Vivo's TOF 3D camera features 300,000 depth pixel resolution, which is said to be 10x the number of existing Structured Light Technology. It enables 3D mapping at up to 3m from the phone while having a smaller baseline than Structured Light. TOF 3D Sensing Technology is also simpler and smaller in structure and allows for more flexibility when embedded in a smartphone. This will enable much broader application of this technology than was previously possible.
Vivo's TOF 3D Sensing Technology is no mere proof of concept. The technology is tested and meets industry standards required for integration with current apps soon. Beyond facial recognition, TOF 3D Sensing Technology will open up new possibilities for entertainment as well as work.
Update: Analog Devices (ADI) claims that Vivo ToF camera uses its technology (Panasonic CCD-based):
Hi Vladimir,
ReplyDeleteAs I know,TOF is base on measure laser time difference. So why is there "300,000 depth pixel resolution"? It is more like Structured Light,isn't it?
And I had some study for TOF, most said TOF is low resolution solution and can't for payment at this moment.So I am confused why Vivo TOF solution can work.
Thanks,
Chi
The 300,000 pixel is spatial resolution with depth measured in each and every pixel.
DeleteThe old generation of ToF cameras from 10 years ago used to have 40um and larger pixels and their resolution was low. There has been a lot of progress since then. Modern ToF cameras have pixel size of order of 6um and progressing tow even smaller pixels.
Microsoft ToF camera with 1MP resolution is a more typical representative of the modern ToF technology:
http://image-sensors-world.blogspot.com/2018/05/mocrosoft-announces-3d-camera-project.html
Pulse based tof camera have low power consumption. + ccd high light sensitivity, 6um pixel size (Smaller size than cmos). So can reach 300,000 pixels. resolution.
DeleteThanks for the information.
DeleteBut I find the information for comparing the resolution for TOF & structured light. TOF is about hundred thousand pixel and structured light is about megapixel. So I confuse again,why do vivo say TOF 300,000 pixel and apple face ID 100,000 pixel?
I focus on the phone application(payment).
So could you provide more detail information for me to compare the the resolution for TOF & structured light? Thanks
@ could you provide more detail information for me to compare the the resolution for TOF & structured light?
DeleteAs I mentioned, ToF is just a common word for many different approaches, each of them being improved over time. Similarly, structured light and active stereo approaches of Apple, Mantis Vision, Intel, Orbbec, etc. are very different between them and evolving over time. So, I can't compare ToF vs structured light as a whole. Some of them work very well indoors but greatly degrade in sunlight. Others work well in direct sunlight but take too much power. Some are cheaper, others are very expensive. And all approaches are evolving, so that whatever I say now might become incorrect next year. Sorry, I can't give a better answer.
For 300,000 depth pixels, the resolution may be 640x480. Which company may supply the ToF sensor? The ToF sensor from PMDTec doesn't support the large resultion, and only support 352x287 (100k) px which is used in Project Tango.
ReplyDeleteThere is a number of possibilities. One is Sony-Softkinetic:
Deletehttp://image-sensors-world.blogspot.com/2017/12/sony-releases-bsi-tof-sensor.html
Another one is Panasonic:
http://image-sensors-world.blogspot.com/2014/12/basler-presents-tof-cameras-with.html
There are some other possibilities too.
Panasonic
DeleteBut those are large format (Sony's is 1/2-inch) sensors. They won't be able to fit the optics with reasonable FOV into a phone form factor. So Vivo must be using a sensor with smaller pixels
DeleteFor ToF sensor from SONY with a VGA resolution, the spec tells us that the distance precision at 1m is 6mm. I guess the distance precision at a close distance about 30cm will be bigger. Is that precision accurate enough for faceID application?
DeleteThe depth accuracy might improve or stay the same at shorter distances, depending on the source of errors. If it's limited by sensor controls jitter, or laser jitter, or pixel-to-pixel mismatches, it would not improve. Most probably, the 6mm accuracy is measured in darkness and would degrade in direct sunlight.
DeleteAlso, there is a number of different faceID algorithms with different degree of reliance on 3D info. So, it might be good enough for some and not good for others.
... yes Vladimir, that is correct. And also, what is always mixed up for structured light is the fact that the real 3D resolution for SL is not given by the sensor resolution but by the number of projected dots which is in the order of 10k to 30k for both Kinect and iphoneX FaceID.
ReplyDeleteThis is not true actually.
DeleteWith 30kdot you can get much higher resolutions, as it's not one depth value per dot, but one depth per symbol.
Symbols are composed by arrangements of several dots.
Thanks to specific dot patterns, different arrangement of dots (symbol) exist with much higher resolution than the reported 30k.
Different constraints/advantages for each technology
Structure light projectors are improving (cheaper, simpler) and the structure light receiver is much smaller than ToF.
Structure light can easily scale up in resolution, with higher accuracy.
Interesting. Suppose our subject is a wall and we are projecting 30,000 dots on it. How do we know what happens between the dots? Is the wall really flat or there is some terrain between the dots?
DeleteWhy said TOF will be smaller, for sensor side, I suppose it needs 50% more size. how about the light sources side, it is smaller?
ReplyDeleteThe requirements to the light source depend on their ToF approach - direct, indirect, pulsed, continuous modulation, mix of them. Some of these approaches require less light power and the illumination source can be quite compact.
DeleteThere are also more exotic ToF approaches, such as MEMS scanning one described in Primesense/Apple patent application:
http://image-sensors-world.blogspot.com/2014/10/primesense-scanning-tof-patent.html
In theory, that one can be very compact, basically a single photonics chip. I'd guess we'll see it in some future Apple products.
You can follow picozense depth camera. Pulse based Tof camera. 300,000 depth pixels resolution 640×320@30fps. The same solution as vivo NEX.
ReplyDeletehttp://image-sensors-world.blogspot.com/2018/06/pico-zense-presents-tof-camera.html?m=1
ReplyDeleteAnother view is tof has no base line concept. Light source need close to sensor side.
ReplyDeleteA VGA TOF sensor is a 300K point resolution but every point noise is way higher than a SL, especially in the front face use case.
ReplyDeleteYou might see some low noise results by utilizing averaging of multiple samples which assumes no movement...
a simple scan of a face while moving reveal the difference.
I agree with you for now. However, the situation is not static. For example, if somebody in ToF camp manages to increase the modulation frequency to, say, 3-5 GHz, the depth noise would become much smaller. I'd guess structured light can be improved too. So, the resolution and accuracy race goes on.
Deletehttps://panasonic.co.jp/es/pespl/products/new/tofcamera.html
ReplyDeleteI wonder if this is the one.