Hardware Implementation of Improved Oriented FAST and Rotated BRIEF-Simultaneous Localization and Mapping Version 2.

Area of Science:

• Robotics and Computer Vision
• Autonomous Systems Engineering

Background:

• Achieving high-level automation in autonomous driving requires accurate mapping, especially in unknown environments.
• Simultaneous Localization and Mapping (SLAM) algorithms are crucial for navigation but often require significant computational power.
• Existing SLAM methods struggle with efficiency in resource-constrained platforms.

Purpose of the Study:

• To enhance the ORB-SLAM2 algorithm for improved mapping performance on a Raspberry Pi 3.
• To address the challenge of balancing accuracy and computational efficiency in SLAM for embedded systems.
• To enable real-time autonomous navigation in environments with limited computing resources.

Main Methods:

• Implementation of an enhanced ORB-SLAM2 algorithm on a Raspberry Pi 3 (ARM A53 CPU).
• Optimization of feature descriptors for increased stereo feature-matching rates.
• Tuning of loop-closing parameters to minimize accumulated errors in mapping.
• Utilizing the Robot Operating System (ROS) for system integration and performance monitoring.

Main Results:

• Significant error reduction in monocular SLAM: RMSE decreased by 18.11%, mean error by 22.97%, median error by 29.41%, and maximum error by 17.18%.
• Minor error reduction in stereo SLAM: RMSE decreased by 0.30% and mean error by 0.38%.
• Stable ROS topic frequency at 10 Hz with approximately 90% quad-core CPU utilization, indicating real-time capability.

Conclusions:

• The enhanced ORB-SLAM2 system provides a viable solution for accurate and efficient real-time SLAM on resource-constrained platforms.
• The optimizations successfully improve mapping accuracy while maintaining computational feasibility for embedded autonomous systems.
• This work demonstrates a practical approach to deploying advanced SLAM capabilities in edge computing scenarios for autonomous driving.