Research On The Technology Of Moving Object Tracking Based On Feature Point

Moving object detecting and tracking technology is a popular topic in the filed of computer vision, which has been widely used in military, civil areas etc. But it has some disadvantages such as the limited ability to occlusion and the low stability to track object.Studying a moving object detecting and tracking technology with good robustness, strong ability of anti-interference and high accuracy of tracking is the goal of many researchers for a long time.Firstly,the paper studies the principle of object tracking algorithm based on the feature point,including various feature extraction algorithms and KLT-based matching algorithm based on the optimal estimation. Because SIFT algorithm is found that it can get the feature point which is more stable and has stronger ability of anti-noise and it can adapt to the object which may change in rotation and shape well, a new tracking algorithm is proposed which combines the SIFT and KLT algorithm. In order to reduce the time-consuming,the SIFT algorithm is optimized. Because the precision of moving object loaction which is determined based on the distribution of feature points is not high,the paper uses Greedy algorithm which is based on the active contour model to get the accurate position of moving object. Experimental results show that compared with KLT algorithm, the stability and the precision of moving object loaction are improved a lot by the proposed algorithm.Secondly, the paper studies the stability of tracking moving objects. First of all,the strategy of updating feature points is proposed to solve the problem which is that the number of feature points decreases when the object poses a significant rotation. Then,a criterion based on the Bhattacharyya coefficient is proposed to estimate whether the object is under occlusion and create a dynamic template library for object.The dynamic template library provides many templates for re-identifing the object.Last, a new tracking strategy which is combined with features such as moving,color and gray is applied to achieve the re-identification of object by expanding the search area is proposed because of the difficulty to predict the trajectory of object under occlusion. Experimental results demonstrate that compared with the Kalman filter algorithm, whether the moving object keeps the linear motion is not required in the proposed strategy.So it has broader applications in the future.Finally,the paper studies the Horn-Schunck algorithm based on the differential optical flow technology for detecting moving regions and also studies the clustering of moving regions with the ISODATA algorithm. In order to get the complete area of the moving object,the paper uses Greedy algorithm which is based on the active contour model to extract the contour of moving object from the classification of the moving regions.Besides these,a moving object tracking system is built in the actual conditions of the laboratory. Experimental results show that:the proposed method for automatic detection of moving object can get relatively complete moving object area.