Discuss: How to Retrofit a Front Bumper Camera on a HW4 Model S and Model X

[Not a Tesla App]

A overview of retrofitting a front bumper camera on HW4 Model S/X, detailing parts, costs, and the DIY process.… Read More

How to Retrofit a Front Bumper Camera on a HW4 Model S and Model X

A overview of retrofitting a front bumper camera on HW4 Model S/X, detailing parts, costs, and the DIY process.

www.notateslaapp.com

 

[PrescottAZRichard]

SO here's my musing of the front bumper camera. (You're welcome ahead of time). Both this camera and the back up camera seem to be designed to show what is directly on the ground in front / behind the car with a fish eye lenses. So the cars and other objects that are at the edges of these lenses are VERY distorted and are a much smaller amount of actual pixels. I suppose a computer can literally calculate and adjust for that distortion to interpret the camera data. The stuff that people WANT the camera to see from the front bumper (for FSD) is usually traffic approaching from the sides, and that's just very distorted and a small amount of the sensor. That computer is gonna have to be REALLY GOOD at interpreting the edges of these frames.
The distortion is very noticeable to our brains when you watch a rear end collision video. When the approaching car is far back the approach seems slow, but as it gets closer the fish eye effect looks like it actually speeds up as it nears the center of the camera frame. This is where a computer interpreting info has that understanding built in. Same thing for cross traffic.
Hope I explained that well. One camera for 3 purposes isn't gonna give data that is as good as 3 cameras designed for more singular purposes. One facing straight down, and two in the front bumper facing left and right for the sole purpose of seeing oncoming traffic from the sides. The thing is there you have more expense, more data to sift through, etc. Everything has trade offs.
I could be missing something of course.

 

[Procal]

SO here's my musing of the front bumper camera. (You're welcome ahead of time). Both this camera and the back up camera seem to be designed to show what is directly on the ground in front / behind the car with a fish eye lenses. So the cars and other objects that are at the edges of these lenses are VERY distorted and are a much smaller amount of actual pixels. I suppose a computer can literally calculate and adjust for that distortion to interpret the camera data. The stuff that people WANT the camera to see from the front bumper (for FSD) is usually traffic approaching from the sides, and that's just very distorted and a small amount of the sensor. That computer is gonna have to be REALLY GOOD at interpreting the edges of these frames.
The distortion is very noticeable to our brains when you watch a rear end collision video. When the approaching car is far back the approach seems slow, but as it gets closer the fish eye effect looks like it actually speeds up as it nears the center of the camera frame. This is where a computer interpreting info has that understanding built in. Same thing for cross traffic.
Hope I explained that well. One camera for 3 purposes isn't gonna give data that is as good as 3 cameras designed for more singular purposes. One facing straight down, and two in the front bumper facing left and right for the sole purpose of seeing oncoming traffic from the sides. The thing is there you have more expense, more data to sift through, etc. Everything has trade offs.
I could be missing something of course.

I think that as long as the camera can detect motion as something approaches, then it should not matter whether it can tell what it actually is. In other words, similar to how a smart doorbell camera works (not exactly but in a general sense), if it detects motion as something is approaching (i.e., getting larger), then that should be enough information to tell the computer “hey something is moving, we should wait a few seconds and see if it is getting closer”. Obviously this would happen in the order of milliseconds, rather than full seconds, but you get the idea. I don't know if that is how that would work, but I can't see how else they would use that data, other than for better parking lot behavior. Either way, I don't think they are currently using that camera for FSD at all, and it might be a while before they do. It would likely require them to retrain the model with the new data from this camera before it can actually be used for anything useful.

 

[K.I.T.T.]

So disappointing reading that Tesla "laid the foundation" all the time for a bumper camera and simply did not deliver it due to stubborness. But hey, the priority was sinking the company with the CyberTruck abomination.

Wouldn't be surprised if this retrofit could be replicated on M3/MY.

Distorted or not, you need it only for parking, so as long as you can see something, even if it was upside down and b&w, that's all it's needed. Surely the software can improve the recognition of obstacles in the 3D rendering, which is widely unreliable in the front section, without a bumper camera.