Research On Position And Attitude Calculating Of Vehicle Based On In-vehicle Video

In recent years,traffic roads have been built more and more normal in China.This status provides the necessary preconditions for the application of computer vision technology to the position and attitude of vehicle.Vehicle camera calibration,3D reconstruction,position and attitude calculations are the basic technologies of calculating position and attitude of vehicle based on visual methods,and they are also the key technologies that affect the calculation accuracy of vehicle position and attitude.This thesis study the position and attitude calculating of vehicle based on in-vehicle video,which based on the mentioned three technologies and the structural characteristics of the road,the research work mainly includes the following aspects:(1)Due to vehicle and environment dynamics when the vehicle is driving,the extrinsic parameters of the in-vehicle cameras could change dramatically and dynamically.In traditional methods,the extrinsic parameters of the in-vehicle camera are usually calibrated in advance,which lead to large computation errors of position and attitude of vehicle.In this paper,three-line calibration method is used to online calibrate the extrinsic parameters so as to minimize the impact of road and vehicle dynamics.From the extrinsic calibration results,vehicle Position and attitude is thus computed by applying the principle of projective geometry and vanishing point.The proposed method has been tested with the actual in-vehicle images collected in different road experiments by setting up the in-vehicle camera with different heights.The experimental results show that the average error of vehicle traversal localizations is 7.3 cm and steering angles is 1.5 degrees in different road conditions,which demonstrate that the proposed localization method is both accurate and adapted by applying the online three-line calibration method.The results also demonstrate that the proposed method outperforms traditional methods by assuming fixed extrinsic calibration parameters.(2)With the improvement of the intellectual development level of road assets and the increase in the demand for road network surveys,it is particularly important to calculate the curvature of the curve quickly and accurately.This thesis proposes a vision-based horizontal road curvature calculation method.Based on this,it also proposes a vehicle position and attitude calculation method based on curve lane lines.Firstly,the sequential video log images are adaptively used to reconstruct the curved lane line.Then the error point of the reconstruction is removed according to the error analysis model.And then the improved ICP algorithm is used to achieve accurate curve registration.Finally,the curvature calculation of curve based on the circle fitting,the position and attitude calculation of vehicle based on visual are realized.Using the proposed method,experimentsare performed on different degrees of curves.The results show that the proposed method can accurately calculate the curvature of the curve and the position and attitude of vehicle,and has a good practicality.(3)Aiming at the problem that the position and attitude of vehicle cannot be calculated using the lane line of the curve When the vehicle is driving on a curved road with blurred or no lanes.This thesis proposes a method for calculating the position and attitude of vehicle based on GPS and road characteristics in curved road.Firstly,a smart car and a roadside laser are used to build a curve map database.And then GPS is used to realize the initial positioning of vehicle in curved road,which matches the GPS information of the vehicle with the curve GPS information sequence in the curve map database.Then based on the ORB global features of the curved road surface image,the image-level positioning of cornering vehicles is achieved,and the most matching image in the curve map database is obtained.Finally,this thesis uses a visual method to achieve a measured level of vehicle position and attitude calculation at the corner.The experimental results show that this method is universal,the position and attitude of the vehicle at the curve are calculated with high accuracy.