Abstract: As real-time compressive tracking algorithm cannot steadily track object when appearance of target or external environment change heavily, compressive tracking method based on adaptive feature fusion is proposed. Two random measurement matrices are adopted to obtain texture and color features from V and H channel respectively. Then relative reliability index is used to calculate weights which can be update online, and two features are combined to improve tracking performance based on their own advantages. Testing results on challenging videos show that the proposed method can track object stably when target appearance or external environment change heavily. The method can run in real-time that the average frame rate is 22 frames/s when the target scale is 70　pixel×100　pixel. The proposed method is more robust in comparison with the compressive tracking algorithm which extracting single feature.

Key words: object tracking, compressive sensing, feature fusion, real-time tracking

摘要： 针对运动目标外观或背景变化较大时，采用基于压缩感知的跟踪算法由于特征单一易导致漂移、跟踪不稳定甚至丢失目标等问题，提出了改进的基于自适应特征融合的压缩感知跟踪算法。该算法采用两种随机测量矩阵，分别投影V、H空间得到压缩后的纹理和颜色特征，利用在线计算的特征可靠性相对程度来自适应调整特征加权系数，充分利用两类特征的互补性来增强跟踪稳定性。对不同视频的测试结果表明，提出的方法在目标外观、背景环境变化时仍能准确跟踪目标，在目标大小为70像素×100像素时平均帧率为22帧/s，达到实时性。与提取单一特征的原压缩感知算法相比，改进后的方法在目标外观和背景变化时具有更强的鲁棒性。

关键词: 目标跟踪, 压缩感知, 特征融合, 实时跟踪