基于机器视觉的无序堆放西瓜识别与定位技术

Research on recognition and localization of unordered stacked watermelons based on machine vision

【目的】研究一种基于机器视觉的西瓜识别与定位技术,满足西瓜搬运设备的自动化要求,为自动搬运设备的视觉部分提供技术支持。【方法】采集西瓜的二维图像数据与三维点云数据。调用OpenCV函数库分割西瓜二维图像,提取出西瓜的外轮廓;采用PCL函数对三维点云数据进行预处理,并与二维图像作图形特征匹配,提取出顶层西瓜的质心点。【结果】单轮次的西瓜轮廓总识别率为97.62%,正确识别率为95.58%,质心识别率为92.72%。20轮测试的总轮廓识别率为98.02%,总正确识别率为96.53%,总质心识别率为94.17%,总质心个数和总西瓜个数之间的误差率为2.36%;西瓜图像处理的总用时为101.8s,效率提高了77.8%。【结论】基于机器视觉的西瓜识别与定位技术识别率较高且误差率较低。

【Objective】To study a watermelon recognition and positioning technology based on machine vision to meet the automation requirements of watermelon handling equipment.【Methods】2D image data and 3D point cloud data of watermelon were collected and the OpenCV function library was called to segment the two-dimensional image of watermelon and extract the outer contour of the watermelon;The 3D point cloud data was preprocessed using the PCL function and perform graphic feature matching with the 2D image to extract the centroid points of the top layer watermelon.【Results】The validation test results showed that the total recognition rate of the number of watermelons in a single round was 97.62%,the correct recognition rate was 95.58%,and the centroid recognition rate was 92.72%.The total recognition rate of 20 rounds of testing was 98.02%,the total correct recognition rate was 96.53%,the total centroid recognition rate was 94.17%,and the error rate between the total centroid number and the total number of watermelons was 2.36%;The total time for watermelon image processing was 101.8 seconds,which improved the efficiency by 77.8%.【Conclusion】The watermelon recognition and positioning method proposed in this article has high recognition rate and low error rate,which can provide technical support and algorithm reference for the visual part of automatic handling equipment.