An efficient edge-feature constraint visual SLAM

摘要

This paper proposes a constraint method based on the relationship between feature points and object regions. Feature-based visual simultaneous localization and mapping (SLAM) is an effective localization approach for robots in unknown environments. Features are some suitable points in the image that can be used as landmarks and the points are represented by descriptors which are computed using local image information. However, for the feature at the edge of an object, under viewpoint changes, its background appearance may change significantly, which makes its descriptor change. Therefore, it is difficult to achieve the correct feature. In this paper, the proposed method uses a deep convolution neural network to get object location information that is helpful to distinguish the features belong to the edge regions. Then, regarding the issue above, we add additional constraints to these features. In other words, the features in the edge regions are usually abandoned for feature matching. In order to evaluate our proposed method, we conduct experiments on ORB-SLAM2 constrained by YOLOv3. The comparison results based on public datasets show the proposed method could effectively reduce the absolute trajectory error.