Optimization And Application Of Camera External Parameters For Automated Vehicle

With the rapid development of self-driving technology and vision sensor,the optimization and application of camera external parameters have gradually developed into one of the core tech-nologies in this field,which has also attracted the attention of academia and industry.Autonomous vehicles need to be able to perform tasks independently,which will improve the safety and relia-bility of urban road traffic systems and reduce environmental pollution and alleviate congestion.In the process of autonomous driving,how to effectively improve the perception ability of visual sensors and provide their own reference positioning objects is still a great challenge.This paper will study the optimization and application of camera external parameters.In this paper,the external parameters of the camera are self-calibrated considering the plane constraint for the self-driving vehicles in real scenes.Based on the multi-view geometry,a new model is derived through the relationship between image frames,and the rotation matrix is de-composed into two parts.Under the condition that the non-integrity constraint of the vehicle is satisfied,a two-step movement is taken to solved the parameters respectively.In addition,by in-troducing the virtual camera,the translation vector in camera external parameters can be solved with higher accuracy and faster speed.Especially when the optical axis of the camera is parallel to the motion plane,we do not need to solve the pose relationship between the camera frames,so we can obtain the external parameters of the camera without knowing the internal parameters of the camera.The simulation experiment is designed,and several related algorithms are compared.The results show that the solution precision of the algorithm is very good.Experimental results in the real scene also verify the effectiveness and simplicity of the algorithm.In real scenes,the dynamic obstacles such as the moving vehicle on the road,the branches swayed in the wind beside the road,the random sample consensus(RANSAC)algorithm is im-proved by using the motion characteristics of the vehicle.Based on the proposed model,a pair of matching points are used to optimize the results.Compared with the result of translation estima-tion based on the epipolar geometry model,the speed and robustness of this algorithm are greatly improved.After obtaining the external parameters of the camera,it is applied to the lane detec-tion to assist the autonomous vehicle to locate itself.The lane is extracted by inverse perspective transformation,and a lane detection method based on camera external parameters optimization is designed,which includes image preprocessing,dynamic region of interest selection,image en-hancement,inverse perspective transformation,edge detection,lane fitting and other parts.The experiment shows that the accuracy and efficiency of lane detection can meet the performance requirements in the process of automatic driving.