Enhancing welfare assessment: Automated detection and imaging of dorsal and lateral views of swine carcasses for identification of welfare indicators

High animal welfare standards are essential for sustainable pork production. Current on-farm welfare assessments present challenges including cost, time consumption, and biosecurity risks. This study presents the initial step towards an automated animal welfare assessment system for swine carcasses. We introduce a computer vision system for capturing dorsal and lateral views of pig carcasses. The proposed system consists of a tracking and image acquisition system, which captures comprehensive lateral and dorsal views of pig carcasses as they pass by a camera installed at a slaughterhouse. During evaluation, the system achieved an accuracy of 92.5 % (74 out of 80 detections in a video with 40 carcasses) with minimal false positives and undetected positions, operating in real-time at 41.65 frames per second (FPS) with high confidence. The Detection manager module led to a remarkable reduction in erroneous detections, dropping from 38.75 % to 3.75 %. In addition to tail position data, leveraging head position data as an auxiliary mechanism substantially reduced missed detections to 6.25 % and erroneous detections to 1.25 %. This underscores the critical role of auxiliary mechanisms in enhancing system performance. This system marks the initial phase in developing an automated welfare assessment tool, paving the way for real-time welfare monitoring of carcasses in pork production.