There have been some discussions and responses in the past about how frequent capture is suggested (one thread is How many times to take the same road? , and I recall somebody from Mapillary saying that once per month is their take on it - but cannot find that particular response right now).
If one sequence is taken forwards, then sideways sequence would still be useful on the next day - but could somebody from Mapillary please comment on how often would they like to see the same sections/directions captured?
Of course, assuming at least some daylight and usefulness of those images for OSM as a golden standard.
This kind of raises another question. Object detection seems to be a function of resolution and frame rate. ie a 4K image at 1FPS maybe equivalent to 1080p size at 4FPS and 720p at 16FPS. So far my 4K images only look “as good” as 1080p size due to the quality loss of mp4-jpg conversion. Then there is uploaded filesize, ISP cost and quality. Is there a hard mathematical spec that Mapillary has defined as optimal?
I suspect Mapillary might not appreciate having the same stretch of road uploaded every day, with 4 directional cameras.
As mentioned, I recall somebody from Mapillary saying monthly frequency is what “backend could support” - which I take as a combination of processing and storage costs. But that was some time ago.
I’d be very interested in the official position currently.
Of course, if a sequence was made yesterday, and a bridge is washed away overnight, new images today would be very useful. But I’m thinking more of a usual European city street - or a countryside road.
I do not think that technology or monetary cost is the core issue here. The question is
When does recurring capturing become surveillance?
If you do it daily then isn’t that like patrolling streets in a prison camp? OK, maybe then let’s capture weekly. Wouldn’t that be comparable to prison cell inspections? Would things be better if captures would be announced for a certain day with time of day, or should captures happen unannounced? What if somebody adjusts their way of life to avoid being captured (for what ever reason)? Streets do not have to be captured factually, just the sole notion that a street can be captured at any time (no matter over what time interval) is enough to change people’s behavior.
I agree that there are different aspects to this question - and probably not a single easy answer. Generally re-capturing is most valuable when:
There have been significant changes on the street (for example in busy urban areas you might expect businesses to turn over on a monthly basis, where as a country road may not change much for a period of years)
You are able to capture with higher fidelity than the one available previously (resolution, density, view angle, or GPS accuracy)
@GITNE, a great point on imagery turning into surveillance.
The “prison cell inspections” part seems to be a bit too over the top ;), but one can surely talk about increased frequency turning into a surveillance-camera like operations.
@boris, that definitely makes sense, and “if something changes, new imagery is warranted” approach seems to be widely used.
But let’s assume nothing changes (or one doesn’t know whether anything has changed) - what would be the lowest frequency Mapillary would find reasonable?
@Richlv If nothing substantive has changed and the capture quality is not going to be better, there is no need to re-capture IMHO. Also, there are still uncaptured areas, and generally I would prioritize capturing those over re-capturing an area with no significant changes.
As a rule of thumb for busy urban centers, maybe once every 1-6 months would be desirable? It also depends on the use case of course - for example businesses may change more frequently than new sidewalks being built, etc.
I think you are on the right track, @boris. Having a “solid reason”, “important reason”, or “valid reason” (a term used in some jurisdictions for rationally justifying something, sometimes an exception from a rule) to capture an area is a sane solution because otherwise it is really difficult to come up with a rationally determined number. This way you do not run the risk of being alleged of surveillance either. Re‑captures should happen as few times as possible but definitely when needed. Hence, volatile areas are going to get captured more often than stable areas. Technological advancements are surely also a valid reason for re‑captures. However, like every leeway rule, every such rule can be abused by overusing it. So, I think it should be best to think of it as keeping good capturing etiquette.
Thanks, often capturing can happen during daily routine, thus it would have to adapt to such routes.
I’ll take it as one month being the lowest desirable (unless there are known or suspected changes).
