Objective: Obfuscate license plates and faces on exports so that we can maintain privacy of individuals in the drive 2FPS images.
The outcome of this card is to produce a prototype and have a good definition of the technical work required to fully integrate face/license plate obfuscation into exports
For the first step, we will talk about how we’d like the architecture/solution for this to look. A few key things:
how you will pull bounding box data
how you will tie this into the export code
where the obfuscated images will be stored
which tool to use to obfuscate (e.g.GraphicsMagick)
consider long-term architecture for obfuscation user upload / replacement of raw images with obfuscated images to protect privacy
obtain a data set of images for licenses and faces to test on (spoiler: we have a few😃)
For the prototype, create a standalone script (not integrated into the driver-cloud repo) that takes an image, an [x,y] coordinate for start, and a [width,height] dimension. The script will draw an opaque box that keeps pixelated properties of the original image, but obfuscates license plate numbers and faces. The license plate number should not be intelligible within a reasonable distance from the camera (6-10' min).
Objective: Obfuscate license plates and faces on exports so that we can maintain privacy of individuals in the drive 2FPS images.
The outcome of this card is to produce a prototype and have a good definition of the technical work required to fully integrate face/license plate obfuscation into exports
For the first step, we will talk about how we’d like the architecture/solution for this to look. A few key things:
how you will pull bounding box data
how you will tie this into the export code
where the obfuscated images will be stored
which tool to use to obfuscate (e.g.GraphicsMagick)
consider long-term architecture for obfuscation user upload / replacement of raw images with obfuscated images to protect privacy
obtain a data set of images for licenses and faces to test on (spoiler: we have a few😃)
For the prototype, create a standalone script (not integrated into the driver-cloud repo) that takes an image, an [x,y] coordinate for start, and a [width,height] dimension. The script will draw an opaque box that keeps pixelated properties of the original image, but obfuscates license plate numbers and faces. The license plate number should not be intelligible within a reasonable distance from the camera (6-10' min).