Study On Fast Algorithm Of Target Detection And Tracking Based On Feature Points

Moving object detection and tracking is a hot research topic in the field of computer vision, its main purpose is to detect the moving object from video sequences, and based on the effective identification and reliable tracking of target, understand and describe the behavior of target. The uncertainty of video background movement and target movement makes it difficult for researchers to find a generally applicable solution to solve the problem of target detection and tracking under dynamic video background, these algorithms mainly include three key techniques such as background compensation and image pre-processing, image segmentation and target detection, feature extraction and target tracking. Algorithm of target detection and tracking based on feature points is a common and widely used one to solve this kind of problem. So algorithm of target detection and tracking based on feature points is systematically analyzed and depth studied in this paper.At present, algorithm of target detection and tracking based on feature points are still many problems to be solved. First of all, because the algorithm of traditional feature point detection has a large amount of time consuming, so that it has a good effect of target detection, but it is difficult to apply in real-time system. Secondly, because the dynamic background of video image especially aerial video images may occur translational change, rotational change, scaling change and the object in the image may be occluded by the background, these factors make the situation that it may loss the target in the process of target tracking. Therefore, this paper carries on the analysis and research on the above problems, and proposes appropriate solutions, through a large number of experiments to verify the proposed method. This main research work and achievements are listed as following:1. For the traditional algorithm of target detection based on feature points, the part of feature point detection takes a long time, a new binary feature points based on adaptive and generic acceleration segmentation of detection algorithm is proposed. In this paper, several traditional feature point detection algorithm is studied, and on this basis, an improved ORB feature points based on adaptive generic acceleration segmentation of extraction and matching algorithm is proposed, the FAST operator in ORB algorithm is changed to AGAST operator to achieve precise positioning and fast generation of feature points. It uses the improved ORB algorithm to extract and match the feature points in dynamic video background image to compensate the dynamic background of the video sequence, and uses method of accumulation of multi-frame interval difference image with morphological operations and edge detection to achieve target detection. Verified by experiments, the algorithm takes less time compared with the traditional algorithm, and has strong adaptability for different types of images.2. For the extraction and matching of feature points for target tracking in dynamic video background image, traditional algorithm takes a long time and may loss the target. To solve these problems, a tracking algorithm of Kalman filter based on improved ORB feature points is proposed. Firstly, it uses the improved ORB feature points to extraction and matching image template and each frame of video image, and calculates the center position of the initial frame of the target image template; then using the Kalman filter to predict the target center position and using expanding area method based on ROI for detection, through the matching feature points to change in the ratio of number of target template and the total number, the occlusion can be analyze, and the algorithm takes the appropriate tracking program. Experiments show that the algorithm is robust to scale, rotation, noise and occlusion, and effective in reducing the target loss happening, the realization of the moving target tracking is stable, accurate and fast.