Pedestrian Detection Technology And Its Application In Surveillance Video

Traditional surveillance system only has video capture and storage capabilities. Therefore, it wastes a lot of front-end computing resources for its over-dependence on the camera hardware. What's more, unstructured video data have a great demand on labor costs. Pedestrian is one of the most important components of intelligent surveillance data, because it is the main cause of accidents. Researching how to de tect pedestrians traveling from videos has great significances to the intelligentization of surveillance systems. In this paper, we propose a fast pedestrian detection method in surveillance videos. In this paper, we do the research mainly from the basic method to detect pedestrians, the offline training of detection classifiers and pedestrian detection in videos based on priori information of motion. Finally, we come up with an application of this method in intelligent surveillance, i.e., structured storage and retrieval of pedestrian information in surveillance videos. In summary, the main contents of this paper include the following aspects:(1) We give a comprehensive analysis of several pedestrian detection methods state of the art, and ultimately selec t Integral C hannel Features based on cascade Ada Boost classifier as the foundation pedestrian detection method in videos. Finally, we evaluate the performance of our method and determine its application environments by a large amount of environments.(2) Aiming at the characters of Ada Boost classifiers and the diversity of negative samples, we propose a weight- loss control sampling method for the offline training of pedestrian detectors. It accelerates the offline training process and promote the accuracy of detectors. The detectors trained by this method lay a foundation of the following detection process.(3) We detect motion in videos by a improved Vi Be method by ourselves. When detecting the pedestrians, we use priori information of motion to block the detection region and restrict the features extraction area. By this way, we minimize the interference of complex background information and accelerate the pedestrian detection speed.(4) We propose the integrated framework of pedestrian retrieval. In order to realize it, we design modules of pedestrian detection in surveillance videos, structured storage of pedestrian data and pedestrian retrieval based on color histogram features. We conduct an exploratory research on the application of pedestrian detection technology in intelligent surveillance.