基于动态目标检测的视觉同步定位与地图构建算法

Visual Simultaneous Localization and Mapping Algorithm Based on Dynamic Target Detection

针对传统视觉同步定位与地图构建(SLAM)算法在动态场景下定位精度低的问题,提出了一种基于动态目标检测的视觉SLAM算法。首先,在视觉SLAM系统的前端对输入图像帧进行预处理,通过目标检测网络YOLO v3(You only look once,v3)剔除图像潜在的动态区域。然后,通过重投影误差优化单应性矩阵,通过求解运动补偿帧得到四帧差分图,并对四帧差分图进行滤波、二值化和形态学处理。最后,结合YOLO v3网络对动态目标检测结果进行优化,以减小图像模糊和强视差产生的噪声。用静态区域的特征点进行视觉SLAM的跟踪、建图与回环检测,在公共TUM数据集上的实验结果表明,本算法能有效提高动态环境下视觉SLAM的精度。

This study proposes a visual simultaneous localization and mapping(SLAM)algorithm based on dynamic target identification to address the problem of low positioning accuracy of conventional visual SLAM algorithm in dynamic scenes.First,the input image frame is preprocessed in front of the visual SLAM system,and the potential dynamic area of the image is deleted by the target detection network you look only once,v3(YOLO,v3).Furthermore,the input image optimizes the homography matrix using reprojection error to obtain the motion compensation frame and four-frame difference image.Then,the four-frame difference image is filtered,binarized,and morphologically processed.Finally,combined with YOLO v3 network to optimize the dynamic target detection results,reduce noise generated by strong parallax and image blur.The feature points of the static area are used for visual SLAM tracking,mapping,and loop detection.The experimental results regarding common TUM data sets indicate that the algorithm can effectively improve the accuracy of visual SLAM in a dynamic environment.