VA-LOAM: Visual Assist LiDAR Odometry and Mapping for Accurate Autonomous Navigation

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Odometry performance is crucial for autonomous systems.
• LiDAR and vision sensors have complementary strengths and weaknesses.
• Traditional sensor fusion methods are vulnerable to environmental conditions.

Purpose of the Study:

• To enhance odometry performance by fusing vision and LiDAR sensors.
• To develop a robust system that overcomes the limitations of individual sensors.
• To improve LiDAR-based mapping accuracy and reliability.

Main Methods:

• Integrated vision sensors with LiDAR sensors, prioritizing LiDAR data.
• Developed a vision support module to enhance LiDAR feature matching.
• Implemented and evaluated the approach on top LiDAR SLAM algorithms using the KITTI dataset.

Main Results:

• Significantly improved LiDAR odometry performance, especially in adverse conditions.
• Demonstrated enhanced LiDAR feature matching by leveraging vision data.
• Achieved robust performance where vision-only or traditional fusion methods fail.

Conclusions:

• The proposed sensor fusion strategy enhances odometry reliability and accuracy.
• LiDAR as the primary sensor with vision support offers superior performance.
• Publicly releasing the VA-LOAM source code promotes research transparency and reproducibility.