Research On Vision Measurement And Vehicle Positioning Based On Roadbed Camera

With the development of 5G technology and Internet of Vehicles technology,driverless cars,as an inevitable trend in the development of today’s vehicles,have also become one of the hot issues studied in recent years.The positioning of an unmanned vehicle is mainly used to determine the location of the vehicle,which is an important prerequisite to ensure that the unmanned vehicle completes the driving task.At present,the positioning of unmanned vehicles in outdoor environments mainly relies on high-precision GPS,multi-line lidar,cameras and other sensors for sensing and positioning.High-precision GPS can obtain high-precision global positioning under open outdoor conditions,but there are problems with poor local positioning accuracy or loss of positioning information in indoor or outdoor sheltered environments,which cannot meet the positioning requirements of unmanned vehicles.This paper studies the centimeter-level positioning of unmanned vehicles,and proposes a method based on camera calibration and feature point extraction to improve the positioning accuracy of unmanned vehicles.This positioning method can solve the problem of vehicles in a wide range of indoor and outdoor scenes.To locate the problem,the specific research content is as follows:First,multiple roadside cameras are used to obtain the moving images of all vehicles in the field of view,and the collected video information is separated from the foreground and the background.Based on the analysis of the advantages and disadvantages of the traditional moving target detection method,the Gaussian mixture based on the background modeling method is adopted.Model to extract moving vehicles.Second,due to the limited resources of the computing platform on the vehicle,in order to make the system real-time,the extracted feature points have good scale invariance and rotation invariance.The ORB algorithm is used to extract and match the feature points of the multi-angle images in the same area,to associate the data between the two pictures,and to filter the matching points through the correspondence of the feature points in different frames of pictures.Then,the linear model of the camera is established to calibrate the road-side camera to solve the corresponding internal and external parameter matrix and to correct the distortion of the camera.The multi-eye matching algorithm is used to match the feature points of multiple images at the same time.According to the binocular vision measurement the distance principle uses the least square method to convert the two-dimensional image coordinates into three-dimensional space coordinates and calls the PCL library function to visualize the point cloud data,draw the three-dimensional bounding box of the point cloud,so as to realize the tracking and positioning of the unmanned vehicle.Finally,the actual vehicle experiment verifies the positioning system of this article,and selects the appropriate camera,experimental vehicle,workstation and venue for the actual vehicle experiment.Measure the specific real position of the vehicle,compare it with the vehicle position output by the program,test the accuracy and reliability of the program,at the same time,discuss and analyze the experimental results,and put forward some suggestions to improve the positioning accuracy of the system.