Research And Application Of Camera Self-calibration Optimization Method In Traffic Scene

In the intelligent transportation system,the traffic monitoring video is an important source of information.Obtaining scene information from the video requires establishing a suitable camera model and calibrating the camera.The purpose of the camera calibration technology is to establish the mapping relationship between the image and the three-dimensional space according to the internal and external parameters of the camera.The stable automatic camera calibration technology is an important guarantee for the extraction of traffic parameters.Most of the current popular camera calibration methods are based on the combination of vanishing point and scene identification information.However,in the actual scene application,there will be problems such as difficulty in detecting scene identification information and the appearance of "illness" in the vanishing point,resulting in unstable calibration results.In view of the existing problems,this paper proposes an optimization method of adaptive matching camera calibration model based on vanishing point.It uses effective information to complete the initial calibration for different scenarios,and establishes a constraint function through the redundant traffic sign information in the scene to perform the calibration results.When the camera calibration is completed,the application based on the camera calibration results is realized.The work of this paper is mainly reflected in the following aspects:1.Stably detecting a set of orthogonal vanishing points and scene traffic signs in the scene.In this paper,by establishing the cascade hough transform,a large number of filtered vehicle tracks and lateral edges are transformed into the parameter space for local maximum accumulation.Thus,a group of orthogonal vanishing points in the scene can be obtained.This method can reduce the influence of lane bending on vanishing points and avoid the error of detecting only by using the information of a single scene.At the same time,the vanishing point and vehicle trajectory are combined to delimit the road flat activity area,and the traffic identification information is detected in the area.This method can reduce the influence of background environment on the identification information detection.2.An adaptive matching camera calibration model was established to avoid the influence of "ill-conditioned" vanishing point on calibration results.Due to the limitation of the scene and special camera angle,the vanishing point in the scene may infinity.Therefore,in this paper,an adaptive model of camera minimum calibration condition is proposed to initialize camera parameters according to different scenes and the known scene information,this makes the calibration method more widely used in the scene.3.The open optimization method is used to optimize the parameter space of the calibration model by using the redundant traffic signal information.When the initial calibration is completed with the minimum calibration condition,the information physical value and the actual physical value are identified by comparing the inverse redundancy,and the posterior error constraint function was established,the calibration results were optimized by several iterations,and the accuracy of the calibration results was more than 95%.4.Traffic video and camera calibration results are used to implement the traffic system.After completing the camera calibration,the application of vehicle space position calculation and vehicle speed calculation is realized by using the internal and external parameters of the camera.By comparing the speed calculation with the standard data set,the accuracy of the speed calculation is more than 92%.