Image Sensor / Optical Components


#1

Image Sensor / Optics

The DSLR solution doesn’t work because of the mechanical shutter that lies on top of the sensor. It’s not the mirror, or even the shutter at the film plane - it’s the shutter that clears the sensor. Also, these sensors are rolling shutters, which require us to run the film at slow speeds to avoid the “jello effect.” Here’s a quick description of the difference between a rolling shutter and global shutter.

The imaging sensor is by far the most expensive part of the system. On the cheap end, you have small board cameras like the Raspberry Pi camera, which is very similar to the camera in a phone. On the high end, you have 4K sensor capable of running at 30fps. In-between are DSLR prosumer options like the Canon cameras, a micro 4/3, or even the RED camera. Ideally, we build a system that can take a variety of sensor options to fit a variety of budgets.

How do we do that? As long as we provide a clear digital signal when its time to take a picture, we should be good. The positioning of the camera will change depending on its dimensions, so ideally the camera platform has adjustments to move it up/down, side/side. Take a look at how Retroscan has done this (even though you aren’t supposed to remove their camera). In my previous design, I used a macro-photography tripod plate and a microscope stand. These are still possibilities and they’re cheap.

Industrial Camera Options are categorized by two main factors: sensor type and the data interface. First is the type of sensor. There are frame sensors and line-scan sensors. Frame sensors are like every other camera. They sense light on a rectangular plane and turn that into an image. The line-scan cameras are like flatbed document scanners. They sense light in a single line of pixels and either it moves or the image being scanned moves to the next row of pixels until scanning is complete.

Line scan cameras offer more speed and do not require us to stop the film. The only thing we need to do is signal when a frame has started and ended so that we create a separate file for each frame.

Frame sensors would force us to either build intermittent motion into the system or to run it at a speed slower than the shutter speed of the camera.

In both of these categories, the speed of capture is determined by the data interface, which means the cable type and data processing necessary to move data from the sensor to a processor and then to storage. To run anything at high speeds, you need to buy additional hardware, called Frame Grabbers. These act like a buffer so that you can keep capturing information without waiting for the previous data to be safely saved first.

In all cases, these cameras do not come with lenses. The do, however, almost always come with their own software SDKs with a license that allows us to use that code in our application (I will verify the details on this as we get closer to a final solution).

So far, these are the industrial cameras I’ve been looking at:

Frame Cameras
Point Grey Chameleon (2K, USB3) $595 (used by Retroscan). Sensor only.
Point Grey Grasshopper (4K USB3) $3700. Sensor only.

Line Scan Cameras
Teledyne Piranha 4 - (4K, CameraLink, requires Frame Grabber hardware). Estimated price = $4000 USD for sensor only. Frame Grabber = $1200. + lens (price unknown). Total estimated at ~$8000 USD.

Here are some board camera options
Rasperry Pi Camera (HD-ish, USB) $25 requires macro lens adapter.
13MP board cam (HD, USB3) $139 (used by Gugusse) requires lens.

There is also news of being able to run a Linux distro on Canon DSLRs using Magic Lantern. I wonder if we can figure out a way to use this to (or just Magic Lantern) to use the Live View RAW image feed for DSLRs to capture our frames for us? This would be the cheapest option by far since most people would have access to a Canon DSLR and therefore our most costly part just disappears!


#2

Sony’s new A7RII ($3200) has a completely digital shutter in its “silent shooting” mode. On top of that it also has an extremely good 7952 x 5304 backlit sensor. I was wondering if a computer controlled stepper motor transport mechanism that pauses on each frame capture would be feasible; this would eliminate rolling shutter problems and potentially yield greater than 4K resolution.


