计算机视觉技术在番茄收获中的应用

Application of Computer Vision to Tomato Harvesting

摘要:

运用双目立体视觉技术对红色番茄进行定位，图像进行灰度变换后，对图像的二维直方图进行腐蚀，膨胀以后除小团块，提取背景区边缘，然后用拟合曲线实现彩色图像的分割，将番茄从背景中分离出来，对目标进行标定后，用面积匹配实现共轭图像中目标的配准，运用体视成像原理，从两幅二维图像中恢复目标的三维坐标，从实验数据可知，当目标与摄像机的距离为300－400mm时，深度误差可控制在3％－4％。

Abstract:

In this paper, the binocular stereovision is applied to orientating red tomatoes. After transformed to grayscale, the 2D histogram of image was eroded and dilated and little area was removed. The edge of background area was extracted, then the image was segmented along the curve gotten from the edge. The corresponding tomato in right image was matched by means of area match after the tomato was labeled, in which means of thresholding was used. The z coordinate of the feature point can be gotten which supported by stereovision imaging technique. In the course of area match, the corresponding feature point in right image was found by error match, thus matching time and possibility of miss match was shorten and decreased respectively. According to the experiment, the error of depth can be controlled within 3~4 percent when the distance between object and camera was ranged from 300~400 mm.