Low-latency Machine learning for smart Shopping system

Sounds VERY invasive.

I don't like the idea of it much myself but the idea is to use video information so it's not morally different from having an assistant standing outside the shop eyeballing the punters. (Some of the less reputable Soho establishments used to do that.)

It looks as if the proposal is all vaporware so I'm not too worried about this specific idea yet.

But what is not too much of a problem when it's done inefficiently becomes quite different when the technology really bites.

More worrying (to me) is if someone big (you know the companies I mean) did it and correlates facial recognition with other data to make instant positive IDs of passers by. I'm expecting laws against this (and similar) within a year or two. Perhaps the answer would be to oblige anyone who uses personal data about an individual for the first time (or for a new purpose) to be obliged to get specific permission, on pain of a fine.

MK

michaelkellett  wrote:

 

I don't like the idea of it much myself but the idea is to use video information so it's not morally different from having an assistant standing outside the shop eyeballing the punters. (Some of the less reputable Soho establishments used to do that.)

That's true (though I tend to avoid those types of store for that very reason), but such human assistants don't have direct API access to database of historic customers and their transactions.

This part sounds very much like facial recognition to me.

Just imagine the customer feeling when he hear a warm welcoming by his name from the system and sees his favorite products and recently added related products displayed on the front monitors.

Feels a bit reminiscent of this:

Edit media

Dimensions x Small Medium Large Custom

Subject (required) Brief Description Tags (separated by comma) Video visibility in search results Visible Hidden

Parent content

RE: Low-latency Machine learning for smart Shopping system

Poster

Upload Preview

Ok

Cancel

At a trade show I saw a camera/GPU system that people-counts, but also identifies if it has seen the person before or not, and also estimates age, gender, etc, in real-time. There are also lower-cost (non-camera/GPU) solutions that can very precisely identify (granularity to within a few feet) if the same person is hanging around a particular shopping aisle too long or re-visiting it (without identifying the specific person), in order to either alert security, or provide targeted instant discounts if you "buy it now". I bet that the way businesses will get the permission, will be to offer free wi-fi, on the condition that they can get the targeted adverts on their phone etc. Many people will do a lot for free wi-fi. From what I've seen I believe (I don't have evidence, possibly not in the interest of mobile providers to reveal it!) a large chunk of the public at least in the UK closely monitor their usage due to being on poor data plans, and often switch off mobile data.

Personally I wish the government would do more about the intrusion with unsolicited phone calls and SMSs. TPS is useless. The fine could be unlimited, but it doesn't help until the public get better tools to identify and flag such abuse, forcing phone companies to check the calling number against the presented number when such abuse is flagged, and stricter regulation of the accident claim/law firms - the Law Society is primarily concerned with its members, not the public).

shabaz  wrote:

 

Many people will do a lot for free wi-fi. From what I've seen I believe (I don't have evidence, possibly not in the interest of mobile providers to reveal it!) a large chunk of the public at least in the UK closely monitor their usage due to being on poor data plans, and often switch off mobile data.

You're not wrong there!

The information people are willing to disclose for "free" WiFi is quite simply staggering, not to mention any traffic interception, analysis that is going on.

I opted for a 4G data-plan at around 20GB when on home territory, meaning I don't ever have to connect to public hotspots. Much safer from a security perspective.

Thank you for the video. I haven't seen it before.

I don't want it to be like a science fiction. I want to make a practical product that can be marketed.

What do you think is missing to be marketable?

Thank you for your opinion. I would like to update my proposal according to our discussion.

So, about the feasibility of the system on the Terasic DE10-Nano, do you have any comment about the system block diagram?

If I were judging this competition I would want to see some calculations that support your assertion that the CNN (DNN ? you call it different things in the text and the diagram) will run on the FPGA fabric but wouldn't run on the Cortex A9s. I'd also want some evidence that it coudn't run on faster or DSP type processors.

The block diagram is much too high level to be convincing - you need to show that you have thought about how much internal block RAM you need, about data path bandwidths and use of multiplier/DSP type resource.

Then you need to talk about development time !

MK

Your biggest hurdle might be take-up by companies who may be concerned about the negative backlash, from customers who don't want to be tracked and targeted electronically.