Difference between revisions of "BirdGuard"

This project aims to develop a solar-​powered and computer vision-​based sticker that can be mounted at the outside of a window to visually monitor the birds, estimate their trajectories, and deter them if an imminent collision is predicted, saving their lives.

BirdGuard is an applied research project at CSEM Zurich, offering interested students to contribute solving a real-world problem with state-of-the-art machine learning (ML/AI) technologies to make the planet more birdfriendly and enable a respectful cohabitation with nature. The goal of this project is to develop the required algorithms and deterrence mechanism to build a prototype for field testing. CSEM has developed hardware and software solutions for a wide range of ML applications. Its latest machine learning platform (shown below) features solar energy harvesting to enable battery-less operation, an image sensor, and CSEM’s ML system-onchip for processing acquired images (e.g. detection algorithms). This system will be used as a basis for prototyping the BirdGuard algorithms and the deterrence sub-system. Swiss research institutes are at the forefront of researching birds and applying this knowledge to make houses more bird-friendly. Thus, our partner institutes will serve as advisors to the project.

Track A): Development of bird detection/tracking system

Robust detection and tracking of birds is essential to monitor their trajectory and estimate the collision probability in real time. Machine learning-based detection algorithms have revolutionized the field of computer vision and will thus be adopted for this application. The algorithm shall be validated on a camera/PC setup in the field and later optimized for low power execution on the ML chip to enable continuous tracking with a limited power budget.

Track B): Evaluation of deterrence and deflection system

To warn a bird on a collision trajectory early enough to change its course, an effective warning mechanism (bird understands that there is an obstacle) must be developed, ensuring low-latency (bird has sufficient time to change course). This requires studying and evaluating potential mechanisms and the birds’ flight behavior (minimum required warning time and distance). The results shall be implemented in a prototype and validated in a field pre-study.