Straight Huntaway sheepdog trials currently rely on subjective human analysis, lacking automated frameworks for performance evaluation. This research proposes an end-to-end computer vision pipeline designed to quantify trial performance from recorded trial footage. The system employs a YOLOv26 detector to extract precise coordinate data for both sheepdog and sheep from video sequences, which are subsequently engineered into interactional behavioural metrics. Performance is evaluated via an ensemble of XGBoost regressors, achieving a mean absolute error (MAE) of 1.984, a coefficient of determination (R2) of 0.487, and a Spearman’s rank correlation coefficient (ρ) of 0.685.

Feature significance analysis identified the sheep flock’s mean lateral deviation from the optimal straight path and mean velocity towards the finish line as the primary predictors of a high scoring trial. These results demonstrate the viability of using an automated vision-based system for evaluating sheepdog performance, though the remaining variance suggests further refinement is required to fully encapsulate the more nuanced heuristic criteria used in human analysis.