New generation of Formula 1 2022 and beyond car will remain utilize the old roll-hoop FOM onboard camera

[dans79]

You have to upload all that data in real time though, from several cameras on each car. FOM said that the helmet camera replaced one of the other onboard cameras for bandwidth reasons, bandwidth was the reason for the relatively low res also.

The cameras could easily be 360 degree, 4k, 50/60 fps. cameras that record & transmit in IPB h.265. That can be decoded and distortion corrected for in the broadcast truck before being broadcast.

This is a $430 USD devise thats does all of the above except 4k 60fps.
https://store.insta360.com/us/product/one_x2

360:
5.7K@30fps, 25fps, 24fps
4K@50fps, 30fps
3K@100fps

The fact that the onboards suck so much, is purely down to FOM being lazy, cheap, dumb, or some combination of the 3!

Everything above is old consumer grade tech, if they went directly to a manufacture they could do far better.

[nzjrs]

The fact that the onboards suck so much, is purely down to FOM being lazy, cheap, dumb, or some combination of the 3!

Everything above is old consumer grade tech, if they went directly to a manufacture they could do far better.

It's not so Bad.

[mzso]

Indeed. But if anyone can afford more bandwidth it's F1. From all those damned billions, they could affor a few million for extra wireless antennas.

My phone was one of those old 2G Nokia brick phones and I wasnt happy with the bandwidth so I went and bought 17 or 18 antennas and stuck them all over the case. It's at 7G speed right now and supplies internet to the neighborhood schools.

You fail at mocking. More antennas can mean more bandwidth in multiple ways. Apart from the obvious: better contact.
If you keep the range low you can use a much larger chunk of the available badwidth, compared to broadcasting through the whole circuit.
More antennas can also mean a broader frequency range.

[tummalakoushik]

I think less stabilization and higher FOV would make a lot of impact and generate great sense of speed in the footage. Even the onboards from early to mid 2000s give better speed sensation.

[mzso]

I think less stabilization and higher FOV would make a lot of impact and generate great sense of speed in the footage. Even the onboards from early to mid 2000s give better speed sensation.

Is there even stabilization. I think the cameras are just mounted very rigidly.
In contrast the helmet cameras are mounted by driver's necks so they're unpleasantly shaky, and should have a good amount of stabilization to provide a pleasant viewing experience.

As for a sensation of speed, a wider FOV would definitely help. But the lack of sensation I think comes from mainly from the cars being very long and heavy hulks, so they don't fidget around much.

[wowgr8]

In this day and age , bit sad to not use the chance to try new things

I can't see why couldn't they experiment with things regardless of large regulation changes.
To me its even sadder that random GoPro footages on youtube look better than F1 onboard cameras...
I want proper UHD footage.

What I would also like to have is between eye cameras. The sort that Grosjean/Gasly trialed, and similar to what Alonso had in Belgium. But with proper image stabilization. And mounted to the helmet instead of a glasses frame which no-one would tolerate long.

The roll hoop cam onboards can be a lot better and of higher quality (like they were in 2017 and 2018) but they lowered the bitrate dramatically in 2019 for smooth F1TV streaming that's the reason the onboards look so bad these days

[mzso]

The roll hoop cam onboards can be a lot better and of higher quality (like they were in 2017 and 2018) but they lowered the bitrate dramatically in 2019 for smooth F1TV streaming that's the reason the onboards look so bad these days

There's nor reason to reduce the bitrate the camera transmits to the receiver. They can do that when as they are broadcasting it.

[Juzh]

In this day and age , bit sad to not use the chance to try new things

I can't see why couldn't they experiment with things regardless of large regulation changes.
To me its even sadder that random GoPro footages on youtube look better than F1 onboard cameras...
I want proper UHD footage.

What I would also like to have is between eye cameras. The sort that Grosjean/Gasly trialed, and similar to what Alonso had in Belgium. But with proper image stabilization. And mounted to the helmet instead of a glasses frame which no-one would tolerate long.

You have to upload all that data in real time though, from several cameras on each car. FOM said that the helmet camera replaced one of the other onboard cameras for bandwidth reasons, bandwidth was the reason for the relatively low res also.

There's enough bandwidth for each car to transfer one camera feed in 1080p50 at a pretty solid bitrate, probably 6+ mbs. Exact number is unknown, as access to source footage is probably impossible unless you work for FOM. Helmet cam's image quality wasn't limited by bandwidth as it was using same broadcast method as all other onboard cameras, it's probably best it can do in such small form factor. Biggest problem wasn't resolution (it was clearly 1080p as you could easily red everything from steering wheel), but rather light distortion from helmet lens and lack of any stabilisation. It'll get better with time.

As a side note, all cameras do record to internal storage which can later be downloaded from the car. Latest such example was Verstappen's forward facing cam from brazil, Stroll's pole lap from turkey in 2020 (broadcasted footage suffered from heavy corruption due to poor signal and if I remember correctly someone in FOM control room was even switching between various cameras in the middle of pole lap), bottas' crash in melbourne 2018...

Obviously go-pros can capture at UHD because they don't do live broadcasts, night and day difference. Another thing that makes all those go pro captured runs look faster despite being 10 or 20 or 30 second slower is because they use massively higher FOV, something that FOM really should implement ages ago.

This video illustrates perfectly how FOV can influence perception of speed:

[Juzh]

In this day and age , bit sad to not use the chance to try new things

I can't see why couldn't they experiment with things regardless of large regulation changes.
To me its even sadder that random GoPro footages on youtube look better than F1 onboard cameras...
I want proper UHD footage.

What I would also like to have is between eye cameras. The sort that Grosjean/Gasly trialed, and similar to what Alonso had in Belgium. But with proper image stabilization. And mounted to the helmet instead of a glasses frame which no-one would tolerate long.

The roll hoop cam onboards can be a lot better and of higher quality (like they were in 2017 and 2018) but they lowered the bitrate dramatically in 2019 for smooth F1TV streaming that's the reason the onboards look so bad these days

Only F1TV streams were downgraded, starting in brazil 2018 because it was stuttering like mad, why that was i don't know. For 2021 f1tv went from 1080p25 @ 5500 kbs down to 720p50 @ 3500 kbs. World feed onboards and everything that isn't from F1TV (so everything from official broadcasters) stayed the same as it was (so 1080p50 at 6 or more mbs). They made some changes to camera lens in 2020 which in my opinion made it worse, but not from an image quality standpoint but rather something about colour balance and brightness/contrast seem off since then.

[Hoffman900]

They really do need to fix FOV (focal length).

The hoop cams in the early 2000s had a greater FOV / slight fish eye effect, and it looks much faster (even on mute).

[mzso]

There's enough bandwidth for each car to transfer one camera feed in 1080p50 at a pretty solid bitrate, probably 6+ mbs. Exact number is unknown, as access to source footage is probably impossible unless you work for FOM.

There's only enough bandwidth because they built infrastructure for only that much. They could brute force high bitrate UHD streams for all cameras on all cars. F1 has more than enough money for such endeavors.
Also, it seems like Mercedes can establish a ~2gbit/s connection. So maybe that's the sort of direction they should be looking at.

Helmet cam's image quality wasn't limited by bandwidth as it was using same broadcast method as all other onboard cameras, it's probably best it can do in such small form factor.

I highly doubt that. Even mobile phone cameras can do better, with actual image stabilization.

Another thing that makes all those go pro captured runs look faster despite being 10 or 20 or 30 second slower is because they use massively higher FOV, something that FOM really should implement ages ago.

One of their many failures with footage.

By the way can't someone fix the topic title. It makes my back shiver any time I see it, and not in a good way.

[piast9]

I have doubts whether these have to do with focal length.

Me too. It seems they have used longer lens with less field of view in old times. But if you use sport camera to film your mountain bike rides the wider the angle the faster the ride looks. I think its more complex. You have the sensation of speed because of no (or worse) stabilization and the grip of the car. Stabilized image from contemporary cameras looks like the cars were on rails not the edge what is humanly possible.
Edit: as for the MTB videos - the sensation depends hugely on the placement of the camera - if mounted on the top of the helmet the footage looks much slower than if the camera was on the chest straps.

[nzjrs]

Helmet cam's image quality wasn't limited by bandwidth as it was using same broadcast method as all other onboard cameras, it's probably best it can do in such small form factor.

I highly doubt that. Even mobile phone cameras can do better, with actual image stabilization.

Some thoughts from my experience building real-time camera/imaging systems (neither refuting nor supporting the points above).