#3

I have been through this a few times, and have designed a lot of scanning rigs over the past few years.
If you want great quality on a budget, the best option as far as dynamic range, low noise and good framerates @ 2K is the Point Grey Blackfly BFLY-U3-23S6C-C.
You can easily get 30fps via USB3, at full bit depth. The camera is easily triggered using a $2 hall effect sensor and magnet, the camera itself is tiny and is easily retrofitted onto an existing projector or edit table.
The manual trigger and high framerate allow you to capture in realtime.
You can use an enlarger lens and cheap extension tubes to give a razor sharp image across the entire frame. The lens is typically under $100. I have used these in Super8, 16mm and 35mm scanners with great success, the results have often won shootouts against the Spirit 2K and other scanners.

As for DSLRs, you can often use a HDMI capture board, such as those from BMD and capture the stills from the computer that way, but the blackfly is very good value and gives astoundingly good results.


#4

Hi Peter,

when you mentioned that a lens under $100 would be sufficient, I did a quick search for enlarger lenses on ebay and found quite a lot of them in different price ranges, manufacturers etc.
Would you recommend any specific brand of lenses for use with the Point Grey Blackfly?
I’m eyeing for a Schneider Componon (-S). Do you think that would be a good match?
I’d also need to figure out what focal length works best to not distort the image. Maybe a standard 50mm would work?

Thanks for your input!


#5

@Peter Thanks for the recommendation. I’d talked to a PointGrey rep about 4K options but I think a 2K option would be fine for many users. It’s a great option. I"ll have to start thinking about how to make the design modular so users can select the image resolution they want depending on their budget.

As for 8mm board cameras, I came across this site for anyone who is interested:
http://www.uctronics.com/
via http://www.arducam.com/

@Peter the lens companies are separate from PointGrey, no? Who have you worked with in the past?


#6

That is the exact lens I am using, the 50mm works well for me. I use adapter rings so I can mount it backwards for best results. Pin sharp from edge to edge, even with 35mm film.


#7

@Peter Thanks a lot for your answer!


#8

@Peter is this the one? And you’re using it with the PointGrey Blackfly with an S-mount, correct?

I just got this module delivered and thought I would try your lens out. Also was recommended this one for the cheaper board cameras and ordered it as well. Will do tests soon as I can.


#9

Hi @matthewepler,

could you please repost the first link (item not found)?
I’ve got a PointGrey Blackfly last week and it has a C-Mount with an optional CS-Adapter.


#10

I’m currently testing out the PointGrey Blackfly camera but I’m having some issues.

I’m trying to hook up my Blackfly with a Schneider Componon-C 50mm lens (the lens has a M39 mount): Lens (M39 mount) -> M39 to M42 (with T2 thread) adapter-> M42 to c-mount adapter-> camera (Blackfly). The problem is that I can’t get focus on anything. I’ve never dealt with c-mount lenses before, but I think the flange focal distance due to the various adapters might be the problem.
@Peter: You mentioned using Componon-C lenses with the Blackfly. Do you have the same problem?
@matthewepler: Have you already been testing your lens with the Blackfly?

Thanks for your help!


#11

I’ll take some photos tonight of my setup. I have a bunch of extension tubes to get the correct distance, and the lens is reversed on my setup.

The focal distance to target is very short, im my case about 50mm (2 inches or so).


#12

Hi @Peter,

that would be very helpful for me! Thanks a lot!


#13

Sorry - here ya go: http://www.adorama.com/SR5028CNS.html?gclid=CjwKEAjw-IOwBRD1wrTC27fSjFISJABUDZ1788_KuakAURWraJHuiTbF0CGlqf5HumnJusaTFQX_HBoCQHvw_wcB

No testing yet. I just found out I have to move again - back to Brooklyn so I have to pack up the lab and it’s thrown a wrench in things. Also, I haven’t really said much about it yet but I’m actually focusing on an 8mm machine for now, then will get back to the big boy.


