YPR-SLAM: A SLAM System Combining Object Detection and Geometric Constraints for Dynamic Scenes

Area of Science:

• Robotics
• Computer Vision

Background:

• Traditional Simultaneous Localization and Mapping (SLAM) systems struggle with dynamic environments due to moving objects.
• Moving objects negatively impact image matching and camera pose estimation accuracy in SLAM.

Purpose of the Study:

• To propose the YPR-SLAM system, an efficient solution for robust SLAM in dynamic environments.
• To minimize the adverse effects of moving objects on SLAM performance.

Main Methods:

• Integration of a lightweight YOLOv5 object detection network for identifying dynamic and static objects.
• Application of a geometric constraint method using depth images and prior dynamic object information.
• Utilizing the Depth-PROSAC algorithm to differentiate and remove dynamic feature points.

Main Results:

• Successful construction of dense cloud maps using only static feature points.
• Demonstrated effectiveness of YPR-SLAM in dynamic environments on the TUM RGB-D dataset.
• Significant reduction in motion feature point interference.

Conclusions:

• YPR-SLAM offers an efficient and tightly coupled approach combining object detection and geometric constraints.
• The system effectively eliminates motion feature points, enhancing SLAM robustness in real-world dynamic scenes.