Industrial CMOS camera as an entry-level HD cinema camera (1080p)

[Charlie Dijak]

The rolling shutter ?skew? example provided in Steve Mullen?s write-up (cited by Jason Rodriquez in Post #11) shows bottles that are moving down a production line.

What is the travel speed of the bottles?

What is the exposure time?

What is the line rate of the sensor?

Mr. Mullen doesn?t list the settings he used. All 3 of these factors influence the degree of skew. Even more important, the subject material emphasizes the affect. Mr. Mullen?s intent was not to exhibit the general overall image quality produced by a rolling shutter. His intention was to dramatize a worse case scenario --- and he succeeds.

 

Mr. Warr?s request was for a low cost camera capable of capturing HD-type resolution.

The SILICON VIDEO 9T001C camera captures a 1920 x 1080p resolution at up to 27 frames per second. The camera, with the XCAP-Std program, captures uncompressed video for hours ? depending on the capacity of the host computer system. The price for the camera and software is approximately $2500.00.

 

Mr. Rodriguez indicates that the Micron sensor is too slow for cinematography. I?m guessing he is basing his conclusion on Micron?s ?Typical Value? pixel clock rate of 48 MHz ? which would allow a maximum of 19 fps at 1920 x 1080p resolution. The SV9T001C camera provides a user-controlled variable pixel clock capable of 70 MHz ? which allows 1920 x 1080p resolution at 27 fps. Reducing the clock rate to 64 MHz provides 24 fps.

 

Further, Mr. Rodriguez says, in reference to the XCAP software, ?it's just not very straight forward when it comes to recording as a cinematographer in the field (which is not what it's intended for).?

 

The XCAP program is designed to maximize image quality and to maximize the user?s freedom to select the parameters used to capture and display images. Advanced users can choose to 'tweak' many esoteric camera parameters - parameters that most cameras simply 'lock in'. Some users even study the camera sensor?s data sheet, and XCAP allows experimentation and application of non-standard modes; all with XCAP?s standard menu options. Other users, with less demanding requirements, are happy simply using the common controls - shutter, gain, white balance, AGC, frame rate - and leave the esoteric settings alone.

 

After a movie clip has been captured it can be transferred to any editing program that can read video files in industry standard formats. XCAP is not specifically intended for movie editing.

 

As Mr. Rodriquez indicates in Post #15, EPIX offers frame grabbers that support many CCD cameras including cameras from Silicon Imaging, Imperx, and Pulinx ? but none of these cameras approach the low cost of the SV9T001C.

 

The SV9T001C camera is available for evaluation. Alternatively, contact me at EPIX and request that EPIX capture a test sequence in its imaging lab. You suggest the subject. We have no interest in selling you a product that doesn?t meet your requirements. If the SV9T001C doesn?t meet your needs, then we?ll be happy to help you find a camera that does (but expect an alternative camera to cost more than $1000.00).

 

Charlie Dijak

EPIX Sales Dept

cad@epixinc.com

[Phil Rhodes]

Hi,

 

I think the most useful things would be for you to shoot some of the standard cinematography-oriented test charts, resolution targets and dynamic range tests, as well as some examples of moving subjects to verify this skew issue. The idea of placing a cylinder painted with vertical black and white stripes on a turntable and taking some frames of that would seem to make any skew issues obvious as the lines wouldn't appear straight.

 

Resolution targets, such as a Siemens star, will test resolution, low pass filtering, and any colour moire issues with the debayering algorithm. I don't know what sort of qualities your usual machine-vision clients look for in lenses, but it'd also be worth seeing a geometry and aberration test, usually done by shooting a chart with circular resolution targets in the corners of the frame and with high contrast in the same locations.

 

One of the most critical issues with electronic imaging in current cinematographic practice is dynamic range, and the usual way to test this is to shoot a lightbox with strips of ND filter on it layered to provide, say, half-stop increments. Usually you'd ensure that the result includes areas that go from completely black to comprehensively overexposed, so as to be able to evaluate the dynamic range of the sensor, and ideally quote the absolute luminance of each patch as measured with a spot meter.

 

These are all technical tests you can do without having to get a cinematographer in to light them, although you should be as transparent as possible regarding your experimental method.

