基于RGB-D相机的矮砧苹果识别与定位

Recognition and localization of dwarf apple based on RGB-D camera

为实现果园的智能化管理,提出一种基于RGB-D相机的矮砧苹果的识别与定位方法。通过KinectV2相机获取果树的彩色图像与深度图像,使用改进的U-Net网络对彩色图像进行语义分割,将分割后的图像与深度图像融合进行滤波处理,获取苹果区域的点云;利用LCCP(locally convex connected patches)算法对苹果点云的局部凹凸性进行分析,对点云进行实例化分割;使用多线程的渐进采样一致性算法对每个实例化点云进行形状拟合,获取果实的半径以及球心位置。实验结果表明,该方法能够高效完成苹果果实的识别并获取果实空间位置与形状信息。

To achieve intelligent management of orchards,an efficient method about apple detection and localization was proposed.KinectV2 camera was used to obtain the color image and depth image.An improved U-Net network was used for semantic segmentation of color image.The semantic image and depth image were fused and the point cloud of fruit area was obtained though point cloud filtering.LCCP(locally convex connected patches)algorithm was used to analysis the local convexity and further acquire the instance segmentation of apples point cloud.Multi-thread PROSAC(progressive sample consensus)algorithm was used to fit the fragmentary point cloud to get the radius and space coordinates of apples.Experimental results show that the proposed method can efficiently complete the apple recognition and obtain the spatial position and shape information of the fruit.