3 ms·

I gave a talk on "Deep Neural Networks for Video Applications" at the GDG DevFest conference last weekend [1]. It's crazy how powerful computer vision is and ho

I gave a talk on "Deep Neural Networks for Video Applications" at the GDG DevFest conference last weekend [1]. It's crazy how powerful computer vision is and how relatively easy it is to apply to video streams. It's not perfect but you can absolutely replace a human watching video in most situations / use cases, for example: * identifying vehicles & reading license plates * identifying people and tracking them across cameras * identifying products on the back of trucks * counting people in queues or getting in/out of vehicles / stores * detecting suspicious activity / shoplifting Generally you don't even need to completely eliminate humans watching footage, you can just present the human with cases detected by the object and let the human decide whether it's a true positive / what to do. It's scary but useful technology because there's no way humans can watch all the footage being produced by a billion cameras... [1] https://speakerdeck.com/alxcnwy/gdg-devfest-2019-deep-neural-networks-for-video-applications https://speakerdeck.com/alxcnwy/gdg-devfest-2019-deep-neural...

MisterOctober 7y ago
Strongly reminds me of "A Scanner Darkly" [the movie, at least] -- where the human scanners have to watch thousands of hours of surveillance video [which they do at increased speed until they spot something notable]
boiler_up800 7y ago
I disagree that identifying people and tracking them across cameras works well. If you have high res cameras and facial recognition then maybe.
- alexcnwy 7y ago
  Yeah sufficient resolution is definitely an issue. I'm not suggesting you can use models in all situations or an all current setups, just that it's possible (and surprisingly easy) to use models to automate a lot of analysis on video. A lot of current commodity $100 cameras are good enough resolution for many use cases but of course you do need sufficiently high res video and good enough lighting conditions. You also don't necessarily need to do facial recognition to track people. I'm nervous of the privacy implications of all this surveillance tech but it's possible to do useful tracking (e.g. I've worked on a project counting people to prevent over-filling busses which is a big issue in Africa) without needing to use facial recognition. In the counting case, you don't need to recognize people to count them and can do edge computing on quite low resolution video so that the raw footage never leaves the device (only computed count statistics).
Florin_Andrei 7y ago
> detecting suspicious activity / shoplifting This one seems hard to do right. How do you avoid false positives / negatives?
- petre 7y ago
  You present them to a human. Cory Doctorow has written some fiction about it lately. It's called Affordances I think.
  - alexcnwy 7y ago
    Yep - you don't need to avoid false positives / false negatives, you just need to reduce the search space from watching 30 cameras' video feeds at once (impossible for 1 person) to watching these clips that have a high probability of being shoplifting (possible for 1 person).

MisterOctober 7y ago

Strongly reminds me of "A Scanner Darkly" [the movie, at least] -- where the human scanners have to watch thousands of hours of surveillance video [which they do at increased speed until they spot something notable]

boiler_up800 7y ago

I disagree that identifying people and tracking them across cameras works well. If you have high res cameras and facial recognition then maybe.

alexcnwy 7y ago
Yeah sufficient resolution is definitely an issue. I'm not suggesting you can use models in all situations or an all current setups, just that it's possible (and surprisingly easy) to use models to automate a lot of analysis on video. A lot of current commodity $100 cameras are good enough resolution for many use cases but of course you do need sufficiently high res video and good enough lighting conditions. You also don't necessarily need to do facial recognition to track people. I'm nervous of the privacy implications of all this surveillance tech but it's possible to do useful tracking (e.g. I've worked on a project counting people to prevent over-filling busses which is a big issue in Africa) without needing to use facial recognition. In the counting case, you don't need to recognize people to count them and can do edge computing on quite low resolution video so that the raw footage never leaves the device (only computed count statistics).

Florin_Andrei 7y ago

> detecting suspicious activity / shoplifting This one seems hard to do right. How do you avoid false positives / negatives?

petre 7y ago
You present them to a human. Cory Doctorow has written some fiction about it lately. It's called Affordances I think.
- alexcnwy 7y ago
  Yep - you don't need to avoid false positives / false negatives, you just need to reduce the search space from watching 30 cameras' video feeds at once (impossible for 1 person) to watching these clips that have a high probability of being shoplifting (possible for 1 person).