 

Regards,

 

Phil

[Charlie Dijak]

Response to Troy Warr ? Post #21

 

The EPIX engineering department is constantly evaluating new CMOS sensors. It is the sensor design that determines whether the camera has a rolling or a global shutter. Sensors with global shutters are more complex and more expensive than sensors with rolling shutters.

We will put up on our web site in a few days a 1920 x 1080 sequence taken with our SV5M12 camera that has a 5 megapixel sensor with 2592 x 1944 pixels. It runs faster than the SV9T001C ? and it also captures 1920 x 1080 resolution at 24 fps and faster. I?d like to emphasize one last time that the SV9T001C camera definitely has the ability to capture a 1920 x 1080p resolution at 24 fps.

 

Charlie Dijak

EPIX Sales Dept

cad@epixinc.com

[Luke Prendergast]

Hi Charlie, and thanks for contributing.

There is a tag under the Frame Rate Examples table on the 9T001C page - 'Note: Pixel clock frequencies higher than 48 MHz may reduce image quality.'

Is this just a resolution hit or something more sinister?

[Charlie Dijak]

Luke -- Post # 29

 

Give me a week to capture an image at 1920 x 1080p resolution and get it posted on our web site. The pixel clock frequency will be 64 MHz -- 24 frames per second. I'll notify you, here, when it is up there. The images already posted were captured at 48 MHz. You'll be able to make your own comparison.

We'll try to get some short sequences posted as well.

 

Charlie Dijak

EPIX Sales Dept

cad@epixinc.com

[Charlie Dijak]

Wednesday, 24 January 2007

 

The EPIX webmaster tells me the promised images from the SV9T001C camera will be posted within the next 2 days.

I?ll be back and let you know exactly where to find the images as soon as they are available.

 

1920 x 1080 resolution -- 24 frames per second.

 

Best Regards,

Charlie Dijak

EPIX, Inc.

cad@epixinc.com

[Troy Warr]

Wednesday, 24 January 2007

 

The EPIX webmaster tells me the promised images from the SV9T001C camera will be posted within the next 2 days.

I?ll be back and let you know exactly where to find the images as soon as they are available.

 

1920 x 1080 resolution -- 24 frames per second.

 

Best Regards,

Charlie Dijak

EPIX, Inc.

cad@epixinc.com

 

Charlie,

 

Thank you! I appreciate this, as I'm sure the others in the forum do, too. I'm very eager to see the images and footage.

 

Incidentally, do you have information on your website and/or publish a price for the SV5M12 camera that you mentioned? I wasn't able to find anything, though I may have been looking in the wrong places.

 

Thanks again.

[Charlie Dijak]

Wednesday 31 January 2007

 

The promised images are posted at http://www.epixinc.com/gallery/sv9t001.htm. There are 2 new samples ? both using the ?Golf Guy? and ?Color Rendition Chart? used in previous images. You will be able to compare the image quality available at 24 frames per second with the previous images ? that were captured at slower data rates.

 

One of the new postings is a single image captured at an exposure time of 1/26th of a second and a Gain setting of 0. This image demonstrates the best color quality available from the SV9T001C camera when operating at 24 frames per second with a 1920 x 1080 resolution.

 

The other posting is a sequence of 29 frames, 1920 x 1080 resolution, capturing the Golf Guy swinging his golf club and hitting the golf ball into the plastic flag pole ? resulting in the golf ball ricocheting back, and bouncing up and down a few times.

I changed 2 parameters before capturing this sequence ?

1. I increased the Gain to near maximum ? so that I could use a shorter exposure time (faster shutter speed).

2. I increased the shutter speed to 1/400th of a second (2.5 milliseconds).

 

The resulting images are nice and sharp ? you can clearly see the dimples in the golf ball as it bounces. The trade-off is that the increased gain increased image noise ? the colors aren?t quite as vibrant as in the single image shot at 1/26th of a second.

 

I?ll be happy to answer any questions.

 

Best Regards,

Charlie Dijak

EPIX, Inc.

cad@epixinc.com

[Charlie Dijak]

Troy Warr -- Post #32

 

The SV5M12 camera hasn't been posted on our web site yet. When it is posted -- which should be within the next 10 days, it will be listed as the "SV5C10". The SV5C10 designation indicates 5 million pixels, Color, 10 frames per second (at full resolution).

