Object Detection Based On Motion Compensation And Global Background Optimization

Real-time monitoring which is a modern city management tool has been more useful in recently years, but a lot of drawbacks still exist in the traditional monitor mode. Therefore, intelligent video analysis technology in recent years has been a great development. Moving object detection is an important basis for intelligent video analysis technology. Researching moving object detection is very important for video analysis.In recent years,moving object detection technology has developed extremely quickly, many algorithms with good performance have been invented. These algorithms have solved and meet some needs of object detection in the surveillance scene. But due to the limit of surveillance scene, these algorithms only can work well when background is static. When background is dynamic, they have poor performance. Because of the moving of camera, the background and foreground are all with the motion attribute, so it is difficult to separate them. For this reason, moving object detection in dynamic background is a challenging subject in the field of object detection.The main work is as follows:(1) For moving object detection in dynamic background issues, we proposed a model to update the background of online algorithm based on motion compensation method and detection of moving objects with a fast way. Specifically, the first we established for each pixel of a pixel-level background model. On the basis of the previous frame background model, we estimate the motion of each pixel with edge preserving filter by Optical Flow algorithm and propagated to the background of the current frame model, judging each pixel is a former attraction or background by the background of the compensation model. Finally, we used a quick random algorithm to compensate for online updating background model to adapt to changes in the background. We have collected a large number of experimental videos and verified advance of the algorithm by experiments.(2) For the widespread phenomenon of false detection of pixel-level object detection algorithm, we have come up with a global optimization method for object detection.The method used Gaussian mixture to model the global background in super pixel Layers,using the global appearance of the background to optimize the pixel level object detection method. Through the experiment, the optimization results of the pixel level object detection method are demonstrated, which proves that this method has excellent optimization ability. Eventually we combined it with the detection of motion compensation method,constructing a local to global object detection method,By comparison, we proved this detection method has a better performance in dynamic scenes.