基于目标检测和动静点分离的视觉即时定位与地图构建技术  

Visual SLAM Technology Based on Target Detection and Dynamic&Static Point Separation

作　　者：来金强 Lai Jinqiang

机构地区：[1]宁波大学机械工程与力学学院,浙江宁波315211

出　　处：《机械制造》2023年第11期80-84,共5页Machinery

摘　　要：针对视觉即时定位与地图构建在动态环境中鲁棒性失效的问题,提出一种基于目标检测和动静点分离的视觉即时定位与地图构建技术。使用目标检测算法YOLO v5对图像预处理,获取动态目标物的图像区域。在图像区域划分的基础上,对框外区域使用随机抽样一致性算法剔除外点。根据点的运动规律分离框内区域的动点和静点,在光束法平差优化中对可用静点赋以权重。为了在建图中滤除动态信息,仅保留框外区域的点云显示。通过性能测试确认,所提出的方法在数据集中表现出较高的精度和鲁棒性,整体上优于同类算法。Aiming at the problem of robustness failure of visual simultaneous localization and mapping in dynamic environment,a visual simultaneous localization and mapping technology based on target detection and dynamic&static point separation was proposed.The image preprocessing was performed using the object detection algorithm YOLO v5 to obtain the image area of the dynamic target.On the basis of image area division,the random sample consensus algorithm was used to remove the exclusion point for the area outside the frame.According to the motion law of the point,the moving point and static point in the area in the frame were separated,and the available static point was weighted in the bundle adjustment optimization.In order to filter out dynamic information in mapping,only the point cloud display of the area outside the frame was retained.Through performance test,it is confirmed that the proposed method shows high accuracy and robustness in the dataset,and is generally better than similar algorithms.

关 键 词：目标检测 动静点分离 即时定位与地图构建 

分 类 号：TP274.5[自动化与计算机技术—检测技术与自动化装置]