In this paper, we propose an improved ORB hybrid algorithm based on the traditional ORB algorithm, which optimizes the method of feature point detection and feature direction calculation, then achieves fast and efficient matching by FLANN -based fast nearest neighbor matching, and finally optimizes the matching effect by using RANSAC algorithm. The Virtual assembly system based on augmented reality technology superimposes the 3D model of the robot with the virtual information in the equipment to improve the assembly efficiency of the staff. Experimental results show that the augmented reality system based on the proposed algorithm has good performance, and the real-time performance and alignment accuracy are improved.