Three-Dimensional Environment Mapping with a Rotary-Driven Lidar in Real Time

Area of Science:

• Robotics and Computer Vision
• 3D Environment Reconstruction
• Simultaneous Localization and Mapping (SLAM)

Background:

• Traditional LiDAR systems face field-of-view limitations, leading to blind spots and incomplete 3D reconstructions.
• Existing SLAM methods struggle with accuracy and robustness in complex, GNSS-denied environments.

Purpose of the Study:

• To develop a novel 3D environment reconstruction approach overcoming LiDAR field-of-view limitations.
• To enhance the accuracy and robustness of 3D reconstruction using a multi-sensor fusion framework.
• To provide a high-fidelity 3D reconstruction solution for challenging environments.

Main Methods:

• Designed a rotary-driven LiDAR mechanism for uniform, full-field-of-view scanning.
• Developed the LiDAR-Visual SLAM (LV-SLAM) framework with multi-threaded feature registration.
• Implemented a two-phase loop closure detection mechanism for improved accuracy and robustness.

Main Results:

• LV-SLAM reduced average absolute trajectory error (ATE) from 6.90 m to 2.48 m on the KITTI benchmark.
• Achieved lower relative pose error (RPE), indicating enhanced global consistency and reduced drift.
• Real-world tests showed more complete reconstructions with fewer occlusions and geometric accuracy within 5 cm RMSE.

Conclusions:

• The proposed rotary LiDAR and LV-SLAM system offers a robust and accurate solution for high-fidelity 3D reconstruction.
• The system effectively overcomes field-of-view limitations and blind spots inherent in traditional LiDAR setups.
• Demonstrated superior performance compared to state-of-the-art methods in both benchmark and real-world evaluations.