0
$\begingroup$

I am designing my robot and I have quite design issues with the available stereo cameras on a market. For the prototype I am using ZED M but it is too expensive and quite large.

The question is, do I really need ZED M if I am using my own algorithms to compute the depth? So is having a global shutter, usb 3.0 connection, fixed baseline and wide fisheye lenses will do the job for me?

I am thinking to get the components which are very small and lense+sensor should be no larger that 14mm x 9mm x 5mm and looks like that the sensors are very small with small lenses also on a market.

I am using stereo camera for SLAM and obstalc detection.

$\endgroup$

1 Answer 1

0
$\begingroup$

There are several factors that you should consider before creating your own stereo camera:

  1. How fast do you expect your robot is going to move. This will influence the synchronization requirements. The main advantage of cameras like ZED or RealSense is that they are usually hardware synchronized. If you just use two USB 3.0 cameras they will not be synchronized and that might have a significant impact on the accuracy of your stereo matching.
  2. You have to make sure that the cameras are rigidly fixed to each other. If they are not robot vibrations would invalidate you calibration and that will also have a significant impact on accuracy.
  3. You have to ensure that the cameras are well calibrated. There are methods to do this but this is an additional step that you will have to do by yourself.
  4. Camera parameter control is another issue that you will have to take care of. Usually USB cameras have an automatic camera parameter control algorithm (controlling of exposure time and sensor gain). This is fine for doing general computer vision, but if you have two cameras you have to make sure that both of them take images with identical parameters. If that is not ensured you might end up with situation where one camera sees a white wall and another does not. In this case they would produce images that do not look similar. This will again have significant impact on stereo matching quality. To overcome this you can set one camera as master and control the second camera parameters manually. But that will again introduce latency.

These are the main points that you have to consider when building a stereo camera.

$\endgroup$
2
  • $\begingroup$ Yes, that what I already made in my mind. I will have the sync, calibration and etc., I just do not understand why ZED costs $450 when my team and I can make a production ready stereo camera for much much less as we have our own dense sensing algorithm $\endgroup$
    – Edmund
    May 3, 2019 at 9:35
  • $\begingroup$ I do not know for sure but hardware synchronization is probably one of the reasons. Another is probably high profit margin. Intel cameras are around two times cheaper. $\endgroup$
    – JuliusG
    May 3, 2019 at 9:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.