动态场景下融合YOLOv5s的视觉SLAM算法研究

Research on Visual SLAM Algorithm Incorporating YOLOv5s in Dynamic Scenes

为了解决视觉同步定位与建图(Simultaneous Localization and Mapping, SLAM)系统在动态场景下容易受到动态物体干扰,导致算法定位精度和鲁棒性下降的问题,提出了一种融合YOLOv5s轻量级目标检测网络的视觉SLAM算法。在ORB-SLAM2的跟踪线程中添加了目标检测和剔除动态特征点模块,通过剔除图像中的动态特征点,提高SLAM系统的定位精度和鲁棒性。改进了YOLOv5s的轻量化目标检测算法,提高了网络在移动设备中的推理速度和检测精度。将轻量化目标检测算法与ORB特征点算法结合,以提取图像中的语义信息并剔除先验的动态特征。结合LK光流法和对极几何约束来剔除动态特征点,并利用剩余的特征点进行位姿匹配。在TUM数据集上的验证表明,提出的算法与原ORB-SLAM2相比,在高动态序列下的绝对轨迹误差(Absolute Trajectory Error, ATE)和相对轨迹误差(Relative Pose Error, RPE)均提高了95%以上,有效提升了系统的定位精度和鲁棒性。相对当前一些优秀的SLAM算法,在精度上也有明显的提升,并且具有更高的实时性,在移动设备中拥有更好的应用价值。

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.