Abstract: A high efficient lighted beacons registration algorithm is proposed to address the limited texture features of cooperative lighted beacons and the depth loss issues with monocular cameras. The algorithm is developed based on the actual distribution environment of beacons within a cubic stereo frame. The algorithm decouples the 2D-3D beacons registration into non-iterative registration of 2D-2D reference plane beacons and reprojection estimation registration of non-reference plane beacons. In the registration of the reference plane, a set of identification reference angles composed of several lighted beacons and their extraction algorithm are designed, ensuring the robustness of the algorithm through the establishment of a weighted connection graph and the search for the maximum cycle. The non-reference plane registration completed after plane registration to calculate the beacon reprojection points by pose estimation, and registration is completed by nearest neighbour search. Simulation experiments confirmed that the algorithm could adapt to a wide range of camera view angle variations, and it is able to complete the registration of 150 lighted beacons with 100% accuracy in an average of 150ms.