Started 3 months ago | Discussions thread
Entropy512 Veteran Member • Posts: 4,862

Yunfan wrote:

Entropy512 wrote:

Yup. You could technically replace the AF system with a TOF sensor and an illuminator in the hotshoe - but honestly that would probably not perform as well as a legacy PDAF system.

TOF sensors are great for initial hinting in the absence of PDAF, and may be beneficial for initial hinting in a case where current lens focus position is so far off that PDAF can only return an "out of range" result, but they're not going to be suitable to replace any final through-the-lens focusing system merely due to issues of lens/sensor assembly tolerances.

Most 3D TOF sensors on the market also have issues with mutual interference when multiple systems are looking at the same area. Upcoming products from IFM make advances in mitigating that, but who knows how well they'll actually perform. Scanning LiDARs are bulky but at least perform better with respect to mutual interference, since they'll only interfere when the scan pattern has them pointing at each other at the exact same time.

A lot of great points here. As a researcher working on ToF related computer vision problems, I would say ToF is far from perfect and reliable due to following reasons:

1. Dark objects, surfaces with a large incident angle to the IR illumination, and far away/tiny objects do not reflect enough active infrared illumination back to the sensor, so you cannot get readings reliably on these surfaces. I think the percentage of missing data can at least be comparable to the occlusion caused by lens/lens hoods.

Don't forget how poorly they usually perform on sharp edges.  You'll often get distance artifacting there.  Combatting distance aliasing requires frequency dithering, which negatively impacts framerate.

2. You need to align the ToF image with the RGB sensor to know which part in the frame to focus on - this can be a really big issue especially for MILC/DSLRs as they can have lens with different FOV and distortion profiles. Correcting them perfectly might be much difficult than one would assume if we also consider the manufacturing tolerances.

This will be a major PITA for interchangeable lens cameras.

3. ToF can have multi-path interferences if there are reflective surfaces in the scene, as the rays bounced back from surface A to the sensor might have hit surface B first and got reflected to surface A. This basically would make the sensor think a surface is further away than it actually is and can lead to terrible measurements.

IFM claims their next-gen sensor architecture will greatly improve performance in multipath scenarios, but TBD whether this will actually work as claimed.

4. I have not tested with this too much but I assume ToF (at least those on phones) can also get blinded by the sun if used outdoors under sunlight.

Even Leddar's 1D rangefinders have been documented by a customer (who installed an aftermarket system using one of them on a manually driven forklift) as false-alarming under skylights.

(My day job is industrial automation.)

I think ToF is therefore useful only useful for distance hinting. For any accurate and reliable AF, use PDAF or CDAF.


-- hide signature --

Context is key. If I have quoted someone else's post when replying, please do not reply to something I say without reading text that I have quoted, and understanding the reason the quote function exists.

 Entropy512's gear list:Entropy512's gear list
Sony a6000 Pentax K-5 Pentax K-01 Sony a6300 Canon EF 85mm F1.8 USM +5 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow