Pre-Trained Model

We have a project using an ESP32-Cam with the Camera and it needs to do person detection with an image from the camera every X seconds (Prob > 30seconds) and if there is a person detected trigger some code to store the image. Hence we are hoping to use an existing model, however would like to train the model to recognise staff members and if these are seen trigger a different code path but that is a Nice to have rather than a need to have.

All image processing needs to be done on device and at present we are looking at an ESP32Cam for battery performance.

Is this possible and has anyone know of a Blog/Docs that can help as we are new to Edge Impulse but keen to learn.


You can check out what I did for a bird feeder: Solar Powered TinyML bird feeder. That might help although it was just a binary classifier (bird/no-bird) rather than multi-class. Good luck.

Hello @CliffAgius,

As @jlutzwpi suggested, I would use a binary classifier to do so. Keep in mind that if you use the ESP32-Cam, you will need small images 64x64 with MobileNetv2, or 96x96 with MobileNetv1 transfer learning as the board resources are limited. So it might be a bit harder to train your model.

Let us know how your project goes,