Abstract:An Improved Approach of indoor space three-dimensional reconstruction with RGB-D SLAM based on ORB (oriented FAST and rotated BRIEF) is proposed. The front end uses the improved RANSAC (Random Sample Consensus) algorithm to improve the feature point matching accuracy, combined with PNP (Perspective-n-point) algorithm to realize the accurate registration of point cloud image, and the back end adopts pose graph to optimize to reduce the influence of noise data on reconstruction, and the error accumulation in the reconstruction process is controlled by using loop detection. Experimental results show that the proposed feature point matching method can significantly improve the matching accuracy of feature points, the correct matching rate is about 94%, which is 6.5% higher than that of traditional RANSAC algorithm, and the proposed reconstruction method is better than the traditional RGB-D SLAM, and the quality of reconstruction results is more excellent. The RMS result of the camera estimated trajectory and the real trajectory is better., and the RMS values are all less than 0.08m.