This paper studies P4P localization algorithm based on quick response (QR) code recognition, and proposes to use the coordinates of the four coplanar vertices stored in the QR code and the pixel coordinate in the QR code images to locate. The pixel coordinates of the four coplanar vertices of the corrected QR code are obtained through camera calibration and image correction, and the actual position coordinates of the four coplanar vertices are obtained through decoding program analysis. The rotation matrix and translation vector output by the P4P algorithm are used as the following iterative initial values of the next algorithm. In perspective N points projection algorithm, the optimization on manifold is converted into optimization in tangent space by means of local isomorphism between Lie group and Lie algebra, and the optimization method in linear space is used to optimize the target value. In any gradient direction of the tangent space, the parameter value is in the space, that is, the matrix is all on the manifold. Experimental results show that during the navigation movement, the camera coordinate position error is less than ±2 mm, and the rotation angle error is less than ±0.5°, showing that the algorithm has high positioning accuracy and stability.