Pedestrian Detection And Distance Measurement Between People And Vehicles Based On Binocular Stereo Vision

In recent years,with the continuous development of technology in the field of artificial intelligence,AI technology has been gradually applied to various fields.Among them,the intelligent assisted driving system is an application of AI technology in the automotive industry,and it is also the foundation of unmanned driving technology.The intelligent driving assistance system sense the information of the road and the areas around the road,simultaneously process these perceptual information,by utilizing computer vision technology.This can reduce the incidence of traffic accidents caused by human factors and further guarantee the property and personal safety.The pedestrian detection in front of vehicle in the intelligent assisted driving system is a major technology relating to the safety of the pedestrians.A real-time measurement of the distance of pedestrians in front of the vehicle by the binocular stereo vision technology is also an important part of it.Therefore,the detection and distance measurement of pedestrians in front of vehicles is particularly pivotal.In view of the above background,this paper mainly studies the technology of pedestrian detection and distance measurement of pedestrians in front of the vehicle,which is based on the binocular stereo vision method.Based on the relevant research of binocular stereo vision theory,in this paper we use binocular camera and computer to build an image acquisition experiment platform to collect pedestrians' images in front of the car.On the one hand,the internal and external parameters of the camera are obtained by calibrating the binocular camera.On the other hand,For the noncoplanar pedestrian image in front of the vehicle,the stereo correction experiment is performed by using Open CV,which is under the Visual Studio 2012 platform.In the pedestrian detection section of this article,the pedestrian binocular images of pedestrians in front of the vehicle is detected a regression object detection and recognition method based on SSD(Single Shot Multi Box Detector)algorithm,for the problem of the poor performonce when detecting a long-distance small targets,in this paper,an improved SSD algorithm based on upsampling feature fusion is proposed.This algorithm improves the SSD algorithm network,increases the feature information of small targets,and improves the detection accuracy of long-distance small pedestrian targets.In the distance measurement part of pedestrians in front of vehicles given by the binocular stereo vision vehicle,we compare the feature matching algorithm SIFT algorithm,the SURF algorithm and the ORB algorithm.In order to solve the problem of real-time long-distance measurement,a distance measurement method based on feature matching of the center point of the binocular vehicle pedestrian image rectangular frame is proposed in this article.This method puts the difference between the horizontal coordinate of the center point of the pedestrian rectangular frame in the left image and the horizontal coordinate of the center point of the pedestrian rectangular frame in the right image as the parallax,and combine with calibration results of stereo vision and the triangulation principle of binocular stereo vision,to calculate the distance of pedestrians in front of the vehicle.In addition,a comparison between the calculation results and the actual distance is conducted,and some analysis with respect to these comparisons are given accordingly.