[OSM-talk] Extracting data from traffic sign recognition aboard modern cars ?

[Philip Barnes]

On Sat, 2013-04-20 at 11:37 +0200, Jean-Marc Liotier wrote:
> On 04/20/2013 09:20 AM, NopMap wrote:
> > Paul Johnson-3 wrote
> >> Sounds like all the more reason to try this.
> > Not a good idea.
> >
> > If you break anything trying to hack into the driver assistance functions,
> > you might cause an accident, either directly by malfunctions of the driver
> > assistance or indirectly by distracting the driver with unexpected results.
> >
> > By design, fiddling with those devices is likely to trigger the theft
> > protection, disabling them completely and voiding the warranty of the car.
> 
> Passively sniffing the CAN bus for sign recognition events would be
> entirely undetected - but that would require reverse engineering the
> protocol and writing specific code... 
I am wondering how good this technology is. Does it cope with implied
speed limits such as a change from single to dual-carriageway, motorway
chopsticks sign, the presence of street lights, a placename, a placename
with a line through it. Does it have a GPS to know if the limit is mph
of kph, or what country it is in.

If it is using a GPS and stored mapping then there is the danger of
polluting OSM with proprietary data.

There is also the danger of picking up temporary limits, such as
roadworks, which should not be imported into OSM. 


> One might argue that rigging a
> dashcam and processing its output on a desktop with some road sign
> recognition software (generic pattern recognition neural net trained
> with road signs ?) would be less effort - albeit an inelegant effort
> duplication.

I have used a dash cam to enable me to tag speed limits, no automation,
just fast step through when I get home. One other advantage is it can be
used to extract a lot more information, crossings, traffic lights, shop
names and so on.

Phil (trigpoint)