The Pi camera is an argument that can take some posts.
The hw installation has been pretty simple , also for me.I follow the indications reported in the https://www.raspberrypi.org/help/camera-module-setup/ and https://www.raspberrypi.org/documentation/usage/camera/README.md.
Then I tested it worked using raspivid: https://www.raspberrypi.org/documentation/usage/camera/raspicam/raspivid.md
So, ready to develop my camera.py class.
- stream images on a browser
- recognize items in the image (ball tracking, as first attempt)
- stream the result in the browser
- run the camera code in a parallel thread
My camera.py module includes 4 objects (classes):
- camera: dedicated to collect images from camera and elaborate this information, in particular performing the ball tracking feature.
- camserver : used to serve the images to the browser.It control the camHandler
- camHandler: the webserver class that pack data and return it to browser.
- camera_data: the class including the data to share ( camera parameters , results, images.
It first inits the camera by :
self.camera = picamera.PiCamera()
Then it is possible to start the man loop of the camera by camera.start().
This loop include a call to :
- the procedure self.configure(). Here it is possible set up the camera parameters by writing a command in the variable self.data.config. It is useful for tuning the thresholds for colors during the ball tracking.
- the self.camera.capture(rawCapture, format=’bgr’, use_video_port=True). It just the the picture
- finally the self.balltracking(image). where it is done the balltracking and returning some information, for example the position of the ball in the image and the estimated distance of the ball respect the rover.
The balltracking is obtained by using the famous openCV library: http://docs.opencv.org/2.4/
In order to install this library I found quite a lot complicated posts googling around, finally I decide to follow this raw method , that worked (at least for me):
sudo apt-get install python-numpy sudo apt-get install python-scipy sudo apt-get install python-imaging sudo apt-get install libopencv-dev sudo apt-get install python-opencv
The ball tracking is also a well known exercise. I used this blog to start my dev: http://www.pyimagesearch.com/2015/09/14/ball-tracking-with-opencv/
Summing up, the image is first filtered.
Then I decide to search for a green ball, so I setup my HSV parameters (I’ll show how later).
The image than become a black and white image: all the pixel inside my HSV thresholds are set to black , all the rest are set to white.
With cv2.findContours(…) you can get a list of contours of the “black items” in the image.So , look for the biggest one ( I suppose in the image there is only one green item: my ball), find its center calculate its radius, and estimate the distance (by a proportion). Finally add some info and draw some lines.Done!
In the next post I’ll explain how to expose the images on the web and how to find the right thresholds for Hue , Saturation and Value (HSV).