Automated individual walking distance of group-housed broilers; A comparison between ground-truth, RFID, and video

Individual broiler behaviour could serve as a proxy for an individual’s health, welfare and performance. However, large-scale individual behaviour recording is difficult. Wearable and remote sensors have both been investigated to record individual broiler behaviour. In this study, we validated the use of automated tracking using video or radio frequency identification (RFID) by comparing the total estimated walking distance to ground-truth annotated from video. By using colour-marked birds we simplified the computer vision task of YOLOv5. The estimated walking distances by video pixel-based closely matched ground-truth pixel-based (≤2 cm difference). Overall, the RFID-system also closely matched ground-truth pixel-based, except for one movement event (75 cm underestimation). Over the entire duration of the trial, RFID and video gave similar animal rankings, but RFID seems more error prone due to missing reads. RFID is suitable to track broilers over time, but video could be more accurate and practical for on-farm application.