Length measurement system to assist automation in the slaughtering process

The meat industry sees a development towards more local, sustainable and self-sufficient food. This means the whole process of livestock farming, butchering and distributing. If a slaughtering process were automated, it would replace the hardest work of a butcher and make it easier to produce local meat products. This is the aim of a larger project which this thesis is a part of; the primary task is to develop a system to measure the length of pigs so that a machine can adapt to the biological variation inherent in the raw material. Hence, the goal with this thesis is to make a precise, robust and contactless measurement system. This thesis begins with addressing the properties of sensors and cameras for measurement purposes. A depth camera had the best potential and was chosen for further usage. Furthermore, the thesis discloses the applications for recognition types, and distinguishes between different detection and segmentation models. Three models were selected for testing and aligned next to each other for comparison. When a pig is located in a picture, it is possible to reduce the amount of data in the depth image and processing it to a 3D point cloud. The point cloud needs filtration due to a rough and noisy surface. At this point, the nose and hind feet can be pinpointed and the distance between them calculated to obtain a carcass length estimation. To get the results, a depth camera was installed in a pork slaughter-line (Fatland, Oslo). The placement of the camera was adjusted during recording, which led to a consequential measurement error. Unfortunately, it was not possible to get more data as the industry no longer allowed outsiders to visit the plant during the COVID-19 outbreak. As a consequence of the measurement error, the system’s accuracy was unobtainable. However, the precision could be found by adjusting the manual measurement using a calibration ratio. A few comparisons were made of the length differences, and the total error was ±4 cm. The accuracy and precision can be improved by a variety of ways. If the pigs in the production line would stop at the exact place, the error is likely to decline. Additionally, some of the aforementioned operations may be easier to implement as a result of this