| Summary: | 碩士 === 國立高雄應用科技大學 === 電子與資訊工程研究所碩士班 === 94 === Video segmentation is key role for developing technique (e.g. index-retrieval, compression or representation) of content-based video processing. In practically, it can be implemented in pre-processing for contend-based video system in order to separate the video into many video objects. Many proposed video segmentation algorithms which are aimed at specific sequence (e.g. indoor environment) or outdoor environment in clear day. However, there restrictions hardly make it to be invalid in bad situation. In this dissertation, we propose a video object segmentation algorithm based on edge and color features combined edge detection and change detection in rainy situation that avoidance accuracy of segment video object reduced in bad environment. The characteristic of moving object will be obtained by HSI color transform and analysis. The edge detection is used to obtain edges of moving object from video frame, which raindrops can be avoid to decision to moving object in dynamic background due to a large number of raindrops moving at high speeds. Then, the object region detection is used to get correct object mask and can solve the uncovered background problem and still object problem. After above step, a problem can be solved that the reflection region of moving object in the environment will be removed by bounding box match method. Finally, subjective and objective evaluation of this algorithm is showed and demonstrates spatial accuracy of our algorithm can maintain a high accuracy above 80 percent in scenes captured by the fixed camera.
|
|---|
