canklecat

The main limitations to recording license plates are:

1. U.S. license plates are tiny compared with most European plates.
2. GoPro and similar cameras were designed for action, not traffic documentation. The ultra-wide angle lenses and fixed focus are geared toward closeup action, not recording fine detail at a distance.
3. Relatively fewer effective pixels, with details that are already rendered tiny by the ultra-wide angle lenses.
4. Slow shutter speeds in anything but bright light.
5. Increased noise at higher ISOs when the camera is capable of defaulting to faster shutter speeds in dim light.
6. Poor dynamic range compared with larger sensor still/video cameras. The highly reflective white plates favored by some states will blow out to blank white at night when most action cams default to greater sensitivity to compensate for overall darkness.

Regarding 1, there's nothing we can do about tiny license plates. I'm surprised law enforcement and the government haven't demanded larger plates to enhance surveillance but it hasn't happened and probably won't anytime soon.

2. The typical action cam has a field of view roughly comparable to a 10mm to 16mm lens on a 35mm film or full frame digital camera. An FOV/angle of view closer to 90 degrees diagonally, comparable to a 20mm-28mm wide angle lens, will better capture license plate detail. I've tested this on my digital cameras with zooms that record video. Even my older Ricoh digicam that records only 640x480 video (lens set to 24mm-28mm equivalent) captures license plate details better than my Ion Speed Pro. My Nikon V1 recorded excellent detail in 1080 (30fps) and 720 (60fps) at the 28mm equivalent focal length.

3. My Ion Speed Pro is billed as a 14mp camera but is actually closer to 5mp. It struggles to reliably record license plate detail, even in daylight if the vehicle is nearby but at an angle. Once the vehicle has passed and presents a more favorable angle, it's too far away to record license plate detail perfectly -- there's usually some guesswork and studying of various frames to identify some letters and numbers.

4. Shutter speeds are always among the main limitations in using video to capture fine detail from individual frames in anything less than ideal light. Most action cams default to around 1/30th-1/60th sec or slower in dim light, guaranteeing license plates will be blurred. Frame rates only increase the chances of capturing a frame without blur, but by itself frame rate doesn't solve the shutter speed problem.

5 & 6. Action cameras still use teensy sensors, compared with most good compact mirrorless still cameras and dSLRs. Tiny sensors are better than ever now, but still very limited in high ISO performance and dynamic range. As light dims and ISO is increased, noise increases and dynamic range decreases. Noise can be compensated with noise reduction but this tends to smooth out fine detail, rendering license plates, hair, foliage, etc., clumpy and indistinct. Dynamic range reduces the ability to capture highlights and shadows. It can be compensated to some extent with exposure biasing, but usually affordable action cams are badly fooled by nighttime and low light conditions and render shadows and bright highlights as featureless blacks and whites.

A better traffic documentation camera would have a 1-inch sensor, which is significantly better than the usual teensy sensors found in action cams and most smartphones. It would have a lens equivalent to no wider than 18mm on a 35mm film/full frame digital, and preferably closer to 20-24mm. The larger sensor would offer better high ISO and dynamic range performance, and enable faster shutter speeds in most conditions -- but not all. There will still be limitations even with higher frame rates. And a true 10mp would be the minimum (found in the original Nikon 1 Series J-1 and V-1 Aptina sensors), although the Sony 1-inch sensors can handle 20mp with good resolution (higher mp isn't always better if it compromises resolving fine detail).

Lenses with less extreme wide angles would require more precise aiming but that's not a huge challenge -- I found it relatively easy on my digicams with rear LCD screens. A little testing or use of wireless connectivity to a smartphone or tablet would help with aiming affordable traffic documentation cameras that lack rear screens.

ItsJustMe

You forgot bitrate. You need probably at least 45mbps bitrate to reliably capture plates of cars travelling at high speed, no matter what the resolution and sensor size is. This isn't really a problem now that you can buy good quality 128GB cards for < $40 - you can still record hours of video at 50Mbps bitrate on these cards, but there's so little demand for this that it's unlikely anyone will build this.

blue192

The great thing about the helmet cameras is that they usually have a microphone, so you as the rider can read out the license plate incase the camera does not read this correctly. At least that is what I do.

wphamilton

Possibly, many people have cheaper cameras that implement anti-shake in the software as opposed to the more effective hardware implementation.

canklecat

That reminds me of one of the main drawbacks to the Ion Speed Pro when mounted on my helmet. The mic located on the bottom of the cylindrical shaped camera pics up horrendous wind noise. Turns out my helmet vents create a lot of turbulence, even at modest speed. Using a wind screen over the mic doesn't help because it's accurately picking up the sound of turbulence created by the helmet. So unless I tilt my head way up or down the wind noise drowns out my voice.

