Pedestrian Detection Method And Device For Complex Traffic Environment

With the popularity of camera performance and the rapid development of artificial intelligence, natural images in the field of intelligent transportation has been widely used.Pedestrian detection due to its significant impact on the traffic safety of the vehicle,has been the core of intelligent transportation field. However, due to pedestrians target itself non-rigid properties, and complex external interference, such as lighting transformation,shelter,environment,Pedestrian detection is still the difficulty in the field of image.This paper summarized the pedestrian detection technology development and research status, introduces in detail the main pedestrian detection data collection,classification, detection operator method, and analyzes the main problems.Next,This paper reports a new pedestrian detection method that combines single and couple pedestrian DPM models in order to solve the pedestrian detection problem due to occlusion of pedestrian targets under common traffic scene;Design a set of real-time pedestrian detection device for intelligent vehicle; Finally to summarize work prospects.In this paper, the main research work and achievements are as follows:1、A double block against each other in the case, according to the blocking probability statistics, extracted with dual training for pedestrians database from TUD-Brussels and ETH database.2、Based on DPM(deformable parts model) detection operator and Latent SVM training method,from INRIA pedestrian databases and database respectively double training people to give a single-line SP-DPM category template and double pedestrian DP-DPM category template.Divided by the method of linear, double pedestrian classified deformed template contains a single row of double pedestrian person member template SDP-DPM.3、Introducing a new pedestrian detection method that combines single and couple pedestrian DPM models in order to solve the pedestrian detection problem due to the feature information of the distribution on the pedestrian traffic scene.4、Through the tachograph captured video to extract the relevant pedestrian database.Training effectively pedestrian classifier,pedestrian detection method based on integral figure HOG, set the detection distance and the corresponding multi-scale detection method.5、Proposes a reduction of the detection area based on diagonal slope ROIalgorithm,By the method of GPU acceleration has realized the real-time detection of the target of pedestrians.6、Based on VS2010 compiler environment and Opencv2.0 interfaces, designed and implemented a complete visualization pedestrian detection system devices.