The video was likely recovered from local flash memory on the camera itself. These kinds of devices are not uploading raw video to the cloud.

There are several reasons for that. The first is that you cannot rely on connectivity 100% of the time. Second, if you can have the camera run image processing and compression locally, you don't have to dedicate a massive amount of processing resources at the data center to run the processing. Imagine ten or a hundred million cameras. Where would you want the image processing to run? Right.

My guess is that they either went to Google to perhaps connect the camera to a sandboxed testing rig that could extract locally-stored video data or they removed the flash device, offloaded the raw data and then extracted video from that data. This last option could also have the advantage of having less compression (architecture dependent).

Decades ago I was personally involved in recovering and helping analyze surveillance video data for the prosecution in the OJ Simpson case. Back then, it was tape.

One of the techniques that was considered (I can't publicly state what was actually done) was to digitize raw data right off the read heads on the VCR's spinning drum. You could then process this data using advanced algorithms which could produce better results than the electronics in even the most expensive professional tape players of he era.

Once you step away from the limitations of a product --meaning, you are not engineering a product, you are mining for information-- all kinds of interesting and creative out-of-the-box opportunities present themselves.