I want to design a computer vision device that will act as a "driving assistant". The device will improve driving safety by acoustically or visually alerting the driver of certain hazardous situations (lane crossing, safe braking distance etc.) but will not be directly connected to any car controls. My question is: how to design my device so that it complies with regulations if I want to sell it as a "car accessory" in aftermarket in the US and EU?
By aftermarket I mean that the device will not be mounted into a vehicle by car manufacturer but by the end user/customer. The mounting should be as simple as placing a camera on the windscreen (out of the driver's field of view), connecting it to the "processing unit" by a cable, placing that unit somewhere (under dashboard, seat etc.) and plugging the unit into the 12V outlet (or rather wiring directly to 12V).
I am a software developer and I have already a prototype running on a x86 platform. The prototype consists of a common mini-ITX board, an USB camera and a picoPSU power source. All these have an "FCC Class B Part 15" certification (I guess this is standard for consumer-grade electronics intended for home/office use). Besides these I will also use an USB 2.0 or 3.0 cable to connect the camera to board and the board will wire to those 12V.
The device will be only an "assistant" and will not carry any life-critical function when driving. I think it may fall in the same category as car GPS navigation devices (Garmin etc.). I know that demands for a mission-critical automotive electronics (things like engine electronics, brake systems etc.) are very high, but as I said my application is not anywhere near this kind of usage.
The best situation for me of course would be no need for re-certification since I only bring/bundle already certified components together (besides the cabling) into a new product without altering any of them. In case the x86 setup could not get car-compliant my second attempt would be finding a powerful automotive-compliant embedded board (nVidia Jetson TK1 or Jetson Pro?) that I could compile my computer vision code on. I would like to avoid doing an FPGA-based design because rewriting my code to hardware is likely to be very costly.