Subaru Outback Forums banner

1 - 6 of 6 Posts

·
Registered
Joined
·
90 Posts
Discussion Starter #1
Does anyone know the actual technical specifications of the eyesight cameras, such as resolution, frame rate, bit rate, data recording, color response / number of color channels ("color" includes infrared)?

Also, is there some obvious reason why the Eyesight system would not be suitable for a Subaru option to allow it to also be used as a recording dashcam?
 

·
Registered
Joined
·
1,628 Posts
that was covered in another thread somewhere around here.

The short answer was that it's not that sort of camera, that it's closer to a wireframe rendition for the system that interprets the input, so it's not suitable for dashcam applications.
 

·
Registered
Joined
·
90 Posts
Discussion Starter #3
Thanks. But, although the data used for the stereoscopy is likely like "wireframe," it's extremely likely that the actual camera sensor is an off-the-shelf sensor, much like you would find in a consumer camera. The camera body probably incorporates a custom chip computer that does the preprocessing to convert the images to the "wireframe" form that is useful for further processing; in fact the assembly probably contains a custom stereoscopy chip. Just the same, I'm asking about the camera, which would be the sensor and its output, prior to further processing.

It's possible that all that is incorporated on-chip with the sensor, but that is unlikely because it would be more expensive and less flexible for future upgrades.

Here's a powerpoint presentation that shows engineering images from Eyesight. It obviously uses a standard camera, and in addition it's a CMOS sensor: http://vasite.org/images/meeting/062415/subaru_next_generation_eyesight_and_future_strategy.pdf
 

·
Registered
Joined
·
1,628 Posts
Thanks. But, although the data used for the stereoscopy is likely like "wireframe," it's extremely likely that the actual camera sensor is an off-the-shelf sensor, much like you would find in a consumer camera. The camera body probably incorporates a custom chip computer that does the preprocessing to convert the images to the "wireframe" form that is useful for further processing; in fact the assembly probably contains a custom stereoscopy chip. Just the same, I'm asking about the camera, which would be the sensor and its output, prior to further processing.

It's possible that all that is incorporated on-chip with the sensor, but that is unlikely because it would be more expensive and less flexible for future upgrades.

Here's a powerpoint presentation that shows engineering images from Eyesight. It obviously uses a standard camera, and in addition it's a CMOS sensor: http://vasite.org/images/meeting/062415/subaru_next_generation_eyesight_and_future_strategy.pdf
Just relaying what was in that thread. I wanted to get a link for you, but was having trouble finding that conversation.
 

·
Registered
Joined
·
2,196 Posts
Thanks. But, although the data used for the stereoscopy is likely like "wireframe," it's extremely likely that the actual camera sensor is an off-the-shelf sensor, much like you would find in a consumer camera...
I agree, and have suggested that myself, but was ignored by the few that insist it's just some kind of useless machine vision. I came across a document called Eyesight teardown (or something like that, google should find it), which had the specs for the "Gen 2" system (it was partially in Japanese, so quite possibly the US Gen 1 system, but not sure). It listed the sensors as something like 0.4 megapixel each.
 

·
Registered
Joined
·
90 Posts
Discussion Starter #6
I agree, and have suggested that myself, but was ignored by the few that insist it's just some kind of useless machine vision. I came across a document called Eyesight teardown (or something like that, google should find it), which had the specs for the "Gen 2" system (it was partially in Japanese, so quite possibly the US Gen 1 system, but not sure). It listed the sensors as something like 0.4 megapixel each.
Thanks. That helped, and I did find info on version 1, which is here: https://www.google.com/url?sa=i&rct=j&q=&esrc=s&source=imgres&cd=&ved=2ahUKEwiJysH47qvkAhWMsp4KHeLkC04QjRx6BAgBEAQ&url=https://www.marklines.com/statics/teardown/fomalhaut_003_ja.pdf&psig=AOvVaw0qHCC8UATNlANu8jSlhjDc&ust=1567298206490348

That one uses a CCD sensor (off the shelf Sony sensor -- Sony is the leading mfr of image sensors). According to the powerpoint I found from 2015, they have switched to CMOS sensors, which follows industry trends.

Because the system is purpose-built, it of course does not have a means to offload video data. However, in a redesign, it would be very easy to incorporate that. From a marketing standpoint, however, it might be embarrassing how low the resolution is of the camera they are actually using. (The higher the resolution, the more compute power is needed to process the video stream for image understanding.)

In that version, the camera sensor used is the Sony ICX278AK, whose datasheet is online, indicating it has a resolution of 768x494, or about the same as standard old NTSC video. The newer versions of Eyesight undoubtedly have higher resolution. Although the data sheet says it is for color video cameras, it's probably a monochrome chip with the expectation that a CCD video camera would use 3 chips.
 
1 - 6 of 6 Posts
Top