blue pedestrian crossing sign detection

idea is to develop algorithm to detect EU pedestrian crossing signs (blue) in images or stream.

requirements

python >2.7, opencv >2.*, python-opencv, numpy

install

1. In ubuntu linux

sudo apt-get update sudo apt-get install libcv2.3 libcv-dev python-numpy python-opencv

usage

prepare file with picture filenames inside with help of command

> ls data/*|grep -i -P (jpg|png)$ > pictures.txt

then run:

> python detect_pedestrians.py 

help of detect_pedestrians.py

please enter text filename with images listed inside to create one you can use something like

ls data/*|grep -i -P (jpg|png)$ > pictures.txt

Usage: detect_pedestrians.py [options] pictures.txt

Options: --version show program's version number and exit -h, --help show this help message and exit -d, --debug show debugging window and do not create blob files -v, --verbose show more information in stdout -c CLASSIFIER, --classifier=CLASSIFIER classifier file name. default: haar_classifier.xml -l DEBUG_LEVEL, --debug-level=DEBUG_LEVEL

you can run program in debug mode to see detected objects in image, if you increse level -l 2 you will even see blue blobs detected

with verbosity you can output more info into console

p.s. debug mode is not creating descriptor files next to images

pipepline

1. detect blue blobs in images by converting image to HLS and extracting tresholded blue color
2. exclude smaller and bigger blobs by ratio of blob area with image area
3. exclude blobs which aspect ratio is more than .49 and less than 1.49
4. detect object with haar like features trained on traffic sign (with min neighbours 2)
5. check for intersected blobs and haar objects (those are what we interested in)