Shapes detector

Detecting figures shapes.

Classical Computer Vision approach.

Algorithm

1. Select ROIs of figures by their color using defined range of values of Hue channel of HSV (Hue Saturation Value) image representation;

2. Remove trash using Morphology operations: dilation and errosion;

3. Use OpenCV findContours method to find contours (obviously);

4. Get convex hull of the shapes;

5. Approximate polygon with cv2.approxPolyDP;

6. Count corners.

Some twist and tweaks are made along the implementation, see the code for more details.

Reproduce

Create virtual environment with pyenv:

$ python3 -m venv env

Activate environment and install dependencies:

$ source ./env/bin/activate
$ pip3 install opencv-python

Place the input video in folder input and run the detector:

$ python3 detect_shapes.py