基于点云投影的货物盘点机器人目标识别算法

Inventory System for Goods Based on Laser Radar Point Cloud Projection

针对物流仓储中常见的非标仓库货物盘点问题,提出一种基于激光雷达点云投影的货物识别算法。首先对采集到的货物点云数据采用RANSAC(随机采样一致性)和ICP(迭代最近点)算法进行点云数据拼接,同时基于改进欧式聚类方法实现货物错层处理;其次将货物点云投影至二维平面并采用点云z轴法向量作为特征信息,形成特征灰度图像;最后借助改进分水岭算法对特征灰度图进行图像分割。通过实物平台及现场数据测试,该算法货物识别精度达到90%以上,相较基于深度学习YOLOv5算法的图像识别方法具有稳定的识别率,且有效避免依赖大规模公共数据集的问题。

Addressing the common non-standard warehouse inventory issues in logistics and storage,a cargo recognition algorithm based on the projection of LiDAR point cloud is proposed.Initially,the collected cargo point cloud data is spliced using the RANSAC(random sample consensus)and ICP(iterative closest point)algorithms,while an improved Euclidean clustering method is implemented for handling cargo mislayers.Subsequently,the cargo point cloud is projected onto a two-dimensional plane,and the z-axis normal vector of the point cloud is used as feature information to form a feature grayscale image.Finally,an improved watershed algorithm is utilized for image segmentation of the feature grayscale image.Through testing with a physical platform and on-site data,the algorithm achieves a cargo recognition accuracy of over 90%,offering a stable recognition rate compared to the image recognition method based on the deep learning YOLOv5 algorithm,and effectively avoiding the issue of reliance on large-scale public datasets.