This camera is based on the MT9P001 sensor from Micron. The Micron sensor has the potential to run at a pixel clock rate as high as 96 MHz -- but the digital serial interface we are using won't support that rate. The maximum pixel clock rate of the SV5C10 is 70 MHz -- the same as the SV9T001C -- which results in a maximum frame rate, at full resolution, of 10 frames per second. The maximum frame rate available at 1920 x 1080 resolution is 22 frames per second.

EPIX engineers will be introducing a new camera design that will offer a 96 MHz pixel clock. I can't predict when this new design will be available.

 

Best Regards,

Charlie Dijak

EPIX Sales Dept

cad@epixinc.com

[Phil Rhodes]

Hi,

 

Well I think it's pretty clear what's needed for cinematography work:

 

- 1920x1080, but preferably 2048x1080. 1280x720 is usable, but probably not worth the hassle.

- 8 bits/pixel precision, but preferably 10

- 24fps, almost essentially 25 and 30, but really as much as you can possibly achieve

- Zero or imperceptible skew

 

I suppose you don't have much to do with HD-SDI interface in your industry, but that's what's used most often and it's easiest to hook up to cinematography-oriented record hardware.

 

Phil

[Adam Paul]

Industrial CMOS camera as an entry-level HD cinema camera (1080p), Possible? Practical? Any experiences or feedback? Options

 

Isn't that what the Si2k basically is? Specially the mini version.

[Brian Wells]

a $1000 1080p setup (actually more like 2K) that includes camera, capture card and frame grabber software is very intriguing. The sensor specs can be found here.

Hi, You can buy a USB2 camera with the same sensor for under $250 at Mightex.com.

[Troy Warr]

Wednesday 31 January 2007

 

The promised images are posted at http://www.epixinc.com/gallery/sv9t001.htm. There are 2 new samples ? both using the ?Golf Guy? and ?Color Rendition Chart? used in previous images. You will be able to compare the image quality available at 24 frames per second with the previous images ? that were captured at slower data rates.

 

Charlie, thank you so much for doing this. To be honest, these images are better than I had expected. The color performance is definitely there, and the motion skew is not really that apparent when watched at full speed.

 

There were a couple of things that I noticed, though, and was wondering if you had any input:

 

- Around some of the highlights (e.g. the spot reflections on the Golf Guy's hat and sides), there is what appears to be some color banding/posterization. I played around with the saturation level in Photoshop and found that the effect diminished as expected as saturation was decreased - do you think that these stills/clips may be a bit overly saturated? Through the camera's software, are there tools to adjust the saturation, contrast, and other image qualities before capture?

- The image skew can be seen clearly on still frames, most notably in the golf ball just after it is struck by the club. The shaft of the club also appears bent during the approach. I'm assuming that the shutter speed of 1/400 sec. has a minimization effect on this - is that the case? Or does the shutter speed not effect the scan rate of the rolling shutter? I would probably be shooting in the 1/48 to 1/60 range most of the time.

 

Please let me know if I'm being greedy, but I think that the last major test of the camera's cinematographic capabilities would come from a capture from a more "life-like" setting or subject. At some point, might you be able to snap a frame of a human subject, and/or a natural outdoor setting? The images that you've provided are great at showing the camera's limits to color rendition and exposure latitude, but it's difficult to imagine a more "typical" film subject in this context. That would be much appreciated and would help greatly to judge the camera's capabilities.

 

Thanks again, Charlie.

 

Industrial CMOS camera as an entry-level HD cinema camera (1080p), Possible? Practical? Any experiences or feedback? Options

 

Isn't that what the Si2k basically is? Specially the mini version.

 

Hi Adam,

 

That's exactly right, and I alluded to that a little upstream in this thread. The issue is that I don't have the $15,000 to spend on an SI-2K mini rig and software, so I'm looking for a considerably lower-budget solution. The SI-2K uses a pretty high-end sensor, but I was intrigued by the other sensors out there that offer similar resolution and pretty decent image quality for the price (albeit not in the same league as the SI-2K).

 

Hi, You can buy a USB2 camera with the same sensor for under $250 at Mightex.com.

