Matching Risk for Feature Selection in Visual SLAM

Abstract Estimating the position of a feature point is very important for the performance of visual simultaneous localization and mapping (SLAM). However, various factors, such as camera motion and matching result, influence the estimation accuracy of feature point position. This paper presents a feature-selection method based on matching risk estimation in visual SLAM. The matching risk is estimated for each feature point in the image. In addition, only the feature points with low matching risk are used in triangulation. The performance of the triangulation and visual SLAM applying the proposed method is demonstrated using simulation and experiments, respectively.