safety-helmet-opencv

4th year Image Processing project to identify whether person is wearing safety helmet on construction site or not

Instructions:

Pre-requisites:

OpenCV
NumPy
cvui
easygui
matplotlib

1. Run SHDetector.py with Python 3.8
2. Click Load Image button in the UI
3. Click Safety Helmet Detection
4. Optionally Save the output to a file using Save Image button

Pseudocode:

preProcess():

• Convert to RGB, and remove the noise using cv2.fastNlMeansDenoising

getSkinMask():

• SkinDetector Class
• HSV color segmentation of skin values
• WaterShedding
• otsu thresholding
• return skin mask

getFaces():

• generate sorted contours from the skinMask
• discard any contour bounding rect that we deem too small by comparing its area to the largest area. If an area is ⅕ the size of the largest we discard it.
• We then combine all the clusters of rectangles that are touching each other.
• If a rectangle contains any children then discard them too
• If by this point we have no values from prev functions then we assume our contours correctly picked out the correct faces and reset back to these values instead.
• A face helmet will always be on the upper body in the image, therefore we discard any contours that appear under a contour.
• We now reliably have a face detection but we need to double its height upwards in order to search for the helmet as well.

getHelmets(src_img, skinMask, faces):

• convert the original image to rgb
• get a region of image (ROI) using the inverted skinMask
• get rid of all values that are below 150, above 250 ro get rid of the background
• for each face region determine whether it includes a helmet using processHelmet(face_image) function
• if the function returns true, draw a green rectangle around the face else draw a red one

processHelmet(face_image):

• calculate the area of the face
• find range of predetermined HSV values of the helmets (specific colours)
• if area of the helmet takes more than 39.0 % then return true else false