 

Hi Brian,

 

Thanks very much for the link. I checked it out and unfortunately it looks like the camera (assuming that you were referring to this model) isn't capable of 24fps at 1080p resolution, as the Epix model is. I did notice that the 1.3 MP model is, though (as is the former model with a 1280x720 ROI), so that might be a good way to get my feet wet before committing $2500 to the Epix setup.

 

Just curious, have you had a chance to use that particular camera before, or did you just notice the sensor specs?

[Phil Rhodes]

Hi,

 

Interesting sample images. Assuming that insidious skew issue doesn't pop up too much in certain situations, I can't see why you couldn't shoot a movie on one of those.

 

It's not absolutely razor sharp, but then it's Bayer, so...

 

Do we know exactly what the true size of the CCD is?

 

Phil

[Brian Wells]

Just curious, have you had a chance to use that particular camera before, or did you just notice the sensor specs?

Yes, mine came in this week. I am extremely disappointed with the noise and rolling shutter. IMHO, the micron sensor is useless for any kind of professional imaging. It's as bad as a webcam. Speaking of, there are webcams that use megapixel CMOS sensors. In fact, I wouldn't be surprised to learn they were using the same CMOS as the industrial camera. They are really awful. I can't begin to understand the fascination with using industrial cameras for professional imaging when there are HDV cameras that look better for less money. For that matter, film stock through a $200 movie camera still looks better than all forms of HD.

[Troy Warr]

Thanks for the feedback, Brian. Were you able to compare the footage that you're getting with the footage that Mr. Dijak provided earlier in this thread? I'm curious to know if that's at all similar to your results. That footage looked somewhat promising to me, but as I mentioned to Mr. Dijak, it can be a bit difficult to judge without some more "cinematic" subjects. My technical knowledge of industrial camera technology is limited, but I would guess that even though both cameras feature the same sensor, there would have to be some design shortcuts taken with a camera system that's 1/10 the price that would ultimately be detrimental to image quality. The rolling shutter problem, though, might very well be the same, and I agree that that would certainly be a deal-breaker for cinematographic use.

 

In any case, my personal fascination with industrial cameras is that they seem to have the potential to deliver great "bang for the buck" in terms of image quality, features, and versatility. Whether or not that's true in practice is really the main subject of this thread. Despite your experiences with a $250 camera, I think that it's dismissive to use that as a basis for criticizing the entire concept. Sure, there are HDV cameras that make great footage for a few thousand bucks, but you're fundamentally limited to interlaced 1080i video (or a fudged 24p equivalent), high compression, relatively poor low-light sensitivity due to small pixel size, aliasing (in some models), fixed frame rates or very few frame rate choices, as well as many other drawbacks that come with the pro-sumer grade of cameras. The industry seems on the verge of solving some of these problems, for example in the Panasonic AG-HVX200, but even it has its own problems (high cost) and limitations (most frame rate choices only available in 720p).

 

Have you seen any footage from the Silicon Imaging SI-2K? Granted, that's more expensive than most, if not all HDV format cameras out there today. But, at about $12,500, it's not worlds above, and it's capabilities, in my opinion, rival those of many near-$100K cameras out there.

 

I don't want to touch the $200 movie camera argument for fear of turning this into another film vs. video thread...

[Charlie Dijak]

Response to Phil: Post #35

 

The SV9T001C delivers good quality color images, 1920 x 1080 resolution, at 24 fps. The maximum at 1920 x 1080 resolution is 27 fps.

 

Response to Brian: Post #37

There is a lot more to a ?camera? than a sensor. The SV9T001C camera with the XCAP-Std program captures 1920 x 1080 resolution images to hard disk array ? uncompressed ? at 24 frames per second. The frame rate can be easily adjusted. The resolution can be easily adjusted. There are controls for color and gain. You don?t need to write a single line of code to get it to work ? you only need to install the system in a computer with a RAID array.

We (EPIX) have color example images posted on our web site demonstrating the fact that we have a working system ? and we offer anyone the ability to purchase the system for evaluation in order to prove to themselves that the system performs as we advertise.

 

Response to Troy: Post #38

Are the images overly saturated?

