基于SR300深度相机的褐蘑菇原位测量技术

Portabella Mushrooms Measurement in Situ Based on SR300 Depth Camera

摘要:

褐蘑菇工厂化种植模式下，为了给蘑菇采摘机器人提供工作参数，采用结构光SR300深度相机采集菇床图像送入工控机进行原位测量。针对菇床上褐蘑菇的菌丝干扰背景，在深度图像中利用土壤表面深度的众数，结合蘑菇菌柄至少20mm的高度，自适应选择动态阈值，从菇床背景中提取蘑菇菌盖二值图；针对粘连的类圆形蘑菇，基于2-1圆形Hough变换初步检测其圆心、半径，进一步对蘑菇的边界点进行跟踪、去噪、插补，分割粘连的蘑菇，准确拟合单体蘑菇边界，获取蘑菇圆心和边界点的三维坐标；校准相机坐标系并基于陶瓷圆板验证原位测量方法的精度，由此计算世界坐标系中单体蘑菇的位置、直径、偏向角、倾斜角。现场试验表明，蘑菇直径最大误差为5.57mm，倾斜角最大误差为6.3°，视频帧的运行时间为206ms，单体蘑菇的运行时间为44ms，可满足蘑菇采摘机器人的现场需求。

Abstract:

In factory-like planting pattern of Portabella mushrooms, aiming to provide working parameters for mushroom picking robot, mushroom bed images were collected into the industrial personal computer (IPC) to measure in situ by structured light 3D camera-SR300. For complex background of plenty of hyphae in mushroom bed, the dynamic threshold was calculated adaptively by the mode of depth data of soil surface combined to at least 20mm of stipe height, and the binary image of mushroom caps were extracted from mushroom bed background. For adhesion and quasi-circular mushrooms, the centric coordinates and radius were detected roughly based on the 2-1 cyclo-Hough transform, and then the boundaries of mushrooms were tracked, denoised and interpolated to segment adhesion mushrooms and fit a single mushroom exactly. And the three-dimensional coordinates of the centric and the boundary points of a single mushroom were calculated. By the coordinate system calibration of camera, the measurement accuracy in site was verified by using a ceramic disk. The location, diameter and deviation-lean angle of a single mushroom were calculated in world coordinate system of the camera. Field test results showed that the maximum diameter error of a mushroom was 5.57mm, the maximum lean angle error of a mushroom was 6.3°. The mean run time of a video frame was 206ms, and the mean run time of a single mushroom was 44ms, which can meet the scene requirement of the mushroom picking robot.