#14

Hi @Peter,

I’m sorry to bother you again, but I’m currently experimenting with the lens setup.
I’m getting something in focus now, but to get a 35mm frame centered and focused I figured out that I’d need something like a total of 60mm c-mount extension tubes and would result in having a working distance from lens to film of about 200mm (20cm!), which I think is a bit much.
Could you please let me know how you manage to have a distance of about 50mm.

Note: I have not reversed the lens yet, which might change the distance also.

Thanks a lot!


#15

You need to reverse the lens.


#16

Just my opinion, however, I think that using sensors that run $4000+ kind of defeats the purpose of this project. There’s already existing solutions in that price range, and my impression (or hope) was that Kinograph, especially for 8mm, would be an open, low-cost solution for the DIY type of person.

Has any thought or work been done previously to explore the use of video cameras rather than still-frame or machine vision cameras? I believe the “Sniper” models used video cameras. I could see that advantages for this would be that there’s many low-cost 3-CCD HD video cameras with zoom lenses that may simplify the hardware requirements and specialty equipment needed.

On the line-scan front, I could see where you may not even need to register the individual frames in real-time, but rather leave that to post-processing to identify and align the frames.

For all sensor types, what about using a B&W sensor and then either use 3 different exposures via different wavelength LEDs, or use 3 B&W sensors with different LEDs exposing each? You could even add a 4th exposure with IR LEDs to capture a dust/scratch reference similar to Epson’s digital ICE mechanism.

Also, given the limited dynamic range of electronic sensors vs. film, if using LEDs, you could even do 3 (or more) exposures per color and then combine in software to essentially create a high-dynamic range image for each frame.

Of course, all these are just thoughts on approach as I’ve never done this. Just trying to reapply some concepts from other photography areas.


#17

Another manufacturer for sensors/mount-cases:
ImagingSource
(I think they are bayered when colour, and I do not know the specs in depth)
there’s an opensource linux API I dont know anything of, and on-cam triggers.

On the lens:
Isn’t it of concern, that one should not use too big a focal length? Because I think that way it relies much more on the build quality of the lens, whereas close positioning to the film could use cheaper lenses… (correct me if I’m wrong).


#18

If you are looking for the 30fps then you can easily get it through USB3. You can also make camera easily triggered using hall effect sensor and magnet. Also the camera is retrofitted on the projector.
The manual trigger and high framerate allow you to capture in realtime.

assembling pcb


#19

digitap, The cheapest machine vision cameras that are worthwhile using are around USD$500 and they give incredible results, considerably better than using 3CCD HD video cameras (I’ve tried both).

I also have built systems that use a mono sensor, and flash LEDs in sequence, this only works if the film is stopped during the exposures, so requires manufacturing an intermittent mechanism, which just isn’t possible to do reliably and cheaply. If you want intermittent then you are much better off modifying an existing projector.

If you want a low cost solution that gives good results, you are still going to have to spend around $1500 plus a projector.

If you want just ‘ok’ results. you may as well just point a video camera at a projected image on the wall, as it gives about as good a result as anything that costs less than that, unfortunately.

To get good results, there are some parts that are just going to cost money unless you get lucky.
You need a good sensor/camera ($500 +)
A good lens ($150 - $600, e.g. http://r.ebay.com/nENrjJ)
Extension rings and adapters ($20-$150)
A good (preferably triggerable) light source ($100 - $800)
A film transport mechanism ($?? for scratch built, $50-$1000 for a good projector),
A trigger and magnet ($15-$25 if you can solder),
An SSD if you want to capture realtime ($150+) and a PC with USB3 ports.
Plus hardware like a focusing rail or linear stage ($50 - $400) and hardware to mount everything.
And of course software, tools and time.

Moving film around reliably, illuminating it, and capturing it in high quality requires some fairly precision parts, so it is unlikely that even in kit form you could make a complete solution for under a grand that would be worthwhile using.


#20

Manu, the imaging source cameras have noise issues and unreliable triggering compared to the PGR cameras.
The prices are a little cheaper, but not worth it in our testing.