动态场景下融合YOLOv5s的视觉SLAM算法研究OA

Visual Simultaneous Localization and Mapping(SLAM)systems are susceptible to interferences from dynamic objects in dynamic scenes,leading to decreased localization accuracy and robustness.To address this problem,a visual SLAM algorithm incorporating the lightweight YOLOv5s object detection network is proposed.This algorithm introduces a module for target detection and removal of dynamic feature points into the tracking thread of ORB-SLAM2,aiming to improve the localization accuracy and robustness of the SLAM system by eliminating dynamic feature points from the images.Firstly,an enhanced lightweight object detection algorithm based on YOLOv5s is developed to improve the inference speed and detection accuracy of the network on mobile devices.Secondly,the lightweight object detection algorithm is combined with the ORB feature point algorithm to extract semantic information from the images and remove the pre-determined dynamic features.Finally,dynamic feature points are eliminated using the Lucas-Kanade optical flow method and epipolar geometry constraints,and the remaining feature points are utilized for pose estimation.Validation on the TUM dataset demonstrates that the proposed algorithm outperforms the original ORB-SLAM2,achieving over 95％improvement in both Absolute Trajectory Error(ATE)and Relative Pose Error(RPE)metrics for high dynamic sequences,thus effectively enhancing the localization accuracy and robustness of the system.Moreover,compared to existing state-of-the-art SLAM algorithms,the proposed algorithm exhibits significant improvements in accuracy and real-time performance,making it more valuable for applications on mobile devices.