Aiming at the problems of existing calibration methods, such as high cost, complicated steps, little robustness and low precision, a stereo vision inspection system is established, and a secondary calibration method based on high-precision special-shaped gauge block is proposed. The error of the system is compensated based on GA-BP algorithm. Firstly, the transformation between the Pixel coordinate system and the world coordinate system composed of a single CCD and a laser is solved by the quadratic calibration algorithm. Then, the method of laser fringe extraction based on Heisenberg matrix is used to extract the center line of the edge line on the light plane where the light plane of the laser projector intersects the high-precision special-shaped calibration block. Finally, GA-BP neural network algorithm is used to compensate the errors of the data collected by the stereo vision inspection system. The experimental results show that the precision of the special-shaped block is less than 0.05 mm after the width and depth compensation, and the stereo vision detection system is stable, reliable, high precision and strong anti-interference ability.