CAJ | 학술논문

基于双目视觉设计了脐橙采摘机器人果实快速识别与三维视觉定位系统.利用归一化2R-G-B色差模型阈值分割提取目标果实；采用游程编码快速进行连通区域标记及区域形状特征描述；通过最优采摘目标策略选取最佳采摘对象.针对目标的外形、纹理区分度小，容易出现多候选匹配情况，采用形心特征点与邻域灰度互相关性相结合的匹配算法立体匹配.通过定位实验验证了系统的实时性和精度性能.实验数据表明：系统定位单个果实时间为0.6s左右，在220~850mm距离范围内误差小于12mm，可以较好满足采摘机器人实时性和精度要求.
기우쌍목시각설계료제등채적궤기인과실쾌속식별여삼유시각정위계통.이용귀일화2R-G-B색차모형역치분할제취목표과실；채용유정편마쾌속진행련통구역표기급구역형상특정묘술；통과최우채적목표책략선취최가채적대상.침대목표적외형、문리구분도소，용역출현다후선필배정황，채용형심특정점여린역회도호상관성상결합적필배산법입체필배.통과정위실험험증료계통적실시성화정도성능.실험수거표명：계통정위단개과실시간위0.6s좌우，재220~850mm거리범위내오차소우12mm，가이교호만족채적궤기인실시성화정도요구.
The robot with a binocular vision positioning system for picking navel orange is designed according to the requirements of fast identify and the real time accurate target location. The target is extracted by using the method of normalized 2R-G-B threshold segmentation. Labeling connected components and recording characterization of regional shapes are carried out with the run-length encoding. The best picking target is selected with the optimal picking target. Hence the multiple candidates matching is apt to occur due to the appearance and texture of the target fruit without significant difference. The algorithm of cancroids feature combined with regional cross-correlations matching is used to realize the stereo matching. The performances of real-time and precision of the system are verified by the positioning experiment. The experimental results show that the localization time for a fruit is about 0.6 s, and the error is less than 12 mm within the distance ranging from 220 to 850 mm, which can meet the requirements of real-time and accuracy in the picking robot.