Research On Target Tracking Algorithm Based On Compression Sensing

Moving target tracking is one of the hottest research topics in the field of computer vision, it combines advanced technology in many areas such as pattern recognition, artificial intelligence and image processing. In recent years, moving target tracking technology has been widely used in military field and civilian field which embodies great research values and broad application prospects.This thesis studied target tracking in-depth, discussed and analyzed the main problems of moving target tracking technology on the basis of previous studies. These problems are mainly:in the process of tracking on video sequences, the prevalence of target appearance changes, illumination changes, occlusion, time complexity and so on. To solve these problems, this thesis deeply studied the technology of compressive sensing, collected signal directly in the sampling rate less than the Nyquist frequency using the unrelated measurement matrix according to the sparse characteristic of the signal which abandon the redundant information. In order to improve the stability and real-time performance of tracking algorithm in complex environment, this thesis proposed APMCCT algorithm, the main work are as following:(1) Researched the compressive sensing theory, analyzed the core part of the compressive sensing theory in-depth:the sparse signal representation, random measurement matrix and reconstruct signals. In this paper, the random measurement matrix was used to map the Harr-like feature vector to the low dimensional space based on the technology of compressed sensing.(2) Researched the principle of angular point detection and focused on Harris operator. The Harris operator was used to extract the angular points, NCC (Cross Correlation Normalized) method was used to match the angular points, and the RANSAC (RANdom SAmple Consensus) method was used to correct and eliminate the error matching points. Meanwhile, the offset of angular points was calculated.(3) Researched the moving target tracking algorithm in-depth, summarized the problems existed in the existing tracking algorithm:appearance change, illumination change, occlusion and other factors. To solve these problems, this paper proposed APMCCT algorithm:the random measurement matrix was used to extract compressed Haar-like features, the Haar-like feature after dimension reduction was used to train the classifier, and the maximum of the response values of the classifier was used to calculate the offset of target position; then, the NCC method was adopted to match the angular points, and the RANSAC method was used to correct and eliminate the error matching points; finally, calculate the offset of the angular points, calculate target location combined with both offset.Experimental results show that the APMCCT algorithm has good real-time performance and robustness, and can adapt to the factors such as pose variation, illumination change, occlusion and so on.