Project price estimation - image scanner

[nctnico]

What I'm also wondering about is why not use a USB camera? If there is a Windows PC involved anyway...

[soldar]

I was involved in something similar about ten years ago. A Hungarian company made an ID/passport reader. It worked with a video camera which took the image and processed it, OCR, etc.

I see the company still exists http://www.arhungary.hu/

[gaminn]

We didn't try usb cameras. There is a limited space available over and under the card in the machine, now there is 6 cm (but could be a little more). What distance would two (top and bottom) standard usb cameras need to output sharp image without distortion? And I also think it would not be easy to perfectly glue at least two video frames (it is not possible to take just one frame as part of the card is always hidden behind mechanism of the scanner) to obtain complete image of the card and not to distort any characters on the card (OCR is then used). And realtime. Another delay of one or two seconds is not very good. But I'm not expert in this area.

[james_s]

Seems like adapting the machine would be easier than trying to design a custom scanner from scratch. Depending on the volume I would at least look into contracting a company that makes scanners to make a custom model meant to fit in place of the old one. I would guess that trying to design something like that from scratch would require a lot of custom mechanical components which would get very expensive in small quantities. I don't see how you'd beat a company that already makes this sort of thing. Surely there are OEM scanner engine type modules meant for integration into customer equipment. Similar to the way early laser printers almost all had Canon engines inside.

[gaminn]

Seems like adapting the machine would be easier than trying to design a custom scanner from scratch. Depending on the volume I would at least look into contracting a company that makes scanners to make a custom model meant to fit in place of the old one. I would guess that trying to design something like that from scratch would require a lot of custom mechanical components which would get very expensive in small quantities. I don't see how you'd beat a company that already makes this sort of thing. Surely there are OEM scanner engine type modules meant for integration into customer equipment. Similar to the way early laser printers almost all had Canon engines inside.

I don't think my client would pay less for adapting an existing scanner. He is also quite happy he has his own solution which he could easily adapt to his needs in future.

You are right there are custom mechanical components which are not very cheap. The price to produce this scanner is same as buying commercial scanner they used before.

[ebastler]

Hey guys, give me some numbers. I will average them and from statistical point of view, we should get the correct number. You have all information I had when I started the project.

$20k

Edit: Yours was probably not the best way to ask. "Give me some estimates, but send them to me via PM, don't post them in this thread" would have worked better, to avoid biasing others by the earlier posts. Too late now... 

[Tomorokoshi]

Production ready: $100,000.

1000 hours at $100 per hour. That is, a 6-month full-time project. And this is ambitious.

Technical areas include: Software, electronics, mechanics, optics, machine integration, etc. EMC? Other approvals?

[gaminn]

Thanks all for their comments. I asked approx. 10 000 USD for this project. Next time, I wouldn't go under 15 000 USD (but then the question is whether my client would start the project...).  And I live in the Czech republic.

[JVR]

And how long did it take to complete?

[gaminn]

Approx 3 months of pure time.

[janoc]

We didn't try usb cameras. There is a limited space available over and under the card in the machine, now there is 6 cm (but could be a little more). What distance would two (top and bottom) standard usb cameras need to output sharp image without distortion? And I also think it would not be easy to perfectly glue at least two video frames (it is not possible to take just one frame as part of the card is always hidden behind mechanism of the scanner) to obtain complete image of the card and not to distort any characters on the card (OCR is then used). And realtime. Another delay of one or two seconds is not very good. But I'm not expert in this area.

Distortion is not a big deal if you have a decent quality lens that doesn't have other problems (heavy chromatic aberration, for ex) - just calibrate the lens using e.g. a checkerboard pattern (do once and then store the calibration in a file) and then undistort images in software. The same could be done to deal with keystone correction if the camera isn't exactly perpendicular. If there is PC involved you have plenty of computational power to do this type of processing - it moves the complexity from hardware to software which may be easier to deal with. Libraries like OpenCV are a popular solution for this type of stuff.

The distance problem can be handled e.g. using a wide-angle (fisheye) lens and/or a mirror. Both are fairly common tricks in these situations.

[Kilrah]

I'd have said $30k.

[gaminn]

Distortion is not a big deal if you have a decent quality lens that doesn't have other problems (heavy chromatic aberration, for ex) - just calibrate the lens using e.g. a checkerboard pattern (do once and then store the calibration in a file) and then undistort images in software. The same could be done to deal with keystone correction if the camera isn't exactly perpendicular. If there is PC involved you have plenty of computational power to do this type of processing - it moves the complexity from hardware to software which may be easier to deal with. Libraries like OpenCV are a popular solution for this type of stuff.

