Research On Camera Auto-Calibration Algorithm Based On Vanishing Point And Model Matching

In the traffic scene,camera calibration is one of the prerequisites for computer vision in intelligent transportation systems.The camera must be calibrated to obtain more accurate and valuable traffic parameters,such as the speed and the spatial position of the vehicle.A geometrical mapping relationship can be established by using the intrinsic and extrinsic parameters of the camera to construct a relationship between the two-dimensional information of the image and three-dimensional information of the space.Currently,the auto-calibration algorithms for cameras are mostly based on vanishing points and various types of traffic markers.However,these methods have some problems: the vanishing point is unstable only by the intersection of lane lines or lane boundaries;the acquisition of calibration markers is affected by natural conditions;the vanishing point tends to infinity at a certain range of angle.In view of the above problems,this paper uses the vanishing point and model(vehicle model or lane line model)matching method to obtain the intrinsic and extrinsic parameters of the camera.The specific researches are shown as follow.Firstly,this paper simplifies the camera model.According to the characteristics of the scenes tested in this paper,the camera model is simplified,and the appropriate world coordinate system is established.The camera parameters are simplified to focal length,pan angle,tilt angle,and camera height,which facilitates the design and implementation of subsequent autocalibration algorithms.Secondly,this paper detects calibration markers and obtains vanishing point along the road direction.For vehicle targets,this paper uses the Single Shot Multibox Detector(SSD)depth learning algorithm for detection.For the vanishing point along the road direction,this paper tracks the vehicle targets to obtain the trajectory set,linearizes the trajectory set and obtains the vanishing point through cascading Hough transform.For lane line markers,this paper uses the vanishing point along the road direction to extract the motion region,and the Hough transform is modified to identify the lane line based on the motion region.Finally,this paper constructs the models of auto-calibration and researches related algorithms.Due to the diversity of traffic scenes,the calibration markers extracted from each scene are inconsistent.And the known information is different,so it is difficult to calibrate using a common calibration model.Therefore,for different scenes,a lane line model or a vehicle model is constructed.And different auto-calibration algorithm is used based on VVH,VVL or VLH.The algorithms proposed in this paper have been extensively tested in given traffic scenes and the accuracy of auto-calibration has reached more than 90%.This paper makes reasonable analyses of the experimental results,explains the causes of the errors and gives optimization methods.The obtained the intrinsic and extrinsic parameters of the camera are used to calculate the speed and spatial position of the vehicle and estimate the three-dimensional size of the vehicle.