Search papers, labs, and topics across Lattice.
This paper introduces OCD SLAM, an advanced stereo visual SLAM framework that enhances ORB-SLAM2 by effectively managing dynamic objects and features within a scene. By employing a novel geometric method based on cross disparity, the system identifies dynamic feature points through the analysis of temporal and stereo inconsistencies, thereby overcoming the limitations of traditional static-world assumptions in SLAM. Evaluations on KITTI datasets reveal that OCD SLAM significantly outperforms ORB-SLAM2 and other contemporary dynamic SLAM approaches in trajectory accuracy, highlighting the importance of its integrated object-level motion classification and filtering techniques.
Achieving a substantial boost in trajectory accuracy, OCD SLAM effectively distinguishes between static and dynamic elements in complex environments.
This paper presents OCD SLAM, a dynamic stereo visual SLAM framework that extends ORB-SLAM2 by jointly addressing dynamic objects and dynamic features in the scene. Usual visual SLAM systems operating in dynamic environments often fail in the presence of moving objects, due to the static-world assumption used in pose estimation and mapping. To address this predicament, we introduce a novel geometric approach based on the discrepancy between disparity and a newly proposed notion called ``cross disparity'', which exploits both temporal and stereo inconsistency to identify dynamic feature points. Complementary to this feature-level motion analysis, OCD SLAM integrates a 3D object detection module (SMOKE) with Kalman filter-based object tracking to perform object-level motion classification, enabling robust separation of static and dynamic scene elements for accurate pose estimation. The proposed approach has been evaluated on various sequences from the KITTI Odometry and KITTI Raw datasets. Results demonstrate that OCD SLAM achieves significant improvement in trajectory accuracy compared to ORB-SLAM2 and several state-of-the-art dynamic SLAM methods. Ablation studies further demonstrate the effectiveness of the cross disparity module in the KITTI Raw dataset and show that this method is able to detect dynamic features that are missed by the 3D object detection scheme alone.
