The Implementation Of The Pedestrian Detection System Based On Opencv

Pedestrian detection has been paid more attention by the researchers in the field of computer vision. Summarily, pedestrian detection can automatically analyze the sequenced images that captured by the camera real-timely to judge whether people is present. If there is someone in the images, it can be labeled by some way and give the prompt for next step. Pedestrian detection has wide applications in many fields, such as vehicle safety assist system, which can detect pedestrian and notice the driver in advance to take action. Besides, pedestrian detection can be used in the intelligent transportation surveillance system, surveillance system of the hotel hall and so on.However, because of the variety of background, the variety of illumination, overlap of people, different postures of people, and the variety of people’s size, color, clothes, etc, the research of pedestrian detection is a difficult research direction. All the above factors bring a lot of challenges to the development of pedestrian detection. But pedestrian detection still gets enormous improvement after many years of research, and more and more advanced algorithms are proposed to enhance the speed and the precision of the detection.In this thesis, we build a whole pedestrian detection system by using a list of tools, but not pay much attention to the algorithm of pedestrian detection to improve the Detection speed, precision, etc. In the thesis, the Haar features of the sample images were extracted and to train classifier with the AdaBoost algorithm, then the classifier was applied in the pedestrian detection system to discriminate pedestrian and non-pedestrian in the video.Here are the main contents in this thesis:1, Build the graphic user interface of the system with the MFC and OpenCV in the environment of VC6.0. This system supports the video which comes from AVI file or camera, and provides the functions of open, play, pause, stop, detection, exit and so on.2, Train the classifiers off-line. Extract the Haar features of the sample images, train the positive samples and negative samples with AdaBoost algorithm to obtain two classifiers, which are used to detect frontal/back and side pedestrians. 3, Apply the two classifiers obtained by training into the pedestrian detection system, and test the actual performance of the system by different video inputs.From the experiments, it’s known that this pedestrian detection system based on OpenCV can discriminate pedestrian and non-pedestrian well. The system can be used directly in the situation that the precision requirement is not very strictly, such as surveillance system in the hall of hotel and so on.