CAJ | 학술논문

为了避免或减少自然光线的干扰,该文设计了一种用于苹果采摘机器人的激光视觉系统.基于飞行时间原理的LMS211激光测距仪可对目标距离进行测量,具有精度高、响应速度快等优点;研制的直线运动单元可自由调整其滑台的移动速度和行程,用来协助测距仪完成对目标场景的三维扫描.试验结果表明:在一定测量范围内,扫描数据能较理想地反映果实的曲面特性,选择合适的水平分辨率可提高数据的成像效果,生成的距离图像易于解析果实、枝叶的空间几何特性及相互间的层次关系,且效果不受光线变化的影响.该系统为后期果实的识别研究提供参考.
위료피면혹감소자연광선적간우,해문설계료일충용우평과채적궤기인적격광시각계통.기우비행시간원리적LMS211격광측거의가대목표거리진행측량,구유정도고、향응속도쾌등우점;연제적직선운동단원가자유조정기활태적이동속도화행정,용래협조측거의완성대목표장경적삼유소묘.시험결과표명:재일정측량범위내,소묘수거능교이상지반영과실적곡면특성,선택합괄적수평분변솔가제고수거적성상효과,생성적거리도상역우해석과실、지협적공간궤하특성급상호간적층차관계,차효과불수광선변화적영향.해계통위후기과실적식별연구제공삼고.
Machine vision was one of the largest external environmental information sources, which was not only related to the capability of recognizing fruit fast and accurately, but also determined the reliability of harvesting robot directly. Designing and developing vision system customized and suitable to harvesting objects was of great significance to realize automatic fruit harvesting. CCD camera was the key component for existing vision systems, but it had a certain limitation for fruit recognition rate and position accuracy, because it was sensitive to unstructured harvesting environment, that was varying, unknown and open, especially to the problems such as the uncertainty of illumination conditions. Therefore, a laser vision system for apple harvesting robot was presented to avoid or reduce the effect of natural light. Based on the principle of time-of-flight, LMS211 was used for data acquisition with higher measurement precision and faster response speed. A linear motion unit was designed for assisting laser range finder to complete three dimensional scanning of object scene, which could adjust movement speed and travel of sliding table freely. A software based on Visual C++6.0 was developed for date collection and management, MATLAB would be used for range images generation, image preprocessing, fruit recognition and position in the later period. The different experimental results showed that scanning data could reflect the characteristics of fruit surface ideally within a certain measurement range (including valid scanning distance range from 200 mm to 1 400 mm, and ideal scanning angle range from 80°to 120°) , and the suitable horizontal resolution was computed by the best estimator, which would increased imaging accuracy. Range image generated was easy to analyze geometrical features of fruits, leaves and branches, and hierarchical relationship between each other. At the same time, the kind of image was immune to various lighting condition, which background could be simplified by range constraint conveniently. All these advantages could provide even richer pattern information for fruit recognition.