Making Animal Re-Identification Accessible: A Web-Based Giraffe ID System for Zoos

Computer vision and machine learning have accelerated the automation of animal re-identification pipelines used in conservation programs worldwide. For species with distinctive markings, such as the spot patterns of giraffes, these automated methods are crucial for research and population monitoring purposes. However, many tools are designed for experts, and their implementation requires substantial technical expertise. Research teams often use specialist software and workflows that are not accessible to the general public. In a zoo setting, visitors lack a simple way to identify an individual animal, and unique features are easily missed by untrained visitors. This study presents a three-part solution: a web interface for zoo visitors to upload photos, a deep learning model for giraffe torso detection, and a fast re-identification method for matching observations to a gallery of known individuals using server-side processing. We compare several re-identification methods (RootSIFT, MiewID, and MegaDescriptor) using a consistent evaluation protocol and report both identification performance and system latency for this closed-set zoo setting. Taken together, this study presents a visitor-facing web system that integrates existing re-identification models into a modular, real-time pipeline for zoo deployment, lowering the barrier to visitor participation and making state-of-the-art re-identification methods more accessible to the general public.