In recent years, due to the continuous development of the transportation industry , Car ownership is rising , Many vehicles are driving , Collisions often occur 、 Rear end collision and other phenomena , Serious circumstances will lead to traffic accidents . And in the investigation and research, it is found that the vehicle with more traffic accidents at present is large buses , It's not just because large buses are relatively large , And its own safety supervision system has not been fully established , This increases the possibility of bus accidents , It poses a very serious threat to people's lives and economy .
2017 year 3 month 7 Japan , The Ministry of transport issued a circular on the implementation of transportation industry standards 《 Safety technical conditions for operating passenger cars 》（JT/T 1094-2016） The notice of .《 Safety technical conditions for operating passenger cars 》 Technical requirements under 4.1.5 Article specifies ：9M The above operating buses are required to be equipped with lane departure warning system （LDWS） And the front collision warning that meets the standard （FCW） function , And in the first place 5 The transition period requirements for the implementation of this standard clearly stipulates 13 A transition period of months .
The continuous development of urban economy 、 The scope continues to expand , Transportation companies will also move towards high-speed development . As the policy level pays more and more attention to public transport safety . Although the vehicle is equipped with on-board video monitoring system , It can realize real-time preview and scheduling of vehicles , But the safety of driving cannot be guaranteed , Driver's operation specifications 、 The driver's driving behavior cannot be supervised , Speeding 、 Tired driving 、 Traffic violations 、 Drive and smoke 、 Inattention 、 The traffic accidents caused by bad behaviors such as making phone calls are as high as 95%, Vehicle loss is normal 5-10 times , Fuel consumption increases 10%-50%.
This project is open source , It is a relatively simple demand analysis , It is only used to detect eye fatigue , I will not open source other requirements for the time being , When you can do this according to business needs analysis , This project mainly gives you a simple guide to think .
Open source face recognition haarcascade_frontalface_default.xml And eye recognition haarcascade_eye.xml, This is a trained model , We can call .
The calling method is also very simple ：
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
eye_cascade = cv2.CascadeClassifier('haarcascade_eye.xml')
The so-called real-time detection , We can simply understand it as a video , The video is frame by frame , You can understand a frame as a picture , We need to test whether a person is tired , Is to detect the person's eye condition in a short time , It is to analyze the changes of eye pictures in each frame during this period , Set a standard to judge whether you are tired , This standard has been tested .
To turn on the camera in real time , Use the following function ：
cap = cv2.VideoCapture(0)
The transmission parameter is 0 Represents the use of a camera , Because our notebook has only one camera , If the parameter is 1 It uses two cameras , I am a notebook , So I pass on parameters 0, Of course, you can also import videos here , You want to check whether the driver in this video is tired , Instead of real-time detection .
About opencv I've talked about getting started with video processing ：opencv Introduction to video processing You can further modify and understand according to this .
import cv2
from functools import wraps
from pygame import mixer
import time
I don't need to teach module download ？ The foundation is not proficient , Please see my basic column ：python Basic course
def counter(func):
@wraps(func)
def tmp(*args, **kwargs):
tmp.count += 1
global lastsave
if time.time() - lastsave > 3:
# This is in seconds , therefore 5 minute = 300 second
lastsave = time.time()
tmp.count = 0
return func(*args, **kwargs)
tmp.count = 0
return tmp
Meet the output of eye closure ：
@counter
def closed():
print ("Eye Closed")
Similarly, we also define an output function ：
def openeye():
print( "Eye is Open")
The alarm sound also defines a function ：
def sound():
mixer.init()
mixer.music.load('sound.mp3')
mixer.music.play()
use while Loop real-time detection , Grayscale the image ,detectMultiScale Determine the target size , The next step is to circle the eyes , If you have a foundation , It's not too difficult to understand him .
while 1:
ret, img = cap.read()
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray, 1.3, 5)
for (x, y, w, h) in faces:
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
roi_gray = gray[y:y + h, x:x + w]
roi_color = img[y:y + h, x:x + w]
eyes = eye_cascade.detectMultiScale(roi_gray)
if eyes is not ():
for (ex, ey, ew, eh) in eyes:
cv2.rectangle(roi_color, (ex, ey), (ex + ew, ey + eh), (0, 255, 0), 2)
openeye()
else:
closed()
if closed.count == 4:
print ("driver is sleeping")
sound()
cv2.imshow('img', img)
k = cv2.waitKey(30) & 0xff
Overall evaluation , The whole structure of judging fatigue is not very reasonable , It looks a little rough , This project is just a guide for everyone , If you want to do a graduation project or something , This alone is definitely not enough .
''':cvar
official account ： Play with big data
I wechat ：hxgsrubxjogxeeag
python Group ：428335755
'''
# coding=utf-8
import cv2
from functools import wraps
from pygame import mixer
import time
lastsave = 0
def counter(func):
@wraps(func)
def tmp(*args, **kwargs):
tmp.count += 1
global lastsave
if time.time() - lastsave > 3:
# This is in seconds , therefore 5 minute = 300 second
lastsave = time.time()
tmp.count = 0
return func(*args, **kwargs)
tmp.count = 0
return tmp
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
eye_cascade = cv2.CascadeClassifier('haarcascade_eye.xml')
cap = cv2.VideoCapture(0)
@counter
def closed():
print("Eye Closed")
def openeye():
print("Eye is Open")
def sound():
mixer.init()
mixer.music.load('sound.mp3')
mixer.music.play()
while 1:
ret, img = cap.read()
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray, 1.3, 5)
for (x, y, w, h) in faces:
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
roi_gray = gray[y:y + h, x:x + w]
roi_color = img[y:y + h, x:x + w]
eyes = eye_cascade.detectMultiScale(roi_gray)
if eyes is not ():
for (ex, ey, ew, eh) in eyes:
cv2.rectangle(roi_color, (ex, ey), (ex + ew, ey + eh), (0, 255, 0), 2)
openeye()
else:
closed()
if closed.count == 4:
print("driver is sleeping")
sound()
cv2.imshow('img', img)
k = cv2.waitKey(30) & 0xff
cap.release()
cv2.destroyAllWindows()
