Dynamic SLAM System Using Histogram-based Outlier Score to Improve Anomaly Detection

It is well known that the traditional visual SLAM systems have performed well in a relatively ideal environment, but for the dynamic environment, there are some problems in the instance segmentation algorithm such as incomplete segmentation or wrong segmentation which will result in outliers. Aiming at this problem, this paper proposed a dynamic SLAM system using Histogram-based Outlier Score (HBOS) to improve the reprojection residual constraint to increase the robustness in removing outliers. In this method, HBOS method is used to process the reprojection residual, and the feature association data obtained after the segmentation mask processing is further filtered to improve the robustness. This method can combine the semantic segmentation network with traditional geometric methods to solve the data association problem caused by the wrong information generated from the dynamic objects. To verify the effectiveness of the proposed method, the experiment is carried out on the public TUM data set and the results show that the proposed method improves the accuracy of SLAM positioning in the dynamic environment.