I saw the glare spot on the Golf Guy?s hat. No argument ? it?s certainly there. I tried to balance the exposure of the Golf Guy with the exposure of the panels on the Color Checker ? and that?s the result. The GUI includes separate controls for Brightness, Saturation, and for Gamma.

 

Does the 1/400th of a second shutter speed minimize the skew?

Or does the shutter speed not affect the scan rate of the shutter?

Answer ?

The scan rate (the rate at which rows of pixels are exposed and read-out) is affected by the frame rate ? an increase in the frame rate accelerates the rate at which the rows need to be exposed and read out. The faster exposure time (1/400th of a second) reduces the amount of blur exhibited in any particular line of pixels. If I used a slower exposure time the distortion due to skew wouldn?t change?but the skew would be masked by the increase in blur.

 

The request for more ?life-like? setting or subject.

I hope to be able to make the time to capture such a sequence next week. I?m anticipating something outdoors under sunlight with some degree of motion and color with a human in it so you can see skin tones.

 

Phil ? Post #39

The true sensor size?

The Micron MT9T001 sensor used in the SV9T001C camera has a maximum resolution of 2048 x 1536 pixels. The pixels are 3.2 microns square. The active capture area, at maximum resolution, is 6.55mm x 4.92mm ? a ½? sensor is defined as a sensor measuring 6.4mm x 4.8mm.

BUT ? if the goal is 1920 x 1080 resolution, then the active capture area is smaller ? 6.14mm x 3.46mm

 

Brian ? Post #40

http://www.epixinc.com/gallery/sv9t001.htm

 

Best Regards,

Charlie Dijak

EPIX, Inc.

cad@epixinc.com

[Phil Rhodes]

Hi,

 

That's very interesting - thanks.

 

They do seem a bit over saturated, but not anything terribly worrying. The problem is the apparent deformation in the shape of the golf ball at frames 3,4,5,9 and 10 of the test sequence - I don't think that would really be very usable. I presume the readout method used by the sensor in this camera is what allows it to be so competitively priced. What's the nearest equivalent that does full frame transfers?

 

Phil

[Cesar Rubio]

Phil:

 

The AVT Pike F-210c has a 1' Kodak CCD with global shutter (full frame read outs). It can record 1080p at 30 fps. Also the pixels size are 7.4 x7.4 microns. The price is $5650 but it's worth every single dollar!

 

You can also get the AVT Pike F-145c (720p, 30fps) that has a 2/3 CCD Sony sensor, with global shutter and pixels size of 6.45x6.45 microns. The price is $3490

 

I don't really see the need over 720p, if you ain't going to show your work on the BIG screen. Actually I have knowledge of some movies shot with SD cameras, and transferred to film for BIG screen projection. I haven't see the quality thought.

 

Thanks,

Cesar Rubio.

[Luke Prendergast]

Still wondering precisely what potential issue 'Note: Pixel clock frequencies higher than 48 MHz may reduce image quality.' is referring to.

[Phil Rhodes]

Hi,

 

> The AVT Pike F-210c has a 1' Kodak CCD with global shutter (full frame read outs). It can record 1080p at 30

> fps. Also the pixels size are 7.4 x7.4 microns. The price is $5650 but it's worth every single dollar!

 

That ought to be extremely nice. I'm sure you don't hold those in stock, but it'd be interesting to see some images.

 

> Actually I have knowledge of some movies shot with SD cameras, and transferred to film for BIG screen

> projection. I haven't see the quality thought.

 

If it's done extremely carefully it can look alright-ish, in that after fifteen minutes or so, if the story is strong enough, you don't see it every single cut like you did to begin with. I don't think it's really very laudable, and frankly it's now pointless as the HDV stuff is so much better.

 

Phil

[Troy Warr]

Does the 1/400th of a second shutter speed minimize the skew?

Or does the shutter speed not affect the scan rate of the shutter?

Answer ?

The scan rate (the rate at which rows of pixels are exposed and read-out) is affected by the frame rate ? an increase in the frame rate accelerates the rate at which the rows need to be exposed and read out. The faster exposure time (1/400th of a second) reduces the amount of blur exhibited in any particular line of pixels. If I used a slower exposure time the distortion due to skew wouldn?t change?but the skew would be masked by the increase in blur.

