为什么只导入OpenCV导致大量CPU使用？

Question

I noticed something very odd in trying a motion detector for Raspberry Pi:

Removing the camera logging from the script, makes it use almost 0 CPU:

#from gpiozero import MotionSensor
#import cv2
from datetime import datetime
from time import sleep
#camera = cv2.VideoCapture(0)
#pir = MotionSensor(4, queue_len=2, sample_rate=2, threshold=0.5)
import RPi.GPIO as GPIO
GPIO.setmode(GPIO.BCM)
PIR_PIN = 4
GPIO.setup(PIR_PIN, GPIO.IN)
while True:
    sleep(1)
    if GPIO.input(PIR_PIN):
        print( "detected!")
        filename = 'motionpics/' + datetime.now().strftime("%Y-%m-%d_%H.%M.%S.jpg")
        #ret, frame = camera.read()
        #cv2.imwrite(filename, frame)
        #camera.release()
        #pir.wait_for_no_motion()

However, uncommenting just one line - the import cv2 , makes this script go to 300% CPU Usage!!

What is wrong with OpenCV and why can't I even start to use it to grab usb camera images without it using a bunch of cpu, and wearing down the battery?

Answer 1

Hmmmm, if I am not mistaken opencv needs numpy right? Could you try the following:

$ sudo apt-get install libatlas3-base
$ sudo update-alternatives --config libblas.so.3

choose the libatlas option

$ sudo update-alternatives --config liblapack.so.3

choose the libatlas option

$ sudo aptitude purge libopenblas-{base,dev}

Source

Answer 2

I can confirm that Giannis' answer is correct. I just performed the steps listed in his answer and am able to import cv2 in python 3.4 without the high cpu usage. So at least there is that. I am able to grab a frame and display an image. This works for my use case.

I did notice however that during the aforementioned steps, libtiff5, wolfram, and several other libraries were uninstalled.

If you need these libraries and applications (I do not have a full list at the moment) I would reccomend temporarily NOT performing

Sudo apt-get dist-upgrade

And

Sudo rpi-update

At this time, and remain at raspbian jessie. This is just out of my personal experience.

EDIT:

Also I would like to add that Giannis was right, this is seemingly a numpy issue, and can easily be tested by simply:

going on your Raspberry Pi3's desktop->Start Menu->Code->Python 3; type "import numpy" (without quotes).

You should see your cpu usage go through the roof. This is a way of telling that you are eligible to have this fix work.