It’s for the Birds
This solar-powered wildlife monitor uses a Raspberry Pi AI HAT+ to track birds, and it may soon be used in real-world conservation efforts.
Whether you want to further conservation efforts around the world, take a peek at the secret lives of the critters in your backyard, or just learn some new things about the latest technologies, building a wildlife monitoring system is a great way to accomplish your goals. YouTuber Luke Ditria has been hard at work on a variety of wildlife monitors for several months now, and with the lessons learned from past builds he decided to build a more energy-efficient system that might be more practical for real-world use cases, especially in remote areas.
The main purpose of the monitor is to automatically detect and log the presence of certain animals of interest. In this case, Ditria wanted to track the comings and goings of about 30 species of birds that are commonly found in his neck of the woods. This was made possible through the use of a YOLO object detection model, which captures images at regular intervals, then analyzes them to check for known types of birds.
To ensure that the system can be used anywhere, the machine learning algorithm runs entirely locally, such that a wireless connection to remote resources is not necessary. It is also equipped with a solar panel, which serves to keep an uninterruptible power supply topped off.
The plastic case is sealed to keep the components inside clean and dry, even when the weather outside gets nasty. The hardware components were carefully selected — to run a computer vision algorithm, a fair amount of computational horsepower is needed. But when you are running on solar power, you also have to minimize energy consumption. Ditria found that the Raspberry Pi ecosystem struck just the right balance to make this build work as intended.
A Raspberry Pi 5 single-board computer is at the heart of the monitor for any general-purpose computing tasks. To run the YOLO algorithm as fast and efficiently as possible, it is offloaded to a Raspberry Pi AI HAT+ with a powerful Hailo-8 AI accelerator that offers up to 26 TOPS of inferencing performance. There is also, of course, a high quality camera connected to the system to capture images of the outside world.
The algorithm was limited to running at five frames per second — which is fast enough for bird recognition — to minimize energy consumption, which was measured at about 4 to 5 watts. Custom software also limits the monitor to running only during daylight hours. This prevents the batteries from being drained while the Sun is down, and there is no point in running during the night when the camera cannot capture anything of value anyway.
Right now Ditria is mainly using the monitor in his backyard, but he has been contacted by some ecologists that are interested in the device. So in the near future, it may be helping with some real-world research projects. Looking ahead, Ditria is exploring the possibility of adding LoRa transmission capabilities to the system for energy-efficient long-range communications, which would come in especially handy if a number of these monitors were installed in remote regions. Be sure to check back later for updates to see where this project goes next!