Target tracking method based on compressed-sensing theory and gcForest

The invention discloses a target tracking method based on compressed-sensing theory and gcForest. The method includes the following steps: acquiring an initial-frame video image of a tracked target; extracting positive and negative sample image slices, and carrying out multi-scale transformation to obtain multi-dimensional vectors; extracting deep-level features of the tracked target through a gcForest network to obtain deep-level expression of the target; adopting the compressed-sensing theory to carry out dimension reduction on the features thereof to obtain final feature expression, and training a classifier; and sampling n windows around a target location of a last frame on a next-frame image, using the classifier, which is trained by the previous frame, for classification, determiningthat a window, which obtains a largest classification score, is the tracked target, and updating classifier parameters thereby. According to the method, precision of video target tracking is effectively improved, the target