A project by drabanus for allowing research teams on artificial intelligence approaches to generate massive image and video datasets without compromising personal rights and privacy of the persons in thoise datasets.
-
Define the pipeline architecture. must be possible to load onto deep learning accelerator platforms like Movidius, Jetson-Nano, Sipeed-MAix, Edge-TPU and others (this list needs to be watched closely, as about monthly there are new hardware platforms being launched)
-
Obtain a dedicated hardware for developing, testing and training the models
- Define the process for image data (make person recognition from image impossible or at least very unlikely)
- define the process for video data (make person recognition from video impossible or at least very unlikely)
- Define the process for audio data (make voice recognition imnpossible or at least very unlikely)
-
For Prototyping, select hardware platfom to run the pipeline on.
-
Team up with users for this tech and launch prototype surveillance camera and verify the image/video quality as well as the lack of biometric information.