Since it relies on the current new sample, the original compressive tracking algorithm does not perform well, especially in the case of occlusion.In order to improve the tracking accuracy when there is occlusion, the invariant moments of samples are extracted by means of sample blocks, which are used as the classifying standard of Bayesian classifier. Then, the difference of the invariant moments between two sequential frames is calculated, and the value is used to judge whether the object is occluded or not, thus to achieve adaptive updating of the classifier. Experiments show that:1) The algorithm has good tracking results when the object is partially occluded, and is robust to scale changes;2) The tracking accuracy is improved compared with that of the original algorithm; and 3) When the object is 25×69 pixels, the average processing speed is 58 frames per second, which meets the real-time requirements.