A Real-Time Robot Location Algorithm Based on Improved Point-Line Feature Fusion

The traditional robot real-time positioning and mapping (SLAM) algorithm has the problem of over-extraction of line segments in dense environments, which leads to mis-matching and reduces the accuracy of the system. To solve this problem, this paper proposes a visual SLAM method called IPLI-SLAM based on an improved point-line combination tightly coupled with IMU. First of all, Shi-Tomasi feature is used for point feature extraction. Secondly, in the aspect of online extraction, the LSD line detection method is modified to the accuracy of the algorithm. The system sets a filter before line extraction. The filter is based on the pixel gradient density setting to filter the environment with more line textures to reduce the error matching of the algorithm. Finally, the point-line visual information is tightly coupled with IMU and then added to the back-end for optimization. By testing eight different sequences in the open source dataset EuRoc with three difficulty levels: easy, medium and hard, the results show that the trajectory RMSE value (root mean square error) of this algorithm is reduced by 50.7% compared to the VINS-mono algorithm and by 13.2% compared to the PL-vins algorithm, which fully demonstrates the effectiveness of this algorithm.