EEVblog #817 - Mantis Elite Cam

[EEVblog]

Dave checks out the Mantis Elite Cam from Vision Engineering
An upgrade to his existing Mantis Elite, but with built in HD USB camera. Is it any good?
https://www.hawkerrichardson.com.au/optical-inspection/optical-inspection-range/mantis-elite-cam
http://www.visioneng.com/products/stereo-microscopes/mantis-elite-cam-hd-with-camera

Update: More powerful measurement software is an option:
http://www.visioneng.com/products/software/dimensionone

[lukier]

With all the praise you give to Vision Engineering I expected something better. Better sensors, maybe stereo. With clever control of the LED lights they could have done some photometric stereo reconstruction etc. The USB cable is just stupid.

If I get this correct old Mantis head is $1885 and the new one with the camera is $3495. That is a massive difference and all they did is they added extra optical path and plugged there off the shelf IDS uEye UI-155x-LE camera. Not your everyday made in China webcam but not really the proper machine vision camera either, IDS makes better ones. This is probably just CY7C68013A (as these USB 2.0 cameras often are) + MT9D131, 2 MP rolling shutter sensor (that is visible when you wobble the arm), not high speed or anything special. Cheapskates.

But to their defence you need to read a bit on machine vision. You cannot measure anything with a camera without calibration. Every time you unscrew the lenses you should recalibrate. If Mantis has variable zoom/focus (I don't know, don't have one) they it is getting even more complicated.

Only when you calibrate the camera you can measure the angles between pixels. To get proper distance/length measurements you need to know the depth - how far your PCB is from the Mantis Head. For that you need some form of 3D reconstruction. Stereo-vision, photometric or structured light (like old Kinect).

So if you want to take proper measurements you are better with the Leica system you showed some time ago. I think Leica has much more experience in vision based measurement systems.

[CrashO]

If Mantis has variable zoom/focus (I don't know, don't have one) they it is getting even more complicated.

Mantis lenses are fixed. Both zoom and focus.
You can replace the lenses with ones offering different "zoom" levels (2x, 4x, etc) but the lenses themselves are just screw in bricks without any wheels or knobs.
Focusing is done by moving the entire mantis head up or down.

[Rasz]

cropped laggy rolling shutter usb 2.0 piece of shit, can be yours for only $1700 extra! 

17 fps lol, if its useless for live video then what is it good for? taking static pictures? you can do that already with $200 camera at same zoom levels

Dave too bad you didnt compare it to tagano(sp?) camera

Louis summed it up nicely:



ps: iiyama(and nec) are huge in medical and scientific, 10bit color is important for things like xray inspection
but in recent years iiyama started pumping out TN garbage monitors at VA/IPS prices, it looks like you got one of those TN trash

[Arjan Emm]

On this site. It appears the webcam add on comes for free If you buy a whole unit.
http://www.gotopac.com/vision-mantis-elite.html

[vze1lryy]

cropped laggy rolling shutter usb 2.0 piece of shit, can be yours for only $1700 extra! 

17 fps lol, if its useless for live video then what is it good for? taking static pictures? you can do that already with $200 camera at same zoom levels

Dave too bad you didnt compare it to tagano(sp?) camera

Louis summed it up nicely:



ps: iiyama(and nec) are huge in medical and scientific, 10bit color is important for things like xray inspection
but in recent years iiyama started pumping out TN garbage monitors at VA/IPS prices, it looks like you got one of those TN trash

Yep those were my concerns in that video. I got excited when I saw Dave's video, I thought they listened and made a new one! No such luck.

USB 2.0 is over ten years old. Real world sustained throughput of about 37 megabytes per second. This is not for high quality inspection cameras.

I gave up and will stick with my NH1080 optixcam and Omano 2300s for recording live video. I hope sometime before 2025 Vision Engineering decides to catch up with the rest of the world and stop using a protocol best suited for mice.

Keep in mind my NH1080 optixcam with the lens adapter for the 2300S I have was approximately $800. They are charging $1700 more - this is $900 more than my camera setup, for a USB 2 piece of shit! Lots of profit in selling that terrible obsolete camera setup.

[Jeroen3]

Those IDS people make really nice camera's. Definitely not "ordinary". We've used them for vision algorithms at university. Low noise and stuff, much much better than an higher mp webcam. If they didn't get a custom made camera, you can probably replace it by a ethernet of firewire version. Which is the same brick box, with a different set of plugs on the back.

The plugs used on the outside of the mantis looks like a Binder Snap-In.
http://www.binder-connector.nl/nl/producten/subminiature-connectors/snap-in-connectors-ip-40/
Really nice and rigid connector, much better locking in than any USB.
I can see why they used those over USB. Mechanically speaking, USB is not that great.

[f4eru]

Dave, every time you say "this xxxxxx is the duck's guts", i think you should illustrate the point with that Electrolytic capacitor in a duck

[lukier]

Mantis lenses are fixed. Both zoom and focus.
You can replace the lenses with ones offering different "zoom" levels (2x, 4x, etc) but the lenses themselves are just screw in bricks without any wheels or knobs.

