How to put OpenCV in Raspberry Pi and easily collect images of face detection results with Python

Introduction

It was quite difficult to collect images of human faces necessary for machine learning, so I made a script to reduce the workload a little.

OpenCV installation

Install with the following command. By the way, it is for Python 2.7.

python

$ sudo apt install libopencv-dev python-opencv

function

If you specify the image URL of the group photo, the image is downloaded from the Internet, face detection is performed, and the image of the detection result is saved.

Loading the library

Load the library you want to use.

python

import cv2
import sys
from PIL import Image
import urllib
import os

Creating a folder for images

Create folders to save the images for detection and results.

python

if os.path.exists("input") == False:
    os.mkdir("input")
    
if os.path.exists("output") == False:
    os.mkdir("output")

Download image data

Saves the image from the specified URL to the input folder. It also gets the name of the image file from the URL.

• In the example, "photo.jpg "

python

url = "http://hoge.com/photo.jpg "

#Get file name
tmp = url.split("/")
imgname = tmp[len(tmp)-1]
imgpath = "input/" + imgname

#Download image data
response = urllib.urlopen(url)

with open(imgpath, "wb") as fout:
    fout.write(response.read())

Image data processing

If it is full color, it takes time to process, so convert it to grayscale.

python

#Image data reading
image = cv2.imread(imgpath)
#Grayscale conversion
image_gray = cv2.cvtColor(image, cv2.cv.CV_BGR2GRAY)

Face detection process

Reads and detects trained data for face detection.

python

#HAAR classifier designation
cascade_path = "haarcascades/haarcascade_frontalface_alt.xml"

#Acquire the features of the cascade classifier
cascade = cv2.CascadeClassifier(cascade_path)

#Detection process
faces = cascade.detectMultiScale(image_gray, scaleFactor=1.1, minNeighbors=1, minSize=(1, 1))

print "Detection result",len(faces),"Man"

Cut and save the face image from the detection result

Cut the face image from the image data before converting to grayscale and save it. At this time, resize the image size to 50x50 so that it is easy to use for machine learning.

python

img = Image.open(imgpath)

for i in range(len(faces)):
    [x,y,w,h] = faces[i]
    imgCroped = img.crop((x,y,x+w,y+h)).resize((50, 50))
    filename = "output/%s_%02d.jpg " % (imgname.split(".")[0], i)
    imgCroped.save(filename)

did it! !! Some things other than faces are detected, but generally satisfied (^-^)

in conclusion

If you ask Google Sensei for a "group photo", you will see a large number of high-resolution photos. I was surprised a little.