PTZ camera calibration based on improved DLT transformation model and vanishing Point constraints

Abstract

High-precision camera calibration is the key to obtaining accurate 3D information of the target from the image. Due to the shortcomings of both the classical simplified linear model and the complete distortion model in the algorithm, a PTZ camera autocalibration method is proposed, it is an improved DLT method based on three vanishing point constraints and virtual identification point constraints. This method uses the two vanishing points obtained by cascading Hough transform of vehicle trajectory line and car body edge line and the third vanishing point obtained by geometric relationship of three vanishing points to complete the initial parameter calibration of the camera, and calculate the intrinsic parameters, rotation matrix and translation matrix in order. Then artificially filtered spatial virtual identification points are used to establish a camera parameter constraint with the actual scene non-virtual reference points, and the DLT direct linear transformation method is used for final correction. This study is tested under different rotation angles and pitch angles of traffic scene, and it uses the differences between the actual physical distance of lane line in the scene and its tested distance as the measure of the calibration error. The error of proposed method at different conditions is less than 2.5%. The test results of multiple sets of highway scenes show that the method proposed in this paper can meet the application requirements for the physical measurement error of the highway, and it has greater accuracy and better stability compared with the reference method. The obtained camera parameters can be used to accurately calculate vehicle speed and estimate spatial position.