Moving Object Extracting And Tracking In Video Sequence

With the rapid development of computer performance and the progress in image processing technology, moving object extracting and tracking in video sequences has become an extremely important field of computer vision, attracting more and more scholars'interest. Moving object extracting and tracking can be used to assist the informatization of many industries such as human-machine interface, transportation management and smart surveillance. Moving object extracting and tracking is also the basis of image analysis and image understanding. Therefore, the study of moving object extracting and tracking has important theoretical significance and application value.Moving object extracting and tracking in video sequence consist of two main parts: extracting object accurately and tracking object stably. In the object extracting aspect, After analyzing several algorithms'advantages and disadvantages, the paper proposed a background difference method mixed with hue and grayscale information. Background difference was done in both hue space and gray space. The pixel whose changes in hue space and gray space were both greater than their respective thresholds was considered a part of the moving object. After that, mathematical morphology filter was used to enhance accuracy of object extracting. In the object tracking aspect, since the current Mean Shift algorithm often fails when tracking moving object whose grayscale is close to the background, the paper proposed an improved Mean Shift algorithm mixed with texture, grayscale and motion prediction. Bhattacharyya coefficient was used as similarity measurement function. Firstly, object model was established with LBP/C texture and grayscale features. Secondly, the most likely area the object may appear in was predicted with Kalman filter. Lastly, Mean Shift algorithm was used to search the moving object in the predicted area.Experiments showed that the object extracting method can eliminate a large number of interference and detect the moving object accurately. The improved object tracking algorithm can still work well even when the object's grayscale is similar to the background. The algorithm has improved tracking accuracy.