| Summary: | 碩士 === 國立臺北科技大學 === 自動化科技研究所 === 106 === In this thesis, the pedestrian detection on the motion images requires faster computation efficiency than the static images. When detecting a full-frame image, the pedestrian only occupy a small portion of the image. Wasteful operations are applied on a large portion of the image. By detecting the moving object and performing pedestrian identification on that object, the operation speed can be effectively improved. The proposed system architecture is divided into two parts, namely, moving object detection and pedestrian identification. The object detection uses a Gaussian mixture model (GMM) to establish the background image model. The preliminary foreground image is then extracted by background subtraction method. Through a h filter, the noises are removed . Contour detection and edge detection are performed. Finally, the moving object in the continuous image can be identified. To identified the moving pedestrian, the Haar-like feature is extracted from the positive and negative samples of the head and shoulder images. The cascade classifier is trained offline using the AdaBoost algorithm. The classifier is used to identify the head and shoulder parts of the moving object. If the head and shoulder information is identified, it can be judged as a pedestrian. The system uses the OpenCV library which can effectively help the development of image operations. The experimental results show that the algorithm retains useful information, reduces useless calculations, and achieve higher detection speed and accuracy.
|
|---|
