VI Seminar #92: Computer Vision in the Wild: Detecting, Tracking, and Identifying Birds and Bats at Wind Farms

Presented by Mamoona Birkhez Shami, Software Engineer at Spoor AI

Abstract

The rapid expansion of wind energy, both onshore and offshore, requires robust automated systems for monitoring wildlife interactions with turbines. At Spoor, we develop and deploy camera systems on onshore wind farms, offshore platforms, and floating buoys to detect, track, classify, and monitor collisions of birds and bats during daytime as well as nighttime.

This talk presents the core computer vision challenges in this domain and the approaches we have developed to address them. These include detecting birds with a wingspan of 25cm at ranges exceeding 1 km, species classification that combines visual features with inferred flight behaviour and stereo vision-based collision monitoring. I will also discuss our model validation methodology and the pitfalls of evaluating ecological monitoring systems in the real world. Finally, I will present open challenges, including edge deployment of multi-stage pipelines, automatic stereo recalibration after physical camera displacement, and few-shot species generalization to new deployment sites. These challenges sit at the intersection of small object detection, multi-modal time-series classification, and domain adaptation under extreme conditions, and I hope they are of interest to the visual intelligence community.