Catching annoying birds

Have you ever had a bird flying into your kitchen, only to create chaos ? And by the time you heard them and rush in to chase them away, it was already too late. That annoyed me so much that it gave me an idea to use a bird detector and scare them away with eagle sounds.

YOLOv3 Object Detection

I'm using YOLOv3 as the object detector (PP-YOLO seems to be the fastest now). Read this awesome tutorial from pyimagesearch that explain in great details how to use YOLO object detection. For my use case, I just need a reliable enough object detector and so far the YOLOv3 object detector that is pre-trained with COCO dataset is good enough on my Raspberry Pi 4.

This bird is smart enough to avoid detection

Did you spot the bird in the picture below? If it takes you more than 1 second, then I would probably think its forgivable that the YOLOv3 object detector did not manage to detect it. It has been trained on a general COCO dataset anyway. If I were to increase the accuracy of the detection, I would definitely need to build my our dataset of birds that have actually been into my kitchen, and add some data argumentation to it.

Some things in the kitchen look like a bird too

Currently I've set the detector alarm to go off as long as the confidence threshold is more than 0.5. So sometimes I'm getting a few false positives and it's pretty hilarious to know what was detected :)

In a follow up post, I'll explain in more details the code that is driving this bird catcher!