Ah, I *think* that I get it. So, the scan rate is affected by the frame rate, but not the shutter speed? In other words, provided that I'm always shooting at 24fps, using a 1/48 shutter speed would give me no different skew than a 1/400 shutter speed - is that correct? But, as I decrease the frames per second, therefore reducing the scan rate, the skew in turn becomes more pronounced? Provided that I would generally be shooting with a slower speed, probably in the 1/48-1/60 range, the skew might actually be less noticeable due to the increase in motion blur?

The request for more ?life-like? setting or subject.

I hope to be able to make the time to capture such a sequence next week. I?m anticipating something outdoors under sunlight with some degree of motion and color with a human in it so you can see skin tones.

That would be wonderful - I think that we'd all look forward to seeing that. If possible, it would be great to at least briefly see a few varieties of motion that you normally see in a motion picture - for example, a subject walking in a full-body or torso shot, a more rapid motion (e.g. waving a hand near the lens), and if possible, a relatively fast pan or camera movement, and maybe a slower one as a contrast. That would give us a better idea of the extremes of image skew, as that seems to be the major issue that we're concerned with.

 

Beyond that, I think it's only fair that any of us that are interested in the camera further pick one up for evaluation and shoot further tests for our own purposes. Charlie, you've been very helpful thus far, and I, for one, appreciate it!

[Troy Warr]

Phil:

 

The AVT Pike F-210c has a 1' Kodak CCD with global shutter (full frame read outs). It can record 1080p at 30 fps. Also the pixels size are 7.4 x7.4 microns. The price is $5650 but it's worth every single dollar!

 

You can also get the AVT Pike F-145c (720p, 30fps) that has a 2/3 CCD Sony sensor, with global shutter and pixels size of 6.45x6.45 microns. The price is $3490

 

I don't really see the need over 720p, if you ain't going to show your work on the BIG screen. Actually I have knowledge of some movies shot with SD cameras, and transferred to film for BIG screen projection. I haven't see the quality thought.

 

Thanks,

Cesar Rubio.

Hi Cesar,

 

I'm very curious to see footage from your AVT cameras as well, specifically some 720/24p and 1080/24p samples if you plan to shoot at those frame rates. I occasionally check up on this forum thread but was wondering if you had an idea of when you might post the footage.

 

BTW, you haven't experimented with 35mm SLR lenses on your 3-D camera setups, have you? If I eventually pick up an industrial camera, I'm going to be very interested in using 35mm lenses, probably with a 35mm DOF adapter, for that shallow depth of field look. I checked with Dennis Wood of Cinevate, and he seemed to think that would be possible with some tinkering. I realize that footage shot with a 2/3" or 1" sensor will be capable of decently shallow depth of field without an adapter, but I'm not really sure to what degree. I assume that the Silicon Imaging SI-2K interfaces lenses directly to the sensor, and its depth of field control looks promising from the sample footage, but I'm hoping for the shallowest depth of field possible.

[Phil Rhodes]

Hi,

 

Well, a 1" sensor is close to, or even bigger than the size of a 35mm film frame - about the size at which you have to start becoming concerned that even 35mm-format lenses won't cover it.

 

Depends what sort of size of frame you want to window out of the sensor, really.#

 

 

Phil

[Cesar Rubio]

Phil:

 

Right now I am working with my Computer set-up and the triggering system for perfect synch with two cameras. When it’s ready I will post some 3-D videos.

 

I only have a SD set-up at home with two Marlin's F-033c and the quality looks good for me! (Personal use)

 

Probably by the end of next week or beginning of the following one, I will post some footage of a Pike F-145c that I am waiting for. That camera shoots 720p which is 4 times the size of 360p (360x640 pxs. 16:9 of 480p). Most TV stations work with either 720p or 1080i....1080p requires too much bandwidth for transmition!

 

Here is a still image of the Marlin F-033c:

 

http://www.davidrubio3d.com/view_topic.php...amp;forum_id=60

 

 

A 1 1/2" sensor might be comparative to a 16mm film format. See the attachment for the real size in mm of sensors.

 

Troy:

 

There are some 35mm lenses adapters to fit c-mount cameras, but I have not tested one yet. I am planning to buy one for my Nikon lenses. They cost around $75 US dlls.

 

Thanks,

Cesar Rubio.

Sensor_Size.bmp