| Summary: | 碩士 === 臺北市立教育大學 === 數學資訊教育研究所 === 94 === We proposed a simple and robust method to detect and classify the intruders in garages. In order to detect the moving objects, we can not just use the traditional background subtraction method in the environment of parking space because of its complex scene. Instead of using statistics to train out a background model in the past approach, we use a method which is less complexity and don’t need any pre-saved background image to patch and update the background.. Then find out the motion information of moving objects in the frames by background subtraction. The proposed method also can determine a proper updating rate dynamically by combining the motion information of the moving objects, which can update the background image precisely and detect the “moving” objects in the scene and not only “changing” in the scene. After the detecting of moving objects, we classify the objects into human and vehicles according to its spatial geometric relationship and skin tone in color space.
For the whole system efficacy, the system can process up to thirteen frames pere second by using general web camera to process the input images which frame size is 320*240. It is superior to other method which can process only nine to eleven frames. When learning rate of background updating is on level nine (α=56), it only needs 4 frames to finish its updating. It is much quicker than the method built up by statictical which needs 60 to 120 to finish its updating.
|
|---|
