Research Of Object Tracking Based On Discriminant Model

Object tracking is an important topic in the field of computer vision.The main task is to obtain the region of interests in video sequences.It has been widely used in human-computer interaction,video surveillance,behavior analysis,security and robotics industries.At present,great progress has been made in tracking technology,but in realistic scenarios,there are still many challenges.The interference between target and background,the scale change of target itself,the rotation and the illumination change of target area may lead to tracking drift or even failure.In order to solve the above problems,effective decision boundaries of targets and backgrounds are obtained by using knowledge training and updating classifiers based on discriminant model.Robustness and efficiency are the two most important indicators for target tracking.Robustness requires that the tracking algorithm can accurately output the target's motion,attitude change and scene disturbance.In order to solve the above problems,this paper firstly extracts the multi-layer convolution features of the image to be measured by using the multi-layer structure of VGG-Net-19,and then obtains the multi-layer convolution features of Convolutional Neural Network by correlation filtering and fuses them with weights,so as to determine the real location of the target.By combining the characteristics of multi-layer convolution layer and full connection layer,the tracking effect for different interference is greatly improved.The improvement of target tracking efficiency can ensure the real-time performance of the tracking algorithm.If the tracking is not real-time,the target tracking will lose its significance.In this case,the real-time high-speed computation between trackers is realized by improving a perceptual correlation filter.The main contribution to high-speed computing comes from improved depth feature compression,which is achieved by combining content-aware features with multiple automatic encoders.In order to obtain the feature map suitable for target tracking,orthogonal loss function is added to the training phase and the fine-tuning self-encoder phase respectively,and the tracking efficiency is greatly improved.All of experiments show that the improved algorithm can improve the efficiency and robustness respectively.