"Typical approaches to TOF imaging have often included illumination that is periodically modulated, in conjunction with an image-sensor pixel architecture capable of demodulating the reflected signal. This demodulation of an optical signal can then be used to extract the phase change of the optical signal, thus measuring distance.
odos imaging has developed TOF 3D imaging systems based on an alternative approach: pulsed illumination. The systems incorporating this approach operate at high spatial resolutions, matching the typical resolutions of intensity-only machine-vision cameras, but crucially providing the benefit of depth measurement at each pixel. Recently released is the first commercially available pulsed TOF system intended specifically for the machine-vision market, a system with a spatial resolution of 1280 × 1024 pixels."
"Pulsed TOF imaging provides a number of unique benefits. At the application level, system integrators can appreciate the benefits of operation in high ambient light conditions and simple synchronization of multiple systems; however, pulsed illumination is also the key enabler of high-resolution TOF imaging."
 |
| An example scene with 3D information shows the intensity image (left) and the depth image (right) [false color, 816 × 600 resolution]. |
I think that they use dual-pixel with a small phase shifting. The laser pulse is a rectangle in intensifity then by using the ratio between the dual-pixel, the dephasing of light pulse can be calculated. I remember that I've seen this in a seminar that I organized many years ago. That is why there are black holes on the very bright objects, since the exposure time is limited by pixel operation and these points are saturated so no 3D information can be extracted.
ReplyDeleteWhat is your opinion please ?
-yang ni
That sounds just right. But dark areas are also clipped here - have a look at e.g. the image in the upper left area, the vase in the center or the color test chart left of the vase. On the color test chart you can also see that the distance information cannot be properly grabbed for varying reflectance levels. The darker spots on that chart result in further distance estimates in the right picture, which is a common problem appearing at low light levels. Nevertheless, it is an amazing result considering the high resolution (thus low photon counts). It is to be questioned if that image is captured with real-time settings and to what extent image processing, filtering etc. was performed. The illumination unit depicted in the publication in Laser Focus appears also very complex. I know that you can trade peak irradiance against duty-cycle while preserving eye-safety. However, it is still quite difficult to capture enough photons in small pixels to achieve a meaningful depth measurement precision - I'd love to see some more details and actual performance merits of their design at some point.
Delete- Andreas Süss
NHK has made such arrangement by using an image intensifier tube with fast gating function. The image intensifier is coupled to a CCD sensor, they claimed high resolution. The main application was for background decoration generation such as in weather forecast. But I can not find this paper reference now ...
ReplyDelete-yang ni
I doubt that image intensifiers are really helpful here. Good indirect ToF imagers are Photon noise limited - here Photon multiplication doesn't help because that doesn't affect the Poisson statistics. That is only interesting if one worries about read noise issues as in low light applications. I guess they use them only for their gating capabilities. In the early days of ToF people did that also with Kerr and Pockels cells. However, all these approaches imply fancy materials and high voltage levels - which is not particularly low cost...
Delete- Andreas Süss
http://steradian.co.jp/adv3d/techinfo/pdf/NICOGRAPH2004_H16.pdf
ReplyDeletethis one has more details.
-yang ni
The spatial resolution is great, but the depth resolution?
ReplyDeletewhat should be their depth noise level when having small pixels?
It's about time to publish an apple to apple benchmark with current leading depth sensing technologies.
Never trust a depth map :)
ReplyDeleteThis would look awful In 3D.. still don't get it why they have 1Mpx with HUGE noise..