Post-integration based point-line feature visual SLAM in low-texture environments.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional SLAM algorithms struggle with robustness and accuracy in environments with weak textures, such as low light or low contrast.
• Feature extraction and loop closure detection are critical components affecting SLAM performance.

Purpose of the Study:

• To improve the robustness and accuracy of SLAM algorithms in challenging, low-texture environments.
• To develop an optimized feature extraction method and a novel loop closure detection technique.

Main Methods:

• Optimized the AGAST-based feature extraction algorithm to adaptively adjust thresholds based on data feature gradient size.
• Proposed a fusion-based incremental loop closure detection method integrating multi-dimensional feature similarity scores using Borda counting.

Main Results:

• Achieved an average Average Precision (AP) of 92.03% for loop closure detection on KITTI datasets (sequences 00, 05, 06).
• Demonstrated excellent accuracy on the EuRoC dataset with a root mean square error ranging from 0.0061 to 0.0281 m.
• The proposed method shows excellent accuracy and robustness in large-scale data processing.

Conclusions:

• The optimized feature extraction and fusion-based loop closure detection significantly enhance SLAM performance in weak texture environments.
• The developed method offers a robust and accurate solution for real-world SLAM applications facing challenging visual conditions.