复杂背景下油茶果采收机重叠果实定位方法研究

油茶果机械化振动采摘技术关键在于振动点选取,判断振动点选取取决于果实生长密度测算和分布估计.然而自然环境下重叠果实的识别对判定结果有较大的影响,因此提出一种基于凸壳识别的分割边界优化方法,提升重叠油茶果识别与分割准确度.该方法先将原始图像转换颜色空间,经过阈值分割和形态学处理获得重叠果实的凹区域,然后在此基础上通过Harris角点检测得到区域的特征点集,利用主成分分析(PCA)和欧式距离方法分析特征点距离关系得到分割路径,最后采用最小二乘法对分割后的目标区域进行拟合重建得到果实轮廓.对比重建的果实轮廓与真实分布图像,该方法的平均定位误差为8.6％,比Hough方法低5.1％;平均耗时为0.52 s,比Hough方法低0.12 s.结果 表明,提出的方法可以有效解决重叠油茶果实识别与分割问题,为采摘装置的振动点选择奠定基础.

Research on overlapping fruit positioning method of camellia fruit harvester in complex background

The key to mechanized vibration picking technology for oil tea fruit lies in selecting vibration points, which are judged to be dependent on fruit growth density measurements and distribution estimates. However, recognizing overlapping fruits in the natural environment has a greater impact on the decision result. Therefore, this paper proposes a segmentation boundary optimization method based on convex hull recognition to improve overlapping oil tea fruit recognition and segmentation accuracy. The method firstly converts the original image into color space, and obtains the concave region of the overlapping fruits after threshold segmentation and morphological processing, then obtains the feature point set of the region by Harris corner point detection on this basis, uses principal component analysis (PCA) and Euclidean distance method to analyze the feature point distance relationship to obtain the segmentation path, and finally the least square method was used to reconstruct the fruit contour by fitting the segmented target region. Comparing the reconstructed fruit contours with the actual distribution images, the average localization error of the method in this paper is 8.6%, which is 5.1% lower than that of the Hough method; the average time taken is 0.52 s, which is 0.12 s lower than that of the Hough method. The results show that the proposed method can effectively solve the problem of overlapping oil tea fruit recognition and segmentation and lay the foundation for the vibration point selection of the picking device.