-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
AP2.0 Cameras: Capabilities and Limitations?
- Thread starter lunitiks
- Start date
Does this apply for premium audio too? You see I don't have that upgrade but still I found (and made use of) the wiring for subwoofer. Same goes for power liftgate, extra interior lighting +++: All the wiring is there even if you didn't purchase the upgrade. Some exeptions of course, like for example the wiring for the premium audio amp., I dont think that's there.
Would be really interesting to know if wiring for corner radars are in place
I don't have premium audio, so I can report back after China delivers my cable!
(While we're all waiting for some more interesting info from ourselves, I thought we could highlight some of AP1's capabilities and limitations for a moment.)
@JeffK and @scottf200 managed to find Mobileye's EyeQ3 data sheet. A quick read left me with the impression that the chip should be quite capable in terms of video input processing.
And I quote:
"Video Input Interfaces
The video interface block supports the capture of three high dynamic-range (90-120dB), color or monochrome image video streams, each with a maximum of 8k x 8k image resolution (i.e., components). The video interface is always in slave mode (i.e.,sync signals are received from image sensor).
The EyeQ3® provides a glueless interface to the image sensor, which it configures via an I2C or GPIO interface.
Typical image sensors include Aptina’s 12-bit MT9M025 devices and future 16-bit image sensors. The video interface, upon receiving the video stream, performs pixel pre-processing and then stores the modified 8/16bit pixel images.
The video interface features include:
• Three parallel video-in ports
• Programmable capture frame size up to input size (max: 8k x 8k components)
• Support for both progressive and interlaced scans (de-interlacing done in DMA).
• Synchronization support using data bus according to ITU-R BT.656.
• Five channels for the three video-in ports (see Figure 2: Video Channels below), with separated Filter, γ and Histogram for each channel.
o Sub sampling for horizontal and vertical configurable separately (step size 0-31)
o 16bit γ curve approximation. One γ curve for each channel (with bypass option)
o Histogram supports different weights for different areas of the frame; as well as supporting windowing (horizontal and vertical offset and size) and sub-sampling (step size 0-31 in either horizontal, vertical or both)
o Multiple pixel arrangements - bursts from FIFOs may reverse pixel order
o Header and footer from image sensor can be captured without going through sub-sampling and γ correction process.
• Max pclk frequency:
o 100MHz when Histogram is enabled.
o 200MHz when Histogram is disabled.
• Max mclk frequency – 250MHz/n where n>1."
As for video output:
"Video Output Interface
The video output interface supports the display of one high dynamic-range 16-bit color or monochrome image video stream, with a maximum of 8k x 8k image resolution (components). The video output interface supports Master Mode (i.e.,driving syncs to image sensor).
The essential features of the output interface are:
• Supports RGB (5-6-5, 16 bit), Y:Cb:Cr 4:2:2 (8bit/component; 8 or 16 bit/pclk)
• Supports format change from RGB to Y:Cb:Cr 4:2:2
• Output frame size – up to 8k x 8k
• LUT for data generation
• 64 bit Internal memory word width
o 16bit components stored consecutive ly
o 8bit components stored as 3 consecutive components per 32bit (padded).
• Max pclk frequency – 250MHz/n where n>1."
Crappy gif-illustration:
@JeffK and @scottf200 managed to find Mobileye's EyeQ3 data sheet. A quick read left me with the impression that the chip should be quite capable in terms of video input processing.
And I quote:
"Video Input Interfaces
The video interface block supports the capture of three high dynamic-range (90-120dB), color or monochrome image video streams, each with a maximum of 8k x 8k image resolution (i.e., components). The video interface is always in slave mode (i.e.,sync signals are received from image sensor).
The EyeQ3® provides a glueless interface to the image sensor, which it configures via an I2C or GPIO interface.
Typical image sensors include Aptina’s 12-bit MT9M025 devices and future 16-bit image sensors. The video interface, upon receiving the video stream, performs pixel pre-processing and then stores the modified 8/16bit pixel images.
The video interface features include:
• Three parallel video-in ports
• Programmable capture frame size up to input size (max: 8k x 8k components)
• Support for both progressive and interlaced scans (de-interlacing done in DMA).
• Synchronization support using data bus according to ITU-R BT.656.
• Five channels for the three video-in ports (see Figure 2: Video Channels below), with separated Filter, γ and Histogram for each channel.
o Sub sampling for horizontal and vertical configurable separately (step size 0-31)
o 16bit γ curve approximation. One γ curve for each channel (with bypass option)
o Histogram supports different weights for different areas of the frame; as well as supporting windowing (horizontal and vertical offset and size) and sub-sampling (step size 0-31 in either horizontal, vertical or both)
o Multiple pixel arrangements - bursts from FIFOs may reverse pixel order
o Header and footer from image sensor can be captured without going through sub-sampling and γ correction process.
• Max pclk frequency:
o 100MHz when Histogram is enabled.
o 200MHz when Histogram is disabled.
• Max mclk frequency – 250MHz/n where n>1."
