The Research of Target Tracking Algorithm Based on an Improved PCANet
Qi Mu, Yanyan Wei, Yankui Liu, Zhanli Li · 2018
The feature extraction method will greatly affect the performance of the target tracking algorithm. In traditional feature extraction methods, feature descriptors are manually designed, such as HOG features, etc. which does not express spatial information very well. In this paper, a novel target tracking method is proposed. This method uses a lightweight deep learning model, called PCANet network, to extract features. In the previous work, the number of PCA layer filters is determined through a large number of experiments. In this paper, the number of filters in the PCA layer is determined by the cumulative contribution rate, and which achieves the adaptive adjustment of the network parameters. Firstly, the region of interest is acquired by particle filtering; Secondly, the depth characteristics of the image are extracted via PCANet; Finally, the target is determined by the SVM classifier. The results of the experiment show that this algorithm has strong robustness, because the target tracking can be tracked accurately under the condition of illumination change, occlusion and rapid movement.