Automated obstacle detection in dog agility course maps

Author(s)
Tang, Junjie
Advisor(s)
Joseph, Daurette
Editor(s)
Associated Organization(s)
Supplementary to:
Abstract
This thesis presents a computer vision pipeline for automated obstacle detection in dog agility course maps, aimed at improving canine safety during training and competition. Leveraging Meta’s Segment Anything Model (SAM), the system segments agility maps into candidate regions, which are then filtered, clustered, and classified using template matching. A parallel deep learning classifier based on ResNet18 is also evaluated. Extensive testing on labeled agility maps demonstrates that the rule-based method outperforms the ResNet model in accuracy and interpretability, particularly in distinguishing visually similar obstacles. The proposed system lays a robust foundation for future development in injury-aware route planning.
Sponsor
Date
Extent
Resource Type
Text
Resource Subtype
Undergraduate Research Option Thesis
Rights Statement
Rights URI