Luke Ditria's Shrunken AI Wildlife Cam Cuts the Size and Price with a Raspberry Pi AI Camera

Maker and teacher Luke Ditria has built a compact gadget designed to identify passing wildlife — thanks to a Raspberry Pi Zero 2 W and the Raspberry Pi AI Camera accessory.

"I've made a few videos now about this concept," Ditria explains by way of introduction to the project, "and I've had a lot of interest from people wanting to start to build their own. Now, the original version, which had a Raspberry Pi 5 and Hailo AI accelerator, cost quite a bit and was a bit inaccessible for some, so I decided to create this mini version which is not only a lot more compact but it's also a lot cheaper."

Ditria's original smart camera system, inspired by trail cams, relied on the relatively high-performance Hailo AI accelerator connected to the top-end Raspberry Pi 5 single-board computer to process incoming video and run a machine learning model capable of distinguishing and identifying animals. The new version does exactly the same thing, but in a more pocket-friendly way — thanks to the Raspberry Pi AI Camera.

Launched back in September last year, the Raspberry Pi AI Camera looks like any other Raspberry Pi Camera Module — but its imaging sensor is a Sony IMX500, positioned by the company as an "intelligent vision sensor." This integrates vision-based model acceleration on-sensor, running models capable of fitting in its dedicated 8MB of RAM before the video is transferred to the host device — in this case, the Raspberry Pi Zero 2 W single-board computer. As a result, the processing takes up no host CPU and requires no separate accelerator — perfect for a budget-friendly build.

"I'm just using a YOLO[v]8n [model] trained on a whole bunch of Victorian bird species," Ditria says of the model running on the IMX500 sensor. "It's constantly streaming the detection information from the camera, and once it detects a high enough confidence detection it will save the image and log the detection to a little USB [flash drive] that I've got in there, and at the end of the day I can take the USB out, plug it into my computer, and see the results we've got."

The project is documented in the video embedded above and on Ditria's YouTube channel; STL files for the 3D-printed trail cam-like enclosure have been uploaded to Maker World under the Standard Digital File License.