A Visual SLAM With Tightly Coupled Integration of Multiobject Tracking for Production Workshop

• Published in: IEEE internet of things journal Vol. 11; no. 11; pp. 19949 - 19962
• Main Authors: Gou, Rongsong, Chen, Guangzhu, Pu, Xin, Liao, Xiaojuan, Chen, Runji
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.06.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Cameras; Conferences; Datasets; Dynamic environment; Dynamics; Extended Kalman filter; Feature extraction; Heuristic algorithms; multiobject tracking; Multiple target tracking; Object motion; Performance evaluation; production workshop; Simultaneous localization and mapping; simultaneous localization and mapping (SLAM); Visualization; Workshops
• ISSN: 2327-4662; 2327-4662
• DOI: 10.1109/JIOT.2024.3368417

The application of simultaneous localization and mapping (SLAM) technology has a noteworthy potential for enhancing the cognitive capability of production workshops, particularly for complex and ever-changing industrial settings. In the production workshop, obtaining the information of dynamic objects in a scene is key of accurate Visual SLAM. This article proposes a tightly coupled visual SLAM (TC_vSLAM) method, which can accurately obtain SE(3) pose information on cameras, as well as motion information on moving objects in a scene. The proposed method first uses a simple extended Kalman filter-based tracker to determine the actual motion state of objects and associate it. Next, the extracted features are classified into static background features and object features based on the object instance mask information and object state information. Static features are used to initialize camera poses, whereas dynamic features are used to obtain SE(3) pose information on tracked objects in the considered scene. Finally, a new graph optimization method is proposed to optimize the static 3-D landmarks, object 3-D landmarks, camera poses, and poses of moving objects in a scene jointly. The TC_vSLAM is verified on the KITTI dataset and the OMD dataset compared with the ORB_SLAM2, the ORB_SLAM3, and the VDO_SLAM, then the performance evaluation of the TC_vSLAM in real environment is also conducted. The experimental results validate the effectiveness of the TC_vSLAM.