Yes that would destroy some data points but we are talking about a statistical approach here. How often do you give other people cash directly in comparison to spending it at stores. Given enough time and collected information local concentrations will show up. Its surveillance through data analysis with a new veneer.
Netzpolitik recently did an article about that. I consider them a credible source. How often bills are used before there are returned to the bank heavily depends on the denomination. Larger bills don’t circulate as much and at least in my country most stores return their cash income to the bank on a daily basis. People also tend to spend their money around the area where they live, so even if you couldn’t figure out which exact store a targeted person spends their money at due to circulation (which I doubt), you can still quickly find the general area in which they are staying.
Given a large enough time frame this can be treated as random noise which is easily filtered out, and this data isn’t necessarily meant to track your supermarket shopping. For example, you can use it to figure out where somebody went who has gone into hiding. They might have cleared out their bank accounts before leaving and with that data you can see where these banknotes are now showing up. Just wait at the store they apparently visit every Tuesday.
Banks can track each banknotes serial number when you receive them from the ATM and when they are returned from the store you spent them at. This data could then be used to create a complete profile of your spending habits.
This still fundamentally suffers from the oracle problem like all blockchains solutions. You can always attack these blockchain solutions at the point where they need to interact with the real world. In this case the camera is the “oracle” and nothing prevents someone from attacking the proposed camera and leveraging it to certify some modified footage. The blockchain doesn’t add anything a public database and digitally signed footage wouldn’t also achieve.
No manual intervention needed just track everything and retain the data for a long time. Then you can query the database for all serial numbers your target has ever taken out of an ATM and where they where returned. Circulation will diffuse these results somewhat but that follows a normal distribution. That means it is much more likely that a bill that was spent at a store by your target which is given out again as change, is more likely to be spent at another store near that original store. Aggregate this information over a couple of weeks or months and you will get a heatmap of locations to check out. You can run this operation automatically on all bank accounts if you want. Threshold the heatmap at some significant value and you get a list of stores visited by each person with X% probability.