6
$\begingroup$

I would like to build a motorized robot with stereo vision that sends visual input to a PC over wifi for processing. A Raspberry Pi would have been perfect for the task if it would be possible to connect 2 cameras. The Pi 2 would be powerful enough for two USB webcams (with minimum 8fps) I guess, but the shared USB seems to be a bottleneck (2 cameras + wifi dongle).

What other options are there to send input from 2 cameras and control a motor (or an arduino)?

$\endgroup$
6
  • 1
    $\begingroup$ With two USB webcams you would have to worry about synchronization. $\endgroup$
    – Dima
    Feb 3, 2015 at 20:05
  • 1
    $\begingroup$ have you looked at the Jetson TK1? $\endgroup$ Feb 6, 2015 at 19:44
  • $\begingroup$ You should consider an ODROID U3 from Hardkernel. I'm using them for stereo vision on MAVs and they work great. $\endgroup$
    – abarry
    Feb 20, 2015 at 6:18
  • $\begingroup$ @abarry: Do you do your stereo matching on the U3 directly or do you send the image frames elsewhere for processing? The board is rather weak to the the processing itself, right? I did not find it to support OpenCL for GPGPU. $\endgroup$
    – Kozuch
    Apr 7, 2015 at 9:15
  • 1
    $\begingroup$ (sorry, I can't make comments) @Kozuch: all the processing is on the ODROID-U3. You can expect to get 5-10fps on 320x240 stereo USB cameras (black and white) in OpenCV without parallelizing any of the code. You can read about how we get 120 fps (8.3ms / frame) in our ICRA 2015 preprint below. The USB bus is sufficient for those frame sizes and rates with two cameras. Overall, the ODROID-U2/U3 will preform much better than a Pi. Andrew J. Barry and Russ Tedrake. Pushbroom stereo for high-speed navigation in cluttered $\endgroup$
    – abarry
    Apr 8, 2015 at 4:52

6 Answers 6

5
$\begingroup$

You probably are asking too much of inexpensive components.

The raspberry Pi does not have 2 camera connections, but its brother, the Pi Compute board does.

enter image description here

http://makezine.com/2014/11/03/stereo-depth-perception-with-raspberry-pi/

Even then, you will have to write optimized GPU Code in assembler to get anything near acceptable performance if you try to do it all on the Pi.

I doubt there will be any cpu cycles left to do anything else.

For now I believe the best inexpensive solution remains the Kinect. I've seen some recent new products described, but no news of how they work for robotics (Intel RealSense, Creative Senz3D).

Wait for something new or spend more money or use the Kinect.

$\endgroup$
1
$\begingroup$

The lowest cost options for stereo vision that I have used are the Nvidia K1 boards (around 200$), with integrated GPU that can be taken advantage of by the OpenCV GPU libraries I can get 30+ fps with 2 720p streams doing just stereo depth mapping.

$\endgroup$
1
$\begingroup$

I'm working in a project where I'm using a Netduino+2 and 2 Cameras (maybe it will be more than that), I'm working with I2C. I can connect all the cameras I want.

I don't know if it is an option for you, because in my case I only need positions. The image processing is on the camera side.

$\endgroup$
1
$\begingroup$

If you're working with the K1 and stereovision you might be interested by this http://jetsonhacks.com/2015/03/25/stereolabs-zed-camera-3d-depth-sensing-gtc-2015/

$\endgroup$
1
$\begingroup$

Not sure how well this will run on a Raspberry Pi but the Zed stereo camera does most of the processing istelf with makes it ideal for this type of thing.

https://www.stereolabs.com/

$\endgroup$
2
  • $\begingroup$ Doesn't the Zed explicitly require CUDA capability? $\endgroup$ Mar 3, 2017 at 4:56
  • $\begingroup$ Ah, yes it does. Thanks for pointing that out. I'm running the ZED from a NVidia Jetson TX1 which has CUDA. I don't think the ZED would work on a raspberry pi. $\endgroup$
    – NickF
    Mar 3, 2017 at 19:00
0
$\begingroup$

If you're working with the TK1 and stereo camera, you might consider this:

https://devtalk.nvidia.com/default/topic/994170/jetson-tk1-and-stereo-vision/

These stereo cameras are supports ROS: http://rosindustrial.org/news/2016/1/13/3d-camera-survey

$\endgroup$

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.