基于深度图像的球形果实识别定位算法

Spherical Fruit Recognition and Location Algorithm Based on Depth Image

为了解决近色背景果实识别困难问题,针对果实近球形的形态特性,提出了一种利用深度图像从果实形态角度进行果实识别定位的算法。该算法使用深度摄像头获取果树的深度图像,通过深度图像计算出各像素点的梯度向量,将梯度向量看作运动矢量场,并计算出矢量场的散度,根据散度最大原则,从矢量场中搜索出辐散中心点;然后利用果实和叶片等深图像的差异从辐散中心点中筛选出果实中心点,以果实中心点为起点采用八方向搜索方法搜索出果实边界点,将果实边界点依次连接后形成的封闭区域内的果实图像导入点云;最后根据果实图像部分点云利用RANSAC算法求出目标果实的拟合球形,进而得出果实的尺寸以及三维空间位置。该算法无需传统算法需要利用的颜色特征,而仅利用了深度图像中的深度信息进行果实识别定位,能够克服传统算法受色彩、光照等因素影响的弊端,并且由于该算法完全没有利用到彩色图像信息,因此不仅可以实现绿色果实的识别定位,还可以实现采摘机器人在夜间环境下正常工作,为复杂环境下的果实识别定位算法研究提供了技术支撑。

In order to solve the difficulty of fruit recognition in near color background,an algorithm for identifying and locating fruits from the depth image was presented based on the near-spherical morphological characteristics of fruits.A depth camera was used to get depth image of a fruit tree.The gradient vectors of each pixel point from the depth image were calculated.The gradient vector was considered as a vector field of motion and the divergence of the vector field was calculated.Searching for divergence center points from vector fields according to the principle of maximum divergence.The fruit center point was selected from the divergence center point by using the difference of the contour image between the fruit and the leaf.The fruit boundary points were searched in eight directions from the center point of the fruit.The fruit images in the closed area formed by connecting the fruit boundary points were imported into the point cloud.Finally,the point cloud was used to find the fitting circle of the target fruit according to the random sample consensus(RANSAC)algorithm,and the size and spatial location of the fruit were obtained.The algorithm discarded the color features commonly used in traditional algorithms but used only the depth information in the depth image for fruit recognition and positioning.It can overcome the drawbacks of traditional algorithms affected by color,illumination and other factors.Because the algorithm did not use color image information at all,it can not only recognize and locate green fruits,but also enable the harvesting robot to work in dark environment.The research result can provide a method for fruits recognition and location in complex environment.