针对现有在线学习跟踪算法中目标在线模型更新错误导致跟踪漂移的问题, 提出一种在线模型自适应更新的目标跟踪算法: 首先利用压缩感知技术的高效性, 对多尺度图像特征进行降维, 并提取多尺度样本来实现目标尺度自适应更新, 再由提取的正负样本低维图像特征训练朴素贝叶斯分类器, 利用分类器输出置信度最大处目标样本完成目标跟踪, 并依据当前目标置信度来自适应在线模型更新速率, 减少了遮挡带来的目标错误更新。实验表明: 该方法在尺度变化、局部和全局遮挡、光照变化等情况下均能完成鲁棒跟踪, 平均跟踪成功率较原始压缩感知跟踪算法提高了 20.3%。

In order to deal with the drift problem by updating error in current online learning tracking algorithms, a new adaptive update tracking algorithm is proposed. First of all, based on the efficiency of compressed sensing, the multi-scale image feature space is decreased, and multi-scale samples are exacted to update the target scale. Secondly, a naive Bayes classifier is trained by low dimension image features from positive and negative samples. Experimental results show that the proposed algorithm can complete the robust tracking under the condition of scale changes, partial and full occlusion, illumination changes, etc. Tracking successful rate is improved by 20.3% compared with the original compressive tracking.