The video I posted above about the 2018 season is probably built with 2017 era cameras. That might sound old but most sensor manufacturers do a yearly or 18 month refresh of sensors in the HD to 2k range. The throw cameras on at the high end of the range but even now that is slowing down because multiple-sensors+lenses is the new consumer battleground. The sensor is probably good enough.

Most of what we are seeing in mobile phone image improvements in the last 3 years are AI driven. Maybe 1/3 of the AI bling in the last years are enabled by the pretense of multiple lenses or sensors - simultaneous imaging lets one do super-resolution, focus blending, blurring, (better) HDR, etc etc.

None of that is feasible however to do in real-time at the moment.

There would be some value in a partnership with Huawei or some mobile imaging / camera partner as 2022 era camera SOCs could run in real-time some AI upsampling/cleaning in on 2017 era camera resolution cameras. They could engineer something themselves. But quite a lot would need to be on-device which would end up being itself a compromise that would be interesting to design.

There are also optical and form-factor constraints in the roll-hoop enclosure that are going to be different to phones. For example, the motion will necessitate larger pixel sensors (of which their are fewer) in order to keep short exposure times to have some control over motion blur. This affects compatible lenses and what lenses are available. Meanwhile on the mobile phone they just fake this in software - possible 'generally' only because they are not doing it in realtime.

People should not underestimate how many hundreds of thousands (millions?) of man hours of engineering have gone into mobile phone imaging tech.

[Zynerji]

Helmet cam's image quality wasn't limited by bandwidth as it was using same broadcast method as all other onboard cameras, it's probably best it can do in such small form factor.

I highly doubt that. Even mobile phone cameras can do better, with actual image stabilization.

Some thoughts from my experience building real-time camera/imaging systems (neither refuting nor supporting the points above).

The video I posted above about the 2018 season is probably built with 2017 era cameras. That might sound old but most sensor manufacturers do a yearly or 18 month refresh of sensors in the HD to 2k range. The throw cameras on at the high end of the range but even now that is slowing down because multiple-sensors+lenses is the new consumer battleground. The sensor is probably good enough.

Most of what we are seeing in mobile phone image improvements in the last 3 years are AI driven. Maybe 1/3 of the AI bling in the last years are enabled by the pretense of multiple lenses or sensors - simultaneous imaging lets one do super-resolution, focus blending, blurring, (better) HDR, etc etc.

None of that is feasible however to do in real-time at the moment.

There would be some value in a partnership with Huawei or some mobile imaging / camera partner as 2022 era camera SOCs could run in real-time some AI upsampling/cleaning in on 2017 era camera resolution cameras. They could engineer something themselves. But quite a lot would need to be on-device which would end up being itself a compromise that would be interesting to design.

There are also optical and form-factor constraints in the roll-hoop enclosure that are going to be different to phones. For example, the motion will necessitate larger pixel sensors (of which their are fewer) in order to keep short exposure times to have some control over motion blur. This affects compatible lenses and what lenses are available. Meanwhile on the mobile phone they just fake this in software - possible 'generally' only because they are not doing it in realtime.

People should not underestimate how many hundreds of thousands (millions?) of man hours of engineering have gone into mobile phone imaging tech.

Having a 17 year special relationship with Samsung and still receiving engineering samples, I feel that I can attest to your final paragraph.

But the belief that these things cannot be done in real time in 2022 is patently false.

The processing power available in a top end smart phone is slightly larger than a stick of gum, and easily parallelizable. You could literally have a supercluster the size of a cigarette pack if you wanted.

Apple, Samsung, nVidia, Huawei, Qualcomm could all provide military grade hardware for these cameras. It literally just a lack of leadership at FOM at this point. "Perfect Archiving" should be a pursuit for the powers-that-be, and high-tech on-board imaging should be the simplest path forward.

[nzjrs]

But the belief that these things cannot be done in real time in 2022 is patently false.

Haha, you tell me exactly what you want done and I'll tell you if it can be done in real time. I won't even charge you for the advice .

I think you will be surprised

(very few conv. nn / DL image processing operations are able to be spatially parallelized and doing round-robin parallelization just pushes the latency such most people, myself included, don't define it as real-time any more. there are trade offs here but to say 'patently false' is laughably patently false and shows a misunderstanding of how current AI juiced imaging is implemented and the trade offs involved)