Facial recognition on RT1062

OpenMV Boards
1

Hello, we are trying to do some facial recognition experiments with the RT1062. This doesn’t have to be fancy or even very fast. We’d simply like to see if the camera can determine if a face belongs to a specific person. I found a few posts about this, which all seemed a bit discouraging but didn’t provide many details. We are completely new to the OpenMV world.

From what I gather, there are some facial recognition models people have tried with OpenMV (where?). I am not sure how we would train one of those models to recognize a specific person. I gather we’d use EdgeImpulse but it’s all a bit overwhelming.

Any guidance is greatly appreciated!

2

As a test, I tried running the tf_object_detection_1 example. It looks like that uses the FOMO model to detect faces. However, when I try to run it on the RT1062, I get an ENODEV error at line 26:

model = ml.Model("/rom/fomo_face_detection.tflite")

I guess this means the model isn’t on the ROM? Any example I’ve tried that references /rom gets an ENODEV error. What am I doing wrong?

3

Hi, you need to install ROMFS on the camera. Please click the firmware version displayed in the status bar, it will ask you if you want to reset to the default firmware, say so and select the ROMFS as one of the things to install back to default.

4

Note, we’re about to do a firmware release that brings blazeface to the RT1062. This runs at about 2 FPS but is much better than fomo face detection.

As for facial recognition, we’ll be able to do this on the N6 and AE3.