成熟柑橘的图像识别及空间定位研究

Image Recognition and Three-dimensional Location of Mature Oranges from Nature Scene

采用RGB颜色系统中2R-G-B色差的Ostu自适应阈值分割的方法可以将单果、多果、部分重叠及遮挡的成熟柑橘自动、快速识别;优化圆形Hough变换逼真地拟合出目标图像的形心坐标及半径;提出一种新的基于特征的匹配策略,结合双目立体成像技术实现空间目标的准确定位。经过验证,识别率>95%,当工作距离小于1.0m时,距离误差可以控制在±0.8cm以内,算法准确快捷、鲁棒性好、通用性强。

A new method is presented according to the self-adjusting threshold of 2R-G-B, the algorithm can recognize single or multi-fruits easily and accurately, especially a portion of the fruits overlapped or sheltered; the Circular Hough transform arithmetic can simulate the object image centroid coordinates and radius realistically;The new match technique of the binocular stereo vision can locate the objects accurately, The result shows that the recognition correctness rate is high up to 95%, and when the work distance is less than 1.0m, almost all the errors are less than 0.8cm, the algorithm is effective,robust and universal.