利用深度攝影機進行自動化的豬隻體重量測

Abstract

豬隻體型量測是豬隻養殖戶重要的工作之一，在育種、檢視飼料效率、豬隻健康控管上皆扮演重要角色，其中體重也是決定豬隻是否達到出貨標準的條件。目前豬隻體型量測以人工為主，逐一量測體型是十分耗費體力的工作，豬隻在受到控制的時候會用力的掙脫，過程中常常導致量測人員受傷，也容易使豬隻產生壓力，對其健康造成不良的影響。 本研究提出一套非接觸式的豬隻體重量測系統，使用單一RealSense深度攝影機拍攝欄位中豬隻的彩色影像與深度影像，藉由Yolo物件辨識模型在彩色影像中偵測出豬隻所在位置，再使用RCF邊緣偵測模型擷取出範圍內豬隻身體的封閉輪廓。使用深度影像所提供的距離資訊，以針孔成像原理將輪廓內面積還原成現實尺度的可視面積，以影像實現簡單又有效率的豬隻體重量測。 本系統減少需要人為決定的參數，使系統在不同光源及環境下都可以正常使用，豬隻不同的位置與站立方式之下也能進行可視面積的修正，增加系統的強健性。在研究中我們對同樣的豬隻進行前後兩次的體型量測，觀察到投影面積及立體模型的表面積在豬隻生長過程中有顯著的成長，豬隻體重也與投影面積有一定的關係。

Measuring pig size is one of the import tasks for the farmers. It plays an important role in breeding, check feed efficiency and health control. Among all the body features, weight is used to determine whether a pig is reaching the shipping standard. These days, pig body measurement is mainly manual. It is a labor-intensive job. When a pig is under control, it will struggle to escape, which often causes farmers to be injured. It also put stress on pigs, which has a negative impact on pig’s health. This research proposes a non-contact pig weight measurement system that uses a RealSense depth camera to capture color images and depth images. We use Yolo object detection model to find the location of the pig in color image, and use RCF edge detection model to extract the close contour of the pig. We use pinhole imaging principle to calculate the realistic scale area of the contour by the depth information from depth image. A simple and efficient pig body weight measurement can be achieved with images. The system reduces the parameters that need to be determined manually, so that the system can be used normally under different environments, and the pig contour area can be corrected under different positions and postures of the pigs, which increases the robustness of the system. In the study, we took two measurements on the same pigs before and after, observed the contour area of pigs and the surface area of the pigs model. We found that contour area and surface area had significant growth during the growth of the pigs, and the weight of the pigs also had a certain relationship with the projected area.