田间果蔬采摘机器人视觉传感器设计与试验

Design and Experiments of Vision Sensor for Harvest Robots in Field

针对田间果蔬采摘机器视觉识别与定位的需求,设计嵌入式采摘视觉传感器,内嵌130万像素CMOS图像传感器和开放性MCU,配置亮度处理模块和室外光自适应控制专家库,解决室外光变化影响视觉检测的难题。用主从传感器构造双目视觉系统,配置图像分割和测距算法,进行采摘目标的识别与定位。经试验和测试,在采摘臂长范围内,对成熟西红柿的识别有效率为96%,位置误差±12 mm。

Aimed at the demand of recognition and location in computer vision for harvest robots,the embedded vision sensor plat was designed which contained 1.3 million pixel CMOS image chip,MCU with open structure,and brightness-adapting module controlled by the expert database to solve the problem of outdoor light changing which affected the visual inspection.Two sensors were used to construct binocular vision sensor system which was coordinated by master-slave frame.The algorithm of image segmentation and distance measurement was applied to get the location data of harvesting object.The binocular vision sensor system ran smoothly and reliably.The performance test results showed that the success ratio of identifying is 96% and the measure error is within ±12mm when the system took ripe tomato as an measuring object and the measure distance limits was in machine pick-arm range.