It picks up my voice better when mounted on the handlebar, but I use that option only on rides longer than 90 minutes so I can hook the camera up to an external battery (in a bag on the handlebar).

ItsJustMe

I think mounting a camera with optical image stabilization like that would kill the camera in minutes for me. That kind of vibration on an optical device that moves would just destroy the thing, and even if not, would probably make the shake worse than nothing since the voice coils in there would not be able to overcome the G forces from the vibration.

wphamilton

I'm not sure that I follow you. The hardware implementation of anti-shake is solid-state signal processing, not mounting.

gbru316

This would probably work well. Mount it to your helmet and you only need 1 camera.

KeyMission 360 Camera: 360 Degree Videos & Photos | Nikon

01 CAt Man Do

Going forward I think what we ( the cycling public ) need is a compact "license plate reader" coupled with a video recorder. Once someone begins to market something like that and make it small enough, I figure there should be a way ( to make it small ) using BT to work with your smart phone.

To those who don't know, license plate readers are those things you sometimes see mounted on Police cars. They basically work on the same principles as facial recognition technology but in this case only for numbers and letters. There are even companies that employ these on private vehicles and are contracted with the DMV in order to recover vehicles that have loans that are delinquent. Of course we wouldn't need to access the plate info unless in an accident so that shouldn't be a problem. Just give the number(s) to the Police and let them do their work.

ItsJustMe

OK, done differently in action cams than in proper cameras then (or indeed in my Nexus 5). Those have optical image stabilization, which uses a meniscus lens mounted in an X-Y controlled voice coil mount to counter camera shake. That's the TRUE really good anti-shake technology. Relying on postprocessing isn't as good, though if it's done in something like an FPGA or other dedicated circuitry it's probably going to be better than firmware control. Though of course anything that relies on just jiggering the data off the sensor around can largely be replicated in computer post-processing. Not quite of course since it's been through a lossy compression layer first.

canklecat

Eventually 360 video cameras will be the standard. But the combination of adequate tech and affordability isn't quite there yet. I've watched several Ricoh Theta sample videos and the resolution isn't quite good enough. Nikon is better but it'll also require some easy to use software that doesn't demand huge processing power to isolate specific segments of a 4k video for documenting traffic problems.

nuclear_biker

DCR had some footage from the Garmin VIRB 360 camera and I couldn't quite make out plates. It's coming, though.

FlashBazbo

Please excuse my rewind to the original post . . .

What are your suggestions for mounting rear-facing cameras where you have an under-the-saddle bag already mounted? I've tried the kind that mount on saddle rails, but they don't seem to work for me even without a saddle bag. To complicate things more, I ride a BMC TeamMachine SLR01 that has a trapezoidal shaped seat post rather than round. Any ideas?

nuclear_biker

What exactly didn't work? Pictures? I use the GoPro mount on my saddles and it works well, I could see it being trouble if your seat is mounted pretty far forward I suppose, but with some extenders it should be possible in almost any configuration.

I was actually thinking about coming back to this post since I recently started taking some footage to test the feasibility of recording a license plate with a GoPro (or in my case a VIRB Ultra 30, although they're very similar in terms of the sensor/capabilities). A preliminary outing suggests at 1080p30 it is in fact doable, I can record most plates of cars passing me (even in 55 mph zones where cars are likely going 65 or more). It's much clearer when the cars are driving into the sun then away from the sun, but I think it would be enough with the partial plate and make/model/color.

1nterceptor

Depending on the cam; fork mounted facing backwards. Polaroid CUBE with strips from an old tube:

Contour by the rear triangle facing forward; but could also face backwards I suppose:
Rear triangle mounted by 1nterceptor, on Flickr

FlashBazbo

I tried one of the fantaseal GoPro clone saddle rail mounts. With oval carbon rails, the fit was not good. Too tight and the edges would have galled the carbon. Too loose and it would likely have fallen off with vibration. (No, I didn't take pictures. What would be the point?) Round bar/tube mounts (Garmin, GoPro, cheapo, etc.) don't work on the trapezoidal seatpost. And the BMC doesn't have an open triangle above the rear brake, so there is no access to tiny tubes.

And I can't imagine rigging up that inner tube setup! Wow.

nuclear_biker

Oh yeah I'd be hard pressed to clamp on carbon rails myself. Garmin makes some quarter turn mounts that work on aero seatposts (for Varia lights) and from there you could in theory mount a GoPro style attachment. Might be pushing the weight limits. Skewer mounts (shown above posts) might also work.