Optimization of visual SLAM in dynamic environments using object detection

To address the decreased robustness and accuracy of traditional visual SLAM systems in dynamic environments, this paper proposes a visual SLAM method that integrates FastestDet object detection to enhance performance. First, dynamic objects are detected and classified by FastestDet. A depth map-based clustering algorithm is then used to segment the static background from highly dynamic objects. Epipolar geometry constraints are applied to eliminate dynamic points from low-dynamic objects, retaining only static points for pose estimation, thus improving accuracy. Validation on the TUM dataset shows that, compared to ORB-SLAM3, the proposed method reduces the root mean square error of absolute trajectory error by an average of 94.55% in highly dynamic sequences, with notable improvements in low-dynamic sequences as well. These results demonstrate that the proposed method significantly enhances both robustness and accuracy in visual SLAM under dynamic environments.