Visual SLAM Dynamic Disturbance Suppression Algorithm Combining Blur Denoising and Object Detection
Jiangting Zhao, Xiaoyu Zhang, Yandong Yang · IEEE Internet of Things Journal · 2025
Visual simultaneous localization and mapping (SLAM) is a crucial technology for solving the navigation problem of unmanned systems in unknown environments, playing an important role in Internet of Things (IoT) applications. However, the dynamic objects in the environment bring great challenges to visual SLAM. The dynamic objects and the blur errors caused by the motion seriously affect the feature extraction quality and localization accuracy of the system. To address these problems, this paper proposes a dynamic disturbance suppression algorithm combining blur denoising and object detection, referred to as DOSLAM, which improves the front end of ORB-SLAM2 to make the system more robust in dynamic environments. First, DO-SLAM employs a deblurring algorithm based on generative adversarial networks (GANs) to reduce the blur noise caused by dynamic objects or the jitter of unmanned platforms. Then, the optimized object detection is combined with the geometry constraint and the overlapping box strategy to comprehensively eliminate the dynamic feature points. Finally, the remaining static feature points are used to complete the subsequent pose estimation. Local and global experiments in TUM dynamic datasets demonstrate that DO-SLAM greatly improves the performance of the system in dynamic environments compared with ORB-SLAM2 and has better robustness than other similar methods.