Moving Target Detection And Tracking Method Based On Surf Research

With the fast development of computer technology and image processing method, the research of tracking and analyzing moving targets which is based on video sequence has been drawn more and more attention. While multi-targets tracking has become hot issues in computer vision areas which has been applied widely in various areas such as intelligent transportation, video monitoring, video compression coding, image guidance, etc.In accordance with the achievements summarized by predecessors, this paper mainly studies the detecting of moving objects, local characteristics object matching and the tracking method based on dynamic characteristics, and makes some improvements. The main work is as follows:1. According to common moving target detection algorithm, this paper proposes a method based on edge difference, which combines edge detection and frame difference together. Compared with other algorithms in the scene with large shadows, the proposed algorithm can efficiently overcome the impact of shadows and noise, which has remarkable advantage. The edge profile of moving targets detected by this improved method is clear, which would get very accurate and clear moving targets after the morphological processing.2. Combined with the local characteristics tracking method, the paper makes further research on SURF feature matching. The SURF descriptor has scale-invariant features, and has very good applicability to perspective changes and brightness changes. However, it still exists the problem of low efficiency. Based on the traditional SURF algorithm, this paper makes some optimizations on the matching strategy. First using K-nearest neighbor algorithm on the Similarity Matching, and searching for the nearest neighbor using Kd-trees search strategy, which improves the searching efficiency. Secondly on the process of matching purification, using ratio purification method filter matching, as the results show that there are still wrong and redundancy matching, so we need remove error data using RANSAC algorithm, which gets more accurate matching result.3. In comparison with several common tracking algorithms. The Kalman filter has good performance on the moving target tracking, it plays an important role in improving the system processing speed. The main difficulty in target tracking is that the target information would disappear completely due to object shelter which may result in tracking failure. This paper put forward a multi-moving target tracking method which combined with SURF descriptor matching method and the Kalman filter algorithm. According to make discussions on multiple target tracking, Kalman filter would be used to estimate the target states for tracking; otherwise SURF method would be adopted to deal with occlusion existence. The algorithm receives a good performance.The experimental results show that the improved moving target detection method can get very accurate and clear moving targets, and the multi-moving target tracking method which combined with SURF and the Kalman filter algorithm satisfies both the real-time and accuracy requirements.