Outdoor high-precision 3D dense mapping system based on stereo visual SLAM

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional visual Simultaneous Localization and Mapping (SLAM) struggles with outdoor autonomous navigation due to sparse mapping.
• Existing methods lack the detail required for precise obstacle avoidance and environmental understanding.

Purpose of the Study:

• To develop an advanced visual SLAM algorithm for high-precision 3D mapping in outdoor environments.
• To improve the robustness and accuracy of dense depth estimation and map reconstruction.

Main Methods:

• Utilizes a deep stereo matching network with cross-attention for enhanced disparity estimation.
• Incorporates an adaptive disparity refinement strategy to mitigate mismatches in complex scenes.
• Employs a separate dense mapping thread for fusing depth data with SLAM poses for loop closure and correction.

Main Results:

• Achieved a relative depth estimation error of 8.236% on field test data.
• Reconstructed 3D maps with 92.902% of dense points having an error within 0.443 meters.
• Demonstrated superior accuracy and robustness compared to traditional visual SLAM techniques.

Conclusions:

• The proposed algorithm effectively generates high-quality 3D dense maps essential for autonomous systems.
• Addresses limitations of sparse point clouds and insufficient environmental information in conventional SLAM.
• Offers a robust solution for accurate 3D mapping in challenging outdoor conditions.