxi Liu
This paper introduces a method for evaluating and adjusting feature spaces in real-time object tracking to improve performance. The authors hypothesize that the best features for distinguishing an object from its background are also the most effective for tracking. They develop an online feature ranking mechanism based on the two-class variance ratio measure, applied to log likelihood values from empirical distributions of object and background pixels. This mechanism is integrated into a tracking system that adaptively selects the top-ranked discriminative features. The paper demonstrates how this method adapts to changing appearances of both the tracked object and the scene background through examples. The approach is supported by experiments on challenging video sequences, showing successful tracking even in low-contrast and occluded conditions. The results highlight the effectiveness of the proposed method in maintaining robust tracking performance over time.This paper introduces a method for evaluating and adjusting feature spaces in real-time object tracking to improve performance. The authors hypothesize that the best features for distinguishing an object from its background are also the most effective for tracking. They develop an online feature ranking mechanism based on the two-class variance ratio measure, applied to log likelihood values from empirical distributions of object and background pixels. This mechanism is integrated into a tracking system that adaptively selects the top-ranked discriminative features. The paper demonstrates how this method adapts to changing appearances of both the tracked object and the scene background through examples. The approach is supported by experiments on challenging video sequences, showing successful tracking even in low-contrast and occluded conditions. The results highlight the effectiveness of the proposed method in maintaining robust tracking performance over time.
Reach us at info@study.space
Understanding On-line selection of discriminative tracking features