That simplifies things a bit, but if you replaced the lens and screwed it in the assembly with half a turn less then before this might cause slight errors. If the lenses are good quality the radial and tangential distortions should be small and it should hold calibration for a while.

I have experience with some crazy optics that changes calibration with temperature (camera heats up the lens or the sensor itself is soldered slightly tilted - good manufacturers specify this, and with different thermal expansion of the PCB and the sensor ceramic it moves a bit - doesn't take much to move things few micrometers and create > 1 pixel shifts).

Focusing is done by moving the entire mantis head up or down.

Fixed focus? Moving mantis head up and down? Lol, doesn't that defy the purpose of having comfortable posture? It would imply that I have to bend my neck/spine as I move mantis up and down to focus the PCB. Maybe the depth of field is good enough so this is not an issue.

Those IDS people make really nice camera's. Definitely not "ordinary". We've used them for vision algorithms at university. Low noise and stuff, much much better than an higher mp webcam. If they didn't get a custom made camera, you can probably replace it by a ethernet of firewire version. Which is the same brick box, with a different set of plugs on the back.

Sure, IDS makes various cameras. Used one of them (MT9V034 based) for my master thesis many years ago. Nowadays I mostly use PointGrey or Basler. These companies usually use Cypress EZ-USB -> Sensor in the low end series and in the higher end models it is usually something like USB 3.0/GigE/FireWire transceiver -> FPGA -> Sensor. There are various nice features implemented there, varying among the product ranges and manufacturers, but what is the most important is the sensor.

The point is that VisionEngineering went with possibly the cheapest one IDS has to offer. Cheap Raspberry Pi camera is only slightly worse (SNR, dynamic range, 1/4" and thus pixel size), but has higher resolution and framerate. This particular IDS camera model is really not much different from what you can find in Aliexpress cheap IP cams, $60 trinocular microscope mounted cameras and some webcams.

Recently, I've heard good stuff about new Sony Pregius sensors http://www.sony.net/Products/SC-HP/sensor/technology/pregius.html.

To be honest it is quite disappointing. I've read so many good opinions about Manis, how it all made with rocket science out of pixie dust, but now it all looks rather unprofessional (cheap sensor, USB 2.0, stupid cable and completely insane pricing).

[AF6LJ]

The video add on looks like a hack at best, the PC isn't really powerful enough for the task at hand, no graphics card to do the heavy lifting.
70+% CPU utilization... Give me a break.
Crap.

[CrashO]

Fixed focus? Moving mantis head up and down? Lol, doesn't that defy the purpose of having comfortable posture? It would imply that I have to bend my neck/spine as I move mantis up and down to focus the PCB. Maybe the depth of field is good enough so this is not an issue.

DOF depends on the lens used. I got a Mantis Compact with a 4x lens on it and if I position it so that my benchmat is out-of-focus but the pcb is just in-focus. The DOF is just enough to be able to read a RJ45 connector mounted on the pcb. Anything higher and I need to refocus. Use a lens with larger magnification and the DOF becomes even smaller.

[AF6LJ]

The video add on looks like a hack at best, the PC isn't really powerful enough for the task at hand, no graphics card to do the heavy lifting.
70+% CPU utilization... Give me a break.
Crap.

A CPU second not used is lost forever.

That's true.
That is really a minimalist computer for that app...

[EEVblog]

But to their defence you need to read a bit on machine vision. You cannot measure anything with a camera without calibration.  much more experience in vision based measurement systems.

Nope. As I said, you should be able to get a decent ballpark calibration.
Fixed magnification lends with a shallow DOF means that any in-focus image should be roughly the same known frame size.
So at least they could have had rough settings for all their fixed lenses.

[lukier]

Nope. As I said, you should be able to get a decent ballpark calibration.
Fixed magnification lends with a shallow DOF means that any in-focus image should be roughly the same known frame size.
So at least they could have had rough settings for all their fixed lenses.

With fixed focus and magnification it would be possible in theory, however over the DOF range (let's say 1cm for 4x lens, 30 cm distance) it would cause at least some percents (dunno, 1-5%) of error, and if radial and tangential distortions are not taken into account this would be nonlinear over image area and worse. Also, don't know what's the parameter spread over their lenses. If you buy just the new head and want to use your 10 year old lens it might cause additional trouble. Are such errors acceptable?

Additionally, first scenario would require Vision Engineering to either control the optics across all units and lenses so they can embed fixed camera parameters in the software. They cannot do it as they just provide uEye software.

Another option is embedding camera model parameters in some kind of EEPROM in the camera. This also requires some uniformity over hardware and an additional in-house calibration step at the end of the production line. Here the problem with removable lenses might appear (e.g. using old one). This would also require software support.

Old Kinect uses this route. Camera baselines and focal lengths are embedded in the camera chip. Some of them I suppose are fixed by design (sensors and lenses are rigidly mounted to the frame), some might come from the calibration on the production line (focal length perhaps). The problem is that the camera model the API assumes is very approximate (pinhole) and the parameters values are quite coarse. To get better accuracy you still need to recalibrate your particular camera (and no interchangeable lenses here).