The distance problem can be handled e.g. using a wide-angle (fisheye) lens and/or a mirror. Both are fairly common tricks in these situations.

Good points, thanks. How much you think would be a suitable USB camera?

Illumination could also be challenging. The cards are shiny.

[janoc]

Good points, thanks. How much you think would be a suitable USB camera?

Difficult to say, the largest cost will be the lens. The cameras themselves are cheap compared to lenses. You can get some prices here, for ex.:
https://www.ptgrey.com/cameras

Of course, you can always buy a Chinese board camera from Alibaba/AliExpress which is going to be much cheaper - possibly good enough for a one-off. But if you need more of these, then better go with a vendor like Point Grey.

E.g. these Firefly cameras:
https://www.ptgrey.com/firefly-mv-usb2-cameras

But there are a lot of board camera vendors - pick based on your requirements and budget. A prototype can be built even from two webcams ...

Illumination could also be challenging. The cards are shiny.

Angle the light source so that the reflection isn't going back into the camera. And/or use diffuse lighting, e.g. a panel similar to the LCD backlights.

[gaminn]

We are talking about a device with manufacturing cost of 80 - 90 USD of electronic part (image sensors, illlumination, motor, electronics). These machine vision cameras are out of the budget. If the client would consider going this way then the price of both cameras must be well under 50 USD. I guess webcam or some chinese camera module is the only solution but then resulting image quality would be questionable.

[janoc]

We are talking about a device with manufacturing cost of 80 - 90 USD of electronic part (image sensors, illlumination, motor, electronics). These machine vision cameras are out of the budget. If the client would consider going this way then the price of both cameras must be well under 50 USD. I guess webcam or some chinese camera module is the only solution but then resulting image quality would be questionable.

Not necessarily, the Chinese cameras often use the same sensors, etc. Of course, you will need to find one that actually has a decent lens (or provide your own). Moreover, you want to do some sort of OCR/code reading, not photography, so you don't need perfect image quality - as long as you can decode the content.

If you use a contact sensor, you may save a bit on the sensor itself but will need to deal with all the interfacing (those linear CCDs are  quite non-trivial to use) and the mechanics of the scanning itself, requiring a lot more complex setup than a camera would.

[Kilrah]

Remember one of the reasons for in-house dev was avoiding cases where the manufacturer EOLs the whole scanner. You don't want to get yourself into the same issue by designing around a no name Chinese cam that may or may not exist anymore in 6 months, would be even worse.

[JVR]

Approx 3 months of pure time.

That is damn fast, considering you had to do HW as well. That is generally the timeline for a POC with multiple engineers on a medium sized project.

[gaminn]

That is damn fast, considering you had to do HW as well. That is generally the timeline for a POC with multiple engineers on a medium sized project.

The project as a whole took quite a long time to complete. We started the project in 2017 and finished it (almost) now. There were many breaks during the project, at least for me. After POC the client needed some time to get money for the project, several times I was waiting weeks for prototypes of mechanics of the scanner to be designed by the client and delivered to me, etc... I'm just saying I spent <600 hours of pure time with HW, FW and demo SW design.

[janoc]

Remember one of the reasons for in-house dev was avoiding cases where the manufacturer EOLs the whole scanner. You don't want to get yourself into the same issue by designing around a no name Chinese cam that may or may not exist anymore in 6 months, would be even worse.

That's why I was originally recommending Point Grey - there you don't have this problem (but you do pay for it ...). OTOH, with a bit of a smart design you could make the device flexible enough to accommodate many different board cameras and circumvent this problem entirely.

If you use USB to talk to the PC and standard UVC compatible cameras, the rest is mostly mechanics (how to fix the camera in the frame) and lenses (which are mostly standardized). The exact chipset, resolution etc. don't matter that much (beyond some sane minima) if the goal is only to read what is on the the cards. But yes, I probably also wouldn't design a product around a camera from AliExpress unless I had a guarantee from the manufacturer that it will be available for some reasonable time.

On the other hand, if you do this using a bare linear CCD sensor and the sensor becomes unavailable, you are in much deeper kimchi ...

[Mr. Scram]

Hey guys, give me some numbers. I will average them and from statistical point of view, we should get the correct number. You have all information I had when I started the project.

That's a risky approach. I'd just have to suggest a price of fifteen million dollars to ruin that plan.