Pedestrian Detection Research Based On Color And Depth Image

With the development of science and technology, the pedestrian detection technology became a basic research in the field of the machine vision. The developing pedestrian detection technology has also gradually appeared in the daily life, and it has been widely used especially in the intelligent video surveillance, automotive safety auxiliary driving, intelligent household and other fields. Modern people is paying attention to the public security and having urgent need to improve the quality of life which makes the pedestrian detection:echnology have a broad application prospects and commercial value, and more and more people will be engaged in the research.The current pedestrian detection methods are generally use RGB color camera, which are most widely used in real-life applications. But with the Ievelopment of sensor technology, other types of sensors are also increasingly ised for get pedestrian image data. The most representative is KINECT sensor, vhich makes the pedestrian detection method based on-multiple source sensor nformation fusion become a research hotspot.Aiming at the problem that current RGB color image pedestrian detection methods exists, such as slow speed of HOG method and too much redundant information in CSS method, this paper proposes a new method of improvement. First described the self-similar characteristics of images in the important areas of HSV space by adopting the idea of CSS, and calculate CSS features between adjacent small pieces in the important area. This method then identify the typical area of high stability in the image edge features through statistical methods, and extract the HOG feature of each area, combined with the two kinds of features as pedestrians descriptor. Finally SVM is employed to learn the pedestrian characteristics and classification in order to realize the pedestrian detection.This article also put forward a pedestrian detection method in view of the RGB-D image data obtained by the current depth sensor technology. This article will combine all the sensors and the features fusion, first extract edge image in the depth image and match the standard head and shoulders of the templates, which will eliminate a large number of candidate window in order to improve the detection speed through matching algorithm; Secondly this method extracts depth map image and color image characteristics respectively from the window of the head and shoulders templates which matched successfully. Finally, this method uses multi feature fusion method to detect pedestrians in order to ensure the accuracy of pedestrian detection.