I suppose Mantis is not a precision instrument like Leica, so the measurement function is more of a gimmick. So at least they could have focused instead on sensor resolution/fps/quality/HDMI output etc. This camera as it is now is not an useful feature at all.

[EEVblog]

With fixed focus and magnification it would be possible in theory, however over the DOF range (let's say 1cm for 4x lens, 30 cm distance) it would cause at least some percents (dunno, 1-5%) of error, and if radial and tangential distortions are not taken into account this would be nonlinear over image area and worse. Also, don't know what's the parameter spread over their lenses. If you buy just the new head and want to use your 10 year old lens it might cause additional trouble. Are such errors acceptable?

Like I said, it gets you in the ballpark out of the box!
They also could have included a calibration chart and instructions on how to set it up.

This camera as it is now is not an useful feature at all.

Again, as I said, if you have a requirement for the Mantis, then the ability to get documentation out of it is useful. Saying it is not useful at all is simply not true.

[JWoodrell]

I would think that due to perspective warp and related issues they could implement some sort of uv scanline range finder something like the XBOX one camera sensor works with its dots,  that would let it calibrate its field of view to the perspective.

[rs20]

Was that PCB with the photoimagable silkscreen made by PCBZone/Circuitlabs by any chance?

The 50fps camcorder definitely looks much smoother than the 30fps from the camera! But it still looks like it would be great for videoing demonstrations.

But goodness, has USB camera technology not advanced a single inch since 10 years ago? Why is every USB camera you see barely struggling along at less than 30 fps for anything beyond 720p? Has 1080p30 (USB3 if necessary) not been commoditized yet?

[lukier]

Again, as I said, if you have a requirement for the Mantis, then the ability to get documentation out of it is useful. Saying it is not useful at all is simply not true.

Not a measurement tool, not for video as you in the YT video and vze1lryy mention, so the documentation means still images? For $1800 you can get an entire DSLR kit (body, lens, tripod etc) that will give you beautiful 24 MPix photos and FullHD video.

[Towger]

This video is blocked by youtube on the computer my kids use. Did Dave get too excited taking it apart?

[Smokey]

What's the deal with the AOI functions?  That would be pretty killer if you could do some basic AOI.

[Rasz]

Again, as I said, if you have a requirement for the Mantis, then the ability to get documentation out of it is useful. Saying it is not useful at all is simply not true.

Not a measurement tool, not for video as you in the YT video and vze1lryy mention, so the documentation means still images? For $1800 you can get an entire DSLR kit (body, lens, tripod etc) that will give you beautiful 24 MPix photos and FullHD video.

very much this, I dont know what Vision Engineering was thinking, but this "product" is simply DAMAGING their brand. This is chinese $40 microscope camera level of quality :/

[Kevman]

Has 1080p30 (USB3 if necessary) not been commoditized yet?

It has been. My Logitech C930e will do 1080p30 all day long over USB2.

It requires the software to support decoding the MJPEG compressed video, though (it might be just a different API call into Windows, I don't think its hard) so that might be an issue. It also introduces a delay of a few frames.

Pessimist in me says it might be cost- the C930e costs nearly $100, might be a little rich for Mantis' blood. 

You can get USB3 'webcams' in the form of C-mount machine vision cameras but they typically have a strange resolution and are tough to find in color. They're not crazy expensive, though.

The thing that I hate about webcams is that the cheapest one with optical zoom is $400 for a 3x zoom. A camcorder + HDMI capture card is cheaper. 

[vze1lryy]

Again, as I said, if you have a requirement for the Mantis, then the ability to get documentation out of it is useful. Saying it is not useful at all is simply not true.

Not a measurement tool, not for video as you in the YT video and vze1lryy mention, so the documentation means still images? For $1800 you can get an entire DSLR kit (body, lens, tripod etc) that will give you beautiful 24 MPix photos and FullHD video.

very much this, I dont know what Vision Engineering was thinking, but this "product" is simply DAMAGING their brand. This is chinese $40 microscope camera level of quality :/

They have too many industrial customers who buy without looking at this beforehand to care. It was obvious from the talk I had with the sales rep that someone like me or any home hobbyist engineer is NOT their target customer.

[Howardlong]

On the Amscope I have I use a C mount adapter and 0.5x lens on the trinocular port plus a third party 1080p 50/60fps HDMI camera, chosen for low latency. I capture it with a Startech USB3HDCAP capture device which is reasonable, encoding directly to H.264 at 1080p 50/60fps, and also have a 13" 1080p 50/60fps HDMI monitor directly connected and dedicated to the camera although I much prefer to use the stereo eye pieces to work with. The camera itself is OK but suffers a little from the wobblies if you move things quickly with it, but it's by far the best I've managed to achieve so far for videoing purposes.

I did look at the Mantis offerings, but I just don't have the bench space for one. These look enormous, even the compact one doesn't look that small.

I am surprised that a standard camera port isn't offered on the Mantis. The solution in the video did look like a bit of an afterthought, but I can see why those with deep pockets and plenty of bench space would buy them.

[Psi]

It would be pretty simple for them to sell a specially marked ruler you hold against the PCB for a sec and have it auto calibrate the measuring tool.