Resarch On Compress Tracking In Complex-Scene

Target tracking is a hot research of computer vision technology.It has a good application value in the fields of artificial intelligence,traffic control,public security and video surveillance.Recently,according to the theory of compressed sensing,the original feature is compressed by the sparsity of the signal to improve the real-time performance.After being introduced into the tracking field,it achieves some results.But in the process of tracking,the moving object is Prone to drift or loss affected by the factors such as illumination,texture,pose variation and occlusion.Based on the previous research results,In this paper,some improvement methods are put forward for the existing problems.The main work is as follows:1、A real-time compressive tracking algorithm based on rectangle feature selection was proposed.Firstly,Generate projection matrixes were generated in an initial phase,and the projection matrixes were applied to extract the feature for constructing a feature pool.The characteristics of target used the rectangle feature represent in the feature pool,remove the rectangular features with the target characteristics differences;Finally,The classifier was taken to process candidate samples by Bayes classification and the response results to the classifier were taken as tracking results.The experimental results show that the improved algorithm has a stronger robust in the case of target textures,lighting changes and complex background which has larger changes.2、An fast tracking algorithm is proposed for tracking drift.First,the algorithm obtain features by the input images with different convolution rectangular box.Using random projection matrix to compress high-dimensional feature.Then,it choose to reflect the charateristics the tracking target of center characteristics,the linear feature and edge feature structure characteristics of the pool;The classifier was taken to process candidate samples by Bayes classification,and the response results to the classifier were taken as tracking results.;Finally,Optimize the parameters of the learning rate,To reduce the effect of noise on tracking drift.The experimental results show that the improved algorithm can effectively improve the tracking drift problem,The accuracy and stability of the fast compression tracking algorithm are improved.Improve the accuracy and stability of the fast compression tracking algorithm.The processing time of the improved algorithm is slightly lower than that of the fast compression algorithm.3、An improved multi feature compression tracking algorithm was proposed.Firstly,two random projection matrices are introduced to extract two kinds of complementary texture features and gray mean features.;And the projection matrixes are used to extract the rectangle feature to construct a feature pool.And the feature of the reaction target is chosen to construct the classifier.Then calculate the complementary characteristics of the classification of sample weights,using the large weights features of the selected to find the target in next frame.Finally,the classifier is taken to process candidate samples by Bayes classification and the response results to the classifier are taken as tracking results.The experimental results show that.As compared with the compressive tracking with single kind of feature,the improved method can track stably under big changed lightings and target textures.