75. Automated Computer Vision Quantification of Chute-exit Behavior and Associations with Performance and Carcass Outcomes in Nellore Cattle.

Abstract Chute-exit behavior has been proposed as an indicator of cattle temperament, yet traditional scoring is subjective and difficult to scale. We developed a fully automated computer vision pipeline to quantify flight speed and related behavioral/postural metrics from short videos of individual beef cattle exiting a handling chute and evaluated associations with subsequent growth and carcass outcomes. Videos from 108 Nellore Bulls (390.2 ±31.1 kg) were captured, at the beginning of the feedlot period, with a fixed GoPro camera (2,704 × 1,520 pixels; 59.94 fps) and processed in Python using Ultralytics YOLO11 for cattle head detection with ByteTrack tracking and an interactive region-of-interest (ROI) polygon to restrict detections to the chute-exit corridor. Kinematic variables were computed from the centroid trajectory of the detected head across time, including mean and peak flight speed, maximum acceleration, path tortuosity, exit direction, and hesitation time near the chute entry zone. Virtual mark lines were also defined to estimate crossing times and split speeds between corridor segments, enabling localized speed measures (e.g., mark 1→2). For statistical analysis, linear mixed-effects models were fit to test associations between early chute-exit behavioral predictors and productive performance outcomes, including hot carcass weight, dressing percentage, final body weight, average daily gain, feed efficiency, and dry matter intake (DMI). Initial body weight was included as fixed effect and diets and animal nested within diet was included as random effects. Multicollinearity was managed through iterative variance inflation factor pruning. Split speed from mark 1→2 (1-meter distance) was negatively associated (P < 0.05) with hot carcass weight, final body weight, and average daily gain, and feed efficiency (P = 0.0593) indicating that faster early exit speed was related to lower growth and carcass-related outcomes. For DMI, average exit speed was positively associated (P < 0.05), with faster-exiting cattle exhibiting greater feed intake. Higher intake in these animals may be linked to differences in efficiency (e.g., higher intake without proportional gain). These results demonstrate that automated, video-derived chute-exit metrics can be generated at scale and can reveal biologically meaningful relationships with performance traits, supporting practical deployment of objective temperament and behavior phenotyping in feedlot systems.