Title: SLAM algorithm based on bounding box and deep continuity in dynamic scene

Authors: Baofu Fang; Xiumeng Han; Zaijun Wang; Xiaohui Yuan

Addresses: School of Computer Science and Information Engineering, Intelligent Interconnected Systems Laboratory of Anhui Province, Hefei University of Technology, Hefei, Anhui 230009, China ' School of Computer Science and Information Engineering, Intelligent Interconnected Systems Laboratory of Anhui Province, Hefei University of Technology, Hefei, Anhui 230009, China ' Key Laboratory of Flight Techniques and Flight Safety, Civil Aviation Flight University of China, Guanghan, Sichuan 618307, China ' Department of Computer Science and Engineering, University of North Texas, Denton, Texas 311277, USA

Abstract: To solve the problems of low positioning accuracy and poor robustness caused by the inability of Simultaneous Localisation and Mapping (SLAM) algorithm to deal with dynamic targets in dynamic scenes, a SLAM algorithm based on bounding box and depth continuity in dynamic scenes is proposed. Firstly, the pixel random search filling process is carried out in combination with the depth and the bounding box to obtain the pixel-level segmentation result of the prior dynamic target. Then, the dynamic features are screened to eliminate the influence of dynamic targets. According to the screening results, relative static features which can be used for posture optimisation are selected. Posture optimisation is carried out by combining the static features to obtain the optimised camera posture. All experimental results show that the proposed algorithm significantly improves the positioning accuracy and real-time performance of SLAM algorithm in complex dynamic scenes.

Keywords: SLAM; simultaneous localisation and mapping; visual sensor; RGB-D perception; YOLO; object detection; dynamic scene; semantic information; depth continuity; data association; mapping.

DOI: 10.1504/IJWMC.2021.121626

International Journal of Wireless and Mobile Computing, 2021 Vol.21 No.4, pp.349 - 364

Received: 30 Sep 2021
Accepted: 17 Jan 2022
Published online: 21 Mar 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article