Using OpenCV to detect and count moving vehicles.
You definetly could install OpenCV from source, but it will be more complicated. Using homebrew is a seay way to go.
- Install homebrew (if you don't have it)
`/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"`
or you could find it [here](http://brew.sh)
- Install Python (if you'd like to use system Python, skip this step)
`brew install python`
Open up the `~/.bash_profile` file
`cd ~`
`open -e .bash_profile`
(if it does not exist, create it use `touch .bash_profile`)
and append the following lines to the file:
`export PYTHONPATH=/usr/local/lib/python2.7/site-packages:$PYTHONPATH`
after that, use `which python` to detect the path of current Python
if the output is `/usr/local/bin/python` , then you are indeed using the Homebrew version of Python. And if your output
is `/usr/bin/python` , then you are still using the system version of Python.
actually I think it is OK to use system Python.
- Install numpy
`brew install numpy` or
`brew install homebrew/python/numpy`
then link numpy
`brew link numpy`
- Install OpenCV
`brew tap homebrew/science`
`brew install opencv --with-contrib --with-ffmpeg --with-tbb`
it is important to add the option `--with-ffmpeg --with-tbb` for decoding some format of video
Then it's done!(I'm not sure if there are useless steps here, if so I'll update it)
To test the installation:
`python`
`>>> import cv2`
`>>> cv2.__version__`
it should output `'2.4.13'`
You could slso install OpenCV3 use `brew install opencv3`
- If you can't import cv2, try
ln -s -f /usr/local/Cellar/opencv/2.4.13/lib/python2.7/site-packages/cv2.so /usr/local/lib/python2.7/site-packages/cv2.so
you may need `sudo`
- Update installed packages
`sudo apt-get update`
`sudo apt-get upgrade`
`sudo rpi-update`
- Install OpenCV
`sudo apt-get install libopencv-dev python-opencv`
If you want the latest version of OpenCV you should make it from source. (but I'm lazy :)
To test the installation:
`python`
`>>> import cv2`
`>>> cv2.__version__`
1.Read the video frame by frame.
2.Apply some fileters to the frame(dilation, etc.).
3.Use BackgroundSubtractor to split the foreground from background(white-foreground, black-background).
4.Detect the contours of the foreground(moving objects).
5.Calculate the centroid of each moving object.
6.For each centroid, detect if there's a nearby centroid of the last frame. If so, assign them to the same vehicle.
7.For each vehicle, detect whether it crossed the target line.
Simply run main.py.
...
Camshift is not good for tracking vehicle due to the sunlight is too strong at day time.
Increase the performace:
http://www.pyimagesearch.com/2015/12/21/increasing-webcam-fps-with-python-and-opencv/
Problems:
1. May apply other image filters to increase the accuracy
2. To enhance the performance on pi: resize the size of frames, use multiprocessing...
3. Multithread is not working on Python(espicially CPU bound program), should use multiprocessing to get advantage of ulticore.
4. I'll update how to switch between OopeCV 2 and 3 later
The accuracy of the near lane is around 95 - 100 %, the further lane's accuracy is lower due to the shelter effect.
Usefull links: https://www.learnopencv.com/how-to-find-frame-rate-or-frames-per-second-fps-in-opencv-python-cpp/