Feature Point Matching Based on Four-point Order Consistency in the RGB-D SLAM System

Random sample consensus (RANSAC) method is often utilized in the RGB-D Simultaneous localization and mapping (SLAM) systems and it is time-consuming because of more repeated fitting the transformation matrix. This paper aims to find a feature point matching method that can reduce computation time in the RGB-D SLAM system. We explore an approach based on four-point order-preserving constraint to determine inliers between two adjacent images. Firstly, the four-point order-preserving constraint between two frames is established to find the good inliers. Then, the 3D points corresponding to the good inliers are obtained to compute the transformation matrix in SLAM system. Finally, the localization and mapping in SLAM system are implemented from transformation matrix and the Global Graph Optimization (g2o) framework. The results indicate that our method is faster and more accurate than the RANSAC algorithm. The less computational time is significant for the real-time SLAM system, and the proposed method is clearly helpful for that.