When I ran the DeepLens model inference to attempt to detect a bird in my window, the model was unable to recognize the birds. However, when testing this object detection algorithm using different types of objects (such as "person" or "cat" or "dog"), those types of objects were all detected fine.
I was able to get a picture of a bird to trigger my object detection logic, and used that for my hackathon submission. However, I would like to learn how to improve the object detection model's accuracy when detecting birds in the wild. This is something that I will be using SageMaker to learn in the future.
Here are some items that I considered doing, but ran out of time.
- Detect when a squirrel is in view, and send a SMS (text) notification. In case you aren't a bird watcher, squirrels are notorious for eating massive amounts of bird food.
- Use AWS Rekognition to annotate images with metadata information that can be detected by the service.
- Use AWS SageMaker to determine what species of bird is being viewed, and add that into the mix.
Try it out
The repo for this projectis private