边缘曲率法用于重叠苹果识别的研究

Edge Curvature Method for Overlapping Apple Recognition

苹果采摘机器人视觉系统的关键问题是果实目标的准确识别和精准定位。在非结构化果园环境下,存在很多影响成熟果实识别正确性的因素,果实的相互重叠就是其中最主要的因素。为实现苹果目标的识别及空间定位,本文提出了一种基于边缘曲率分析的重叠苹果识别方法。该方法首先从二值图像中提取苹果区域的边缘,计算边缘上每个点的曲率,除去曲率异常的边缘点。同时将整个边界划分为若干段,并根据三个准则进行筛选,保留有效段。最后,对每个剩余边界执行圆拟合和回归,以实现重叠苹果的识别。实验结果表明,该方法能有效地从重叠苹果中分离和识别单个果实,总准确率达87%以上。

Vision system is the key part of apple picking robot which determines the accurate identification and localization of fruit targets.In the unstructured orchard,the accurate recognition of apples may be affected by many factors,and the overlapping is the most important factor.In order to realize the recognition and spatial orientation of apples,a method of overlapping apples recognition based on curvature analysis of edge points was proposed.Firstly,the method extractes the edge of apples from the binary image,calculates curvature values of edge points and removes the edge points with abnormal curvature values.And then the entire boundary is divided into several segments and filtered on the basis of three criteria,and the effective segment is retained.After circle fit and regression for the remaining boundaries,overlapping apples is recognized.The experimental results show that the method can effectively separate and identify individual fruits from overlapping apples,and the total accuracy is over 87%.