Research Of Object Tracking Algorithm Based On The Biomimetic Pattern Recognition

Computer vision replace the human brain with computer simulation to aware and understand the surrounding environment, object tracking in the computer vision is one of the important research direction, having important applications in real life, and it has been widely applied to intelligent monitoring, car navigation, medical image analysis and other projects. The biomimetic pattern recognition is given priority to "know" things, using multi-weight neural network in high dimensional space to cover samples, provides a new method for objecting tracking.By studying the biomimetic pattern recognition theory, the principle of homology continuity and multi-weight neural network, the traditional biomimetic pattern recognition object tracking algorithm, according to the practical application, in view of the traditional biomimetic pattern recognition object tracking algorithm only using the gray-scale features of training samples, puts forward the fusion image HOG feature and SIFT feature of biomimetic pattern recognition object tracking algorithm, add HOG and SIFT features in gray-scale features, using the properties of the HOG and SIFT features can effectively resist the illumination and the influence of the deformation. Verified by theoretical analysis and experimental comparison, this algorithm compared with the traditional algorithm is more accurate and stable.Given offline learning fixed in its coverage of biomimetic neural network generalization ability is weak, we studied the online learning algorithm, and apply it to the fusion image HOG and SIFT features of biomimetic pattern recognition in the object tracking algorithm. Using prediction ways to expand the training sample set, solved the disadvantage that offline mode need select a large number of training samples, build hyper sausage neural network to cover, with a minimum distance classifier to determine the target location, when target deformation occurs it can real-time update the network. The algorithm is verified by experiments that it is stable and accurate to track the target in a variety of complex cases.