Abstract :
This paper introduces a novel algorithm to detect and track token markers in automobile crash simulation video. To precisely locate token marker center, a logical correlation method that measure the structure similarity of the template marker with token markers is presented. Then, coordinates and textures of token marker in adjacent frames as features are extracted to be matched and the two matching scores are fused by the weighted average fusion strategy. To obtain the optimal weights, the equal error rate (EER) is minimized. Furthermore, the fused matching score is defined as similarity, and the combination of the largest similarity are recognized by the classifier. Finally, isomorphic mapping principle is applied to decide the optimal match. Experiment results prove that our algorithm has good performance on matching accuracy, which at least increases 5% compared with traditional single-feature algorithms, and could track token markers effectively.
