Computer vision-based detection of locomotor activity patterns associated with play behavior in chickens as a positive welfare indicator: A proof of concept for precision poultry farming

• Proof of concept for automated detection of positive welfare in chickens • Standardized testing procedure stimulating worm running behavior • YOLO vision system with 85.7 % MOTA, 94.7 % precision, and 92.5 % recall • Enables objective 24/7 welfare monitoring and AI tools for farm-level assessment • Identification of specific activity pattern that allows for comparison to baseline The proof-of-concept study conducted under controlled experimental conditions demonstrates that automated computer vision reliably detects locomotor activity patterns associated with 'worm-running' events as an expression of play behavior in chickens with high accuracy. Combining the worm-running test with AI-based video analysis enables objective monitoring of an activity proxy associated with a putative positive welfare indicator and addresses the gap in current welfare assessments, which predominantly focus on negative indicators. A total of 210 dual-purpose chickens (> 15 weeks of age) were recorded in their home pens using a top-view camera. To stimulate play, a worm-like object made from twisted brown paper was introduced. Over 600 minutes of video were analyzed using a YOLO-based automated detection and tracking system, with manual annotations used as ground truth. The methodology achieved a Multiple Object Tracking Accuracy of 85.7%, an Identification Precision of 94.7%, and an Identification Recall of 92.4%. These findings confirm that computer vision reliably detects locomotor activity patterns associated with worm-running events. The study demonstrates the technical feasibility of objectively and automatically detecting locomotor activity patterns associated with play behavior in chickens, thereby supporting the development of AI-based tools to monitor an activity proxy linked to a putative positive welfare indicator in poultry farming.