Visual SLAM Algorithm Based on ORB Features and Line Features

The traditional point-features-based visual SLAM is difficult to track when scene light changes and the camera’s perspective is strenuous. However, in many environments, despite low textures, line-based geometric primitives can still be reliably estimated, such as in urban and indoor scenes dominated by structured edges. And sparse feature point map is also not conducive to human-computer interaction. In this work, we propose a solution to these situations, and we build a visual SLAM system that combines both points features and line features to work robustly in a wider variety of circumstances, specifically in those where point features are exiguous or not well distributed. It uses LSD algorithm to extract and match line features, reconstructs the re-projection error model based on line features, and construct a new error model combines point and line features. The experimental results show that the proposed method can improve the accuracy of the estimated trajectory by 36% compared with the original ORB-SLAM2 algorithm. It can also systematically improve its combination of points and lines in key frames without compromising the efficiency. At the same time, this method can construct a three-dimensional dense map, providing a technical basis for complex application scenarios such as dynamic obstacle avoidance and semantic map.