As for video output:
"Video Output Interface
The video output interface supports the display of one high dynamic-range 16-bit color or monochrome image video stream, with a maximum of 8k x 8k image resolution (components). The video output interface supports Master Mode (i.e.,driving syncs to image sensor).
The essential features of the output interface are:
• Supports RGB (5-6-5, 16 bit), Y:Cb:Cr 4:2:2 (8bit/component; 8 or 16 bit/pclk)
• Supports format change from RGB to Y:Cb:Cr 4:2:2
• Output frame size – up to 8k x 8k
• LUT for data generation
• 64 bit Internal memory word width
o 16bit components stored consecutive ly
o 8bit components stored as 3 consecutive components per 32bit (padded).
• Max pclk frequency – 250MHz/n where n>1."
Crappy gif-illustration:
Last edited:
BTW, the BUC will pulsate when the turn signal is on. The dynamic range isn't so awesome so smoothly deal with the extra lights coming near the sensor. I'm not sure if that affects anything but this did not happen with my LEAF back up camera (which wasn't anywhere near as detailed or capable but maybe that was why it wasn't affected). Also the Model S blinkers are insanely bright.
LOL look at these pictures and CAD-drawings of the Model 3.
Someone please tell me what that cut-out in the rear quarter aluminium is for...
Do you remember these S and X mules?
We already have hard evidence of (planned) rear corner radars in wiring diagrams, software etc. for Model S.
Does the model 3 come with rear corner radars?
@FredLambert
Last edited:
Good find @lunitiks... It could, though, also be Tesla just leaving room for the eventuality - of leftovers from a time when Model 3 was being designed and Tesla considered rear radars... (e.g. AP "1.5" era - Is Model X (or was?) getting an "AP 1.5" before full-autonomous "AP 2.0" suite upgrade?)
Thanks @J1mbo that was exactly the reply I was waiting for. Model S has those two vents as well.
So it might not be anything, then
*Relief*
So it might not be anything, then
*Relief*
Thanks @J1mbo that was exactly the reply I was waiting for. Model S has those two vents as well.
So it might not be anything, then
*Relief*
I checked, apparently Audi puts the radars into the bumpers - this is an after market product in the picture, but the radar slot in the bumper is OEM:
Corner radar has been in the settings since 2014.
strangecosmos
Non-Member
I found a scientific paper with some surprising and impressive results on the accuracy of a multi-camera system (like the one in Hardware 2) in a self-driving car application. The multi-camera system was accurate to under 10 cm, but it was only tested at low speeds.
Based on consulting knowledgeable people, I think adapting a multi-camera system to high driving speeds is just a matter of ensuring that your cameras have good enough specs. But I don’t know a) what specs are good enough and b) whether Hardware 2’s cameras have them.
I don’t have the technical knowledge to figure this out myself, so I was hoping y’all could help. I feel like we’re within spitting distance of hard, quantitative evidence that a multi-camera system is sufficient for self-driving and lidar isn’t necessary. Or the opposite! Uh oh!
Based on consulting knowledgeable people, I think adapting a multi-camera system to high driving speeds is just a matter of ensuring that your cameras have good enough specs. But I don’t know a) what specs are good enough and b) whether Hardware 2’s cameras have them.
I don’t have the technical knowledge to figure this out myself, so I was hoping y’all could help. I feel like we’re within spitting distance of hard, quantitative evidence that a multi-camera system is sufficient for self-driving and lidar isn’t necessary. Or the opposite! Uh oh!
@Trent Eady's been asking in other threads what camera models AP2.0 use, i.e. not just which sensors.
While the exact model numbers are not known, I've seen traces of vendor info in SW:
To find the exact models, I guess we'll need some teardowns of the cameras themselves (unlikely to happen), or someone who's better at searching the interwebs than me.
Bear in mind that vendors can change. But I guess that would come to show in the parts manual with an updated part no. (Since Oct '16, I've only seen one camera revision: The side repeaters, with some strange comment/reference to South Korea.)
While the exact model numbers are not known, I've seen traces of vendor info in SW:
- Main, narrow, fisheye, side-repeaters: "Largan"
- B-pillar: "Asia Optics"
- BUC: "SEMCO"
To find the exact models, I guess we'll need some teardowns of the cameras themselves (unlikely to happen), or someone who's better at searching the interwebs than me.
Bear in mind that vendors can change. But I guess that would come to show in the parts manual with an updated part no. (Since Oct '16, I've only seen one camera revision: The side repeaters, with some strange comment/reference to South Korea.)
strangecosmos
Non-Member
Actually, I don't know the difference! All I needed to know were these specs. Is the fisheye camera the same sensor as all the rest, and just a different lens? Or what?
All cameras use the same type sensor except the rear view cam. So the difference is mainly lens / size, yes
Yes, I remember South Korea note saying that only from certain revisions onwards can be used there.
I guess we assumed at the time it had something to do with camera features and quality needed for the market, but I wonder if it is about side marker differences instead. Maybe Korean law has some specific requirements for side markers and thus a certain camera part version only fits or something like that.
Similar threads
- Article
