Vehicle Detection Technology Based On Stereo Vision And Monocular Recognition

In recent years,the automotive industry is undergoing major changes,and the intelligentization of vehicles has become the focus of the development of the automotive industry.For intellegent vehicles,environmental awareness technology is the foundation for vehicle routing and decision-making techniques.Because the stereo vision sensor is low in cost and can provide rich picture texture information and depth information,which helps to reduce engineering development cost and improve target detection accuracy,this paper uses a combination of stereo vision and monocular recognition to detect vehicle.Firstly,this paper deduces the imaging model of binocular stereo camera to get camera parameters and disparity map.Establishing the mathematical model of three-dimensional reconstruction of stereo vision to provides accurate data source information for subsequent measurement of target distance based on the theory of stereo vision.The statistical principle is applied to the disparity map to construct the V-disparity map.Using random sample consensus(RANSAC)algorithm to fit the inclined line formed by the ground plane projected on the V-disparity map,and the plane equation of the road surface is obtained,so the disparity values of the ground and the sky can be estimated.The road score and obstacle score of each pixel point are calculated to generate cost image.The intersection points of each row of road and obstacle are obtained by dynamic programming.The height of obstacle is determined by membership function,and then the obstacle area is determined.Secondly,based on the obstacle area,the supported vector machine(SVM)is used to train the histogram of oriented gradient(HOG)feature to identify the target vehicle.The obstacle area detection result is combined with the support vector machine to optimize the target vehicle obstacle.The experimental results show that SVM combined with HOG features to detect vehicles in front with high accuracy for ZED cameras.Finally,A method for calculating the distance of a target vehicle relative to the subject vehicle in a three-dimensional coordinate system based on histogram statistics is proposed.The distance value of the pixel in the target retangle of the vehicle is counted and the distance calculation area is determined.The distance between the target vehicle and the subject vehicle is calculated by calculating the center of the histogram area.Although the introduction of the histogram improves the accuracy of the distance detection,the calculation result is still rough.Therefore,the Kalman filter algorithm is introduced to optimize the target vehicle distance.The experimental results show that the optimization algorithm effectively improves the accuracy and smoothness of the distance measurement.The first-order difference is calculated for the distance to calculate the target vehicle relative to the vehicle speed.The accuracy and real-time performance of the obstacle detection algorithm distance detection algorithm is verified by establishing the software and hardware configuration required for the body vision system.