压缩感知近年来在目标跟踪领域得到了广泛应用，它对海量特征压缩降维，在贝叶斯分类器模型下能取得很好的分类效果，处理速度快，具有实时性。但尺寸固定不变的跟踪窗口不能有效跟踪存在明显尺度变化的目标。本文采用多尺度和级联分类器机制，选取最佳尺度下的窗口作为最终目标。实验结果表明，本算法不仅在目标形态变化、光线变化、多目标干扰、运动模糊等复杂场景下有较好跟踪效果，在目标尺度变化时也有较强鲁棒性。

Compressive Sensing has been successfully applied to visual tracking for its high efficiency and robustness in dimension reduction. It can achieve amazing classification results by using naive Bayes classifier with high frame rate. Nevertheless, it performs poorly while tracking targets with obvious scale change. An improved algorithm based on compressive tracking is proposed, which adopts multi-scale and cascade classifier to determinate the best scale and position. Experiment results indicate this algorithm not only performs well in challenging scenes such as pose variation, abrupt illumination, multi-target interference, motion blur, but also prevails in scale variant sequences.