摘要:
为将目标油茶果实从树枝、树叶等外界遮挡中分离出来，以利于油茶采摘机器视觉的图像形态学识别，该文提出了一种改进的类圆随机Hough变换算法，在算法中添加了边缘预检测、快速定位圆心点等模块以提高算法的识别率。仿真结果表明，改进算法对遮挡果实的识别率较其他Hough遮挡识别算法有所提高，最高达到90.70%，识别时间为1.3 s。该研究为采摘机器人的后续采摘工作打下了基础。
Abstract:
Abstract: Camellia tree is widely distributed in the south areas of China. When the camellia fruits mature, their colors appear to pink or yellow. These features and interference factors outside will bring troubles for identification and picking of camellia fruits. In order to separate the camellia-fruit from background impurities (branches, leafs, etc), the paper proposed a quasi-circular Randomized Hough Transform (RHT) algorithm. In the modified algorithm, it adds the some model to improve the recognition rates. The detail steps are list as follows:In order to reduce computation of quasi-circular RHT algorithm，the classical Sobel operator was used to extract the edge of target binary segmentation image. The Sobel operator is a commonly used edge detection method，the algorithm has the less computation time and faster detection speed，it can reflect the disturbing of target edge nicely，so the algorithm is suitable for detecting camellia fruit edge.Quasi-circular RHT algorithm is based on the circular-RHT and circular-detection algorithm research. For the feature of camellia fruit，the algorithm added the model of early-detection, the circular-centre location and the overlapping target merging. Early-detection is considered for the advantage of the quasi-circular geometry feature，the modified algorithm selects the idea of curve slope to approximately reflect the change of the curve-edge. Circular-centre location is the method of refine relative picking points to select circular points compared to the classical RHT. This modification can avoid invalid picking caused by the points compact, first connecting 2 neighbor points as the string of the circular, then selecting the intersection point of two string perpendicular bisectors as the candidate circular centre. Owing to the character of the plant, the camellia fruits central disturbing in the natural environment, so it can not avoid overlapping caused by the camera distance. So the model will merging the heavily overlapped camellia fruits to make picking easy.The simulation proved that the image recognition rate is greatly influenced with the illumination for the shielding camellia fruits. In the weak illumination environment, because the camellia fruits and the external target are not clear, it causes low recognition rate. With increasing illumination, the targets in the image became clear, the recognition rate reached up to 90.70% under 10 000 lx illumination. With the illumination increasing again，the images became more and more complicated, and it caused gradually decrease of the recognition rates, the lowest recognition rates was 52.73% under 22 000 lx. The quasi-circular Randomized Hough Transform algorithm is more practical in computer vision systems of camellia fruit picking robot.
下载:
