Camera provides striped image - what's wrong?

Question

I've been using the raspi cam for some time now. I was using it inside a machine, to take photos from what's going on there. As far as I can tell, the cables are all attached as they should be - but all of a sudden, the images look like this.

Can anybody help? What's going on here?

In the preview window, the photo looks normal. After a restart, the output is completely black instead.

This is my test code which makes the same output as the production code. First, I measure the lighting and fix the camera setting. Then I take a series of photos while the machine works.

I take the photo in the original cam's format (rgb), crop it with numpy slicing and save it afterwards. That had been the fastest image taking process I was able to come up with.

Around 12 000 photos later, the camera does this. Why? What can I do about it?

from time import sleep
from picamera import PiCamera
import numpy as np
from PIL import Image

CROPBOX = (317, 464, 317+370, 464+251) #x0,y0,x1,y1
RESOLUTION = (1024, 768)

cam = PiCamera(camera_num=0, 
      resolution=RESOLUTION, 
      framerate=40, 
      sensor_mode=0, 
      clock_mode='reset', 
      framerate_range=None)

cam.meter_mode = 'average'
#cam.zoom = 0.3, 0.3, 0.65, 0.65
cam.iso = 50
sleep(0.5)
cam.shutter_speed = cam.exposure_speed
cam.exposure_mode = 'off'
sleep(0.5)

# fix white bracket
gains = cam.awb_gains
cam.awb_mode = 'off'
cam.awb_gains = gains

x0, y0, x1, y1 = CROPBOX
output = np.empty((RESOLUTION[0], RESOLUTION[1], 3), dtype=np.uint8)
cam.capture(output, "rgb")
Image.fromarray(output[y0:y1,x0:x1,:]).save("test_image.png")
cam.close()

Answer 1

Found the error.

I define RESOLUTION here and use it correctly

RESOLUTION = (1024, 768)
cam = PiCamera(camera_num=0, 
    resolution=RESOLUTION, 
    framerate=40, 
    sensor_mode=0, 
    clock_mode='reset', 
    framerate_range=None)

But then in the capture area I use it in the same order, and there it is wrong:

output = np.empty((RESOLUTION[0], RESOLUTION[1], 3), dtype=np.uint8)

Because that's the old numpy trap - images have it the other way round than numpy. I should have used this instead:

output = np.empty((RESOLUTION[1], RESOLUTION[0], 3), dtype=np.uint8)

Because, effectively, numpy slicing uses what is y in images as first argument, and what is x in images as second argument. That's the way numpy is organized.