Using OpenCV with the Raspberry Pi camera

If you’re a Raspberry Pi enthusiast like me, you probably love the Raspberry Pi camera. It delivers pictures in such high quality and speed that it’s practically begging to be used in computer vision contexts.

Unfortunately, that’s not as easy as it sounds. Although the userland library offers some nice utilities for working with the Pi’s camera from the command line, and OpenCV, as a great computer vision library, can easily be installed on the Pi using apt-get, that alone doesn’t solve the problem. The userland programs save encoded pictures to disk, so that if you wanted to work with the pictures using them directly you’d have to take the picture, encode it as a JPG, save it to disk, and then read it from disk with OpenCV and decode it back into a picture. Quite unnecessary.

But there’s good news! And, once again, it shows the power of open source in action. There is now a library that allows you to take pictures with the Pi Cam and get them back as an OpenCV matrix. Let’s take a look at how.

As I was looking for ways of using the Pi Cam with OpenCV, I was daunted at first. It’s great to be able to read the sources for the camera, but that means working a lot with semaphores and doing very, very low-level things. I was grimly prepared to try to use the userland libraries as includes in my own code and make a new library that would return pictures as OpenCV objects when I discovered Pierre Raufast’s blog. He wrote up a way of exposing the camera’s pictures as IplImages, which are OpenCV’s structs in C.

The code was great, but it was buried in executable programs. I was planning on extracting its logic into a library that I could use in my own programs, when I found another post by him saying that it’d already been done by Emil Valkov, based on Pierre’s previous work. Well,  what do you know. I went straight to his blog and checked it out.

As it turns out, he had his code on Github so I forked it and got to work on it. It’s super easy to compile and I was able to very quickly put together a small program that either shows you pictures from the camera or video using the OpenCV highgui modules. But one thing bothered me a bit: The resolution was set really low, to 640×480. As it turns out, that was hard coded into the code. So I modified Emil’s sources, leaving the same interface in place and exposing a second function that allows the user to set the camera capture’s dimensions. The default function just delegates work with default resolution to that function. And now I’ve sent Emil a pull request about it.

What am I trying to say here? Open source is awesome. Three enthusiasts from very different parts of the globe were able to collaborate on a pretty complex technical topic, and it happened completely ad-hoc. Plus, other people can profit from our work (which is mainly Pierre’s and Emil’s work, I just changed a teeny tiny bit of it).

So if you’re wanting to use the Raspberry Pi’s awesome camera with OpenCV and would rather write C++ than C, you can use Emil’s library like this to capture and show a single picture:

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include "RaspiCamCV.h"

int main(int argc, char** argv) {
RaspiCamCvCapture * camera = raspiCamCvCreateCameraCapture(0);
cv::Mat image(raspiCamCvQueryFrame(camera));
cv::namedWindow("Display", CV_WINDOW_AUTOSIZE);
cv::imshow(window_name, image);
cv::waitKey(0);
return 0;
}

 

It’s as simple as that!

About erget

My name’s Daniel Lee. I've founded a company for planning solar power. I've worked on analog space suit interfaces, drones and a bunch of other things in my free time. I'm also involved in standards work for meteorological data. I worked for a while German Weather Service on improving forecasts for weather and renewable power production. I led the team for data ingest there before moving to EUMETSAT.

‹ Syntax highlighting with less

Watching Netflix natively in Linux (OpenSUSE) ›

Tagged with: computer vision, OpenCV, raspberry pi
Posted in Uncategorized