For this interview, ipXchange headed to the Discover Hall to chat with Enrico from Birdwatch AI. Based on a Raspberry Pi single-board computer and a camera pointed at a seed plate, this embedded-AI project uses TensorFlow to identify when a bird is feeding in your garden.
As Enrico explains, the camera was placed 30-40 cm from a target area, and various levels of image filtering, including movement and RGB information, were used to differentiate the live image from the base image when a bird is not present in order to train the AI.
Unfortunately, Enrico had only been training the AI for three weeks, and only a single type of bird had visited in this time, so for the time being, the project cannot yet label different bird species though this will be part of the eventual aim. Enrico also outlines some important obstacles to implementing machine vision in robotic systems, which is worth a listen to those working on similar projects.
Stay up to date with Enrico and Birdwatch AI’s progress here.
With AI in mind, did you know that there’s a new microcontroller on the scene for AI vision applications? ipXchange first talked to Alif Semiconductor at Embedded World 2023, and we’re sure that their chip would be a great solution for identifying birds, so check out our technical writeup, and if you’ve got a commercial application, apply for a development board.