A Visual SLAM Image Mismatching Filter Algorithm Based on Progressive Aample Consensus

Visual SLAM based on ORB features will increase the computational pressure of SLAM system due to the large amount of feature extraction and matching computation and the need to screen a large number of mismatched point pairs. It cannot completely eliminate the mismatched point pairs, which will affect the camera positioning accuracy of visual SLAM system to a certain extent. To solve the two questions, the PROSAC algorithm is used, screening point of mismatch on, all matching points by calculation of evaluation function, selection to match point to build the model with the highest quality, through continuous to join interior point, build the final model, screening point pairs of mismatch. Provide high quality data for camera pose estimation and back end optimization of SLAM system. Through the comparison of RANSAC algorithm and PROSAC algorithm false match screening time, as well as tracking error. PROSAC algorithm effectively reduced the time of mismatching screening, with a maximum improvement of 100 times. The tracking error has also improved significantly.