Motion Detection Of Video Objects In Camera-fixed Scenes

This thesis mainly focuses on the application of video surveillance based on background subtraction under immovable scenes, specifically including motion detection, background model, shadow detection and motion estimation.Firstly, motion detection is the basic of video applications such as computer vision, object recognition and tracking, surveillance security etc. Background subtraction is an important approach of motion detection, but it is sensitive to illumination and shadow. A novel background subtraction method using HSV color space and integrating temporal difference is proposed. The experimental results show good robustness. Secondly, background image usually changes as time goes by and the environment changes, which makes some parts of the background misclassified as moving foreground object. Therefore, it is necessary to model background to adapt the changes of background. In this thesis, we use the K-Gaussian mixture model as the background model, not only adaptive for the changing background, but also easy to establish a background model in clutter or busy scenes.Thirdly, motion is usually along with the occurrence of shadow, which impacts the correctness and accuracy of motion detection. Furthermore, there are two same important visual characters between a shadow and its corresponding object: a shadow is obviously different from the background, so it is easily detected as foreground; shadow and its corresponding object have same movement. These two features make shadow detection very challengeable. In this thesis, we improve the shadow detection method used in Sakbot through adding weights to the conditions of shadow detection, which gives a satisfactory result.Finally, motion estimation is useful to forecast the motion at next time. By integrating motion estimation into motion detection, the latter will be...