Abstract In chicken egg production line systems, grading based on vision systems is challenging due to ambient light conditions and egg occlusion problems. This study introduces a depth image-based chicken-egg volume estimation system. Two modes of egg configurations on a sorting line were evaluated; single-egg (no occlusion) and multi-eggs (partially occluded, i.e., simple and complex). Contour curvature analysis and k-closest M-circle-center algorithms were used to segment the occluded eggs. Thirteen regression models based on the egg image (single egg) features were trained. The Exponential Gaussian Process Regression outperformed all the explored models with RMSE of 1.175 cm3 and R2 of 0.984. The same model estimated the volume of the eggs under partial occlusion at RMSE of 1.080 and 1.294 cm3 for simple and complex, respectively. This introduced system can be applied as an accurate, consistent, fast, and non-destructive in-line sorting technique of chicken eggs in a production line system.

中文翻译：

---

基于图像处理和计算机视觉的鸡蛋体积估计

摘要 在鸡蛋生产线系统中，由于环境光条件和鸡蛋遮挡问题，基于视觉系统的分级具有挑战性。本研究介绍了一种基于深度图像的鸡蛋体积估计系统。对分拣线上的两种鸡蛋配置模式进行了评估；单蛋（无遮挡）和多蛋（部分遮挡，即简单和复杂）。轮廓曲率分析和k-closest M-circle-center算法用于分割被遮挡的鸡蛋。训练了 13 个基于鸡蛋图像（单个鸡蛋）特征的回归模型。指数高斯过程回归优于所有探索的模型，RMSE 为 1.175 cm3，R2 为 0.984。相同的模型分别估计了简单和复杂的 RMSE 为 1.080 和 1.294 cm3 在部分遮挡下的卵子体积。