Moving Target Detection And Tracking Algorithm Research In Video Sequences

With the development of computer and image processing technology, the detection and tracking of moving target has been one hot research field in video sequence. It combines many advanced technology, and it is widely applied in military guidance, video surveillance,intelligent transportation, medical diagnosis, machine vision, etc. Among them, the moving target tracking based on feature extraction is the most widely used, with high efficiency,robustness and extracting characteristic features. Image denoising, feature extraction, moving target prediction are the key to moving target detection and tracking technology. Aimed at above problems, this paper carried out related research.Firstly, this paper studies the target detection algorithm based on mixture Gaussian model. Because of illumination change, shadow, camera dithering, the image of shooting has a lot of noise, so need to deal with this noise. First uses the adaptive median filter technology to eliminate noise interference, reoccupy update background template the mixture Gaussian model to detect the moving targets, final morphology filtering and threshold area method are used to eliminate the residual noise, this can be achieved for moving target detection.Secondly, the article mainly studies the scale invariant feature transform algorithm(SIFT algorithm). Classical SIFT algorithm can keep good invariance due to scaling, translation,rotation, scale brightness changes, but there are still some problems. Aim at the computation time of SIFT algorithm is too long, this paper proposes an improved SIFT algorithm. New algorithm introduces a Mallat fast wavelet transform algorithm, reconstructed low-frequency components of the image, in order to improve the matching speed; and then revised the number of the Gaussian pyramid group, reduced the number of down-sampling and Gaussian difference pyramid layers; at last removed the mismatching points by improved RANSAC algorithm. Experiments show that the improved SIFT algorithm has stronger robustness and better real-time performance, better than the original SIFT algorithm.Then combining the improved SIFT algorithm and Kalman filter algorithm to realize the tracking of moving target. Firstly use Kalman filtering algorithm to detect moving targets, and predict the centroid position of moving targets in next frame; then deed the centroid position as the center, to create a search area, in the search area, matching the template image and the input image by using improved SIFT algorithm, to select the most matching target for tracking; regard this position for observation at the same time, optimizing the Kalman parameters, so as to realize the tracking of moving objects.Finally, the article introduces Compressive Tracking algorithm based on simple features(CT algorithm) and Fast Compressive Tracking algorithm(FCT algorithm). FCT algorithm has higher timeliness compared with CT algorithm, but when the target larger texture, the tracking is not stable, easy to lose the goal. In order to solve this problem, the new algorithm does an improvement of the original random projection, and does pre-determination of the candidate sample. The new algorithm tracking stability is improved, and has better real-time.