CODEWITHSUNDEEP
cameracomputer-visionrobotics
Mostafa Khedrawy
Mostafa Khedrawy

Reputation: 11

choosing camera for robot vision system

I am making a robot that navigates based on vision system . vision system use the colour of the object to navigate so how I choose my camera?

enter image description here

Upvotes: 0

Views: 2018

Answers (3)

Vipin Raj C
Vipin Raj C

Reputation: 26

Try Intel Realsense camera! One of the best for development projects like autonomous navigation and deep learning. You can even integrate with ROS.

Upvotes: 0

ScottU
ScottU

Reputation: 75

If you want something simple that locates objects out of the box with little customization, you might try the pixy. It's self contained microprocessor and camera that can be trained to track 7 types of color signatures and 7 color codes. It can track multiple objects at one time returning coordinates for each.

Kickstart Video, Kickstarter, CMUcam5 Pixy, Adafruit

Upvotes: 0

Giuseppe Cotugno
Giuseppe Cotugno

Reputation: 110

If I can give you an advice, be careful on USB controlled cameras as you might quickly get into a bandwidth bottleneck. In other words, the size on the image you need to analyse might be so large that it won't fit on the USB bus (or won't be transmitted fast enough).

Here at work, we use ethernet for transmitting images across and I am also aware that the iCub robot sends images across ethernet as well to other desktop computers which do all the image-processing stuff (and people still complain that it is too slow).

That said, as many other mentioned, the right camera depends on what you want to do and how much you wish to spend. There do exist a lot of robots which perform colour based image processing but such images are also very heavy (3x area of image in pixel) and you might hit that bottleneck pretty fast.

Probably the first thing you want to clarify for yourself is the resolution of your camera. The higher the resolution the more information your algorithm will have (and the more computational CPU power you will need to process it), the lower the resolution the easier it is to process it quickly and the cheaper will be your system (no need of making ad-hoc boards for example).

An higher resolution and better CPU will not automatically give you a better robot if you use your data cleverly. The Nao robot, back in 2009, was using a 320x240 colour camera (like a old mobile phone) interfaced internally to a Geode microcontroller using integrated circuits. Despite the fact that the original image was subsampled, the robot was still able to play football. Of course the code was tightly optimised.

Upvotes: 2

Related Questions