Abstract: In order to solve the problem of motion blur which disturbs target gradient information and causes serious occlusion of targets in video sequences, a novel visual tracking algorithm, which combines the MTRSR model with AdaBoost classifier is proposed. The algorithm uses a descriptive dictionary to estimate the weight of each candidates. Firstly, a MTRSR model is utilized to get the blur kernel [k] to get blur target templates set, and meanwhile the confidence of the candidates is also calculated by the reconstruction error. AdaBoost classifier is trained to evaluate the confidence of all candidates. To this end, the HOG features of the target templates are used to encode a descriptive dictionary to calculate the weights of the candidates. Finally, the best target is retrieved by the sum of production of the weight and the two confidences. The experimental results show that the algorithm can efficiently cope with the gradient change caused by motion blur and severe target occlusion in complicated scenes, and further improve the accuracy and robustness of visual tracking.

Key words: visual tracking, AdaBoost classifier, sparse representation, reconstruction error

摘要： 针对在视频序列中因移动模糊导致的对目标梯度信息的干扰以及目标的严重遮挡等问题，提出使用多任务反向稀疏表示（MTRSR）模型与AdaBoost分类器相结合的目标跟踪方法，同时为有效跟踪目标区域使用一个描述性的字典来估计每一个候选目标的权值。通过MTRSR模型得到模糊核[k]以此得到模糊目标模板，同时通过稀疏匹配计算重建误差得到目标的置信度，以目标模板的HOG特征组建描述性字典并通过重建误差计算候选目标权值，通过AdaBoost分类器计算所有目标的置信度，最后依据权值与二者置信度乘积的和得到最佳目标。实验数据表明该算法能够很好地应对复杂场景下目标的梯度变化、模糊效应以及遮挡，提高了目标跟踪精度与鲁棒性。

关键词: 目标跟踪, AdaBoost分类器, 稀疏表示, 重建误差