基于点云的果园环境苹果定位技术研究

Research on Apple Positioning Technology in Orchard Environment Based on Point Cloud

摘要: 复杂枝叶遮挡的果园环境中的苹果准确检测和定位对于苹果采摘的成功率至关重要。本研究提出了一种目标检测网络和点云处理的方法，用于检测遮挡和未被遮挡的苹果以及计算果实中心及主轴。使用深度相机获取RGB彩色图和深度图像。使用YOLOv5s目标检测网络检测苹果果实，并获取目标点云，通过点云特征以及主成分分析法计算苹果的中心坐标位置；利用最优化方法计算苹果的果轴方向。最后为验证所提出方法的有效性，采用VREP仿真平台部署苹果、果树枝叶遮挡和相机，验证该算法的有效性和稳定性，并在实地果园场景中部署算法和果实采摘机器人，验证了算法在实际场景中的有效性和鲁棒性。

Abstract: Accurate detection and location of apples in orchard environment with complex branches and leaves is very important for the success rate of apple picking. In this study, a target detection net-work and point cloud processing method is proposed, which are used to detect occluded and unob-structed apples and calculate the fruit center and main axis, using depth camera to obtain RGB col-or map and depth image. The YOLOv5s target detection network is used to detect the apple fruit, and the target point cloud is obtained, and the central coordinate position of the apple is calculated by point cloud characteristics and principal component analysis, using the optimization method to calculate the fruit axis direction of apple. Finally, in order to verify the effectiveness of the proposed method, the VREP simulation platform is used to deploy apple and fruit tree branches and cameras to verify the effectiveness and stability of the algorithm, and the algorithm and fruit-picking robot are deployed in the field orchard scene to verify the effectiveness and robustness of the algorithm in the actual scene.