Mobile Robot Localization and Mapping Algorithm Based on the Fusion of Image and Laser Point Cloud

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Visual SLAM algorithms struggle with scale ambiguity due to a lack of depth information in image features.
• This scale blur leads to degraded performance and tracking failures in visual SLAM systems.
• Integrating additional sensor data is crucial for robust ego-motion estimation.

Purpose of the Study:

• To address the scale blur problem in visual SLAM by incorporating lidar point cloud data.
• To improve the stability and accuracy of ego-motion estimation in visual SLAM.
• To develop a more robust and adaptable SLAM algorithm for real-world applications.

Main Methods:

• Introduced lidar point clouds to provide depth information for visual features.
• Improved the front-end of visual SLAM using nonlinear optimization and introducing pole error.
• Implemented a keyframe-based local pose optimization to manage computational complexity.

Main Results:

• The proposed LV-SLAM algorithm demonstrated superior performance on the KITTI dataset and in outdoor environments.
• Achieved a 52.7% reduction in trajectory error compared to pure visual SLAM algorithms.
• The algorithm showed good adaptability and robust stability across diverse environmental conditions.

Conclusions:

• Integrating lidar data effectively resolves scale ambiguity in visual SLAM.
• The enhanced nonlinear optimization front-end improves pose estimation stability and accuracy.
• LV-SLAM offers a robust and adaptable solution for mobile robot localization and mapping.