Why doesn't my program work on my H7 device?

OpenMV Products
1

I created a little program, that detects the color of tokens running past the camera on a conveyor-belt. Depending on the color of the detected chip, I’m then setting the percentage of the pulse of a PWM output. This is then converted into a real analog 0-24V signal.

Everything is working fine when running the program in the OpenMV IDE, but as soon as I save the program to the device and want to run it without, it only seems to be detecting red tokens for everything.

Is there anything that I need to be aware of, when running it without the IDE?

Here’s the code of the program:

import sensor, image, time

from pyb import Pin, Timer

thresholds = [
              # Red
              (70,  80,   10,  30,  50,   60),
              # Blue
              (55,  70, -20,   10,  -65,  -40),
              # Green
              (25, 100,  -69,  -25,  -42,   26),
             ]
global_pixels_threshold = 300
global_area_threshold = 300

NO_TOKEN=0
RED_TOKEN=25
BLUE_TOKEN=50
GREEN_TOKEN=75
DAMAGED_TOKEN=100

# Initialize the output pin
p = Pin('P4') # P4 has TIM2, CH3
tim = Timer(2, freq=1000)
ch = tim.channel(3, Timer.PWM, pin=p)
ch.pulse_width_percent(NO_TOKEN)

# Initialize the image sensor
sensor.reset()
sensor.set_pixformat(sensor.RGB565)
sensor.set_framesize(sensor.QVGA)
sensor.set_auto_gain(False)     # must be turned off for color tracking
sensor.set_auto_whitebal(False) # must be turned off for color tracking
sensor.set_auto_exposure(False) # must be turned off for color tracking
sensor.skip_frames(time = 2000)

while(True):
    color = NO_TOKEN

    # Take a picture and return the image and de-fisheye the image and zoom to the completely black part of the image
    img = sensor.snapshot().lens_corr(strength=1.5, zoom=1.1, x_corr=-0.03, y_corr=0.0)

    # Cut out the part we're interested in (Reduces the false-positives on the sides)
    img = img.crop(roi=[87,0,166,400])

    # Detect Red Tokens
    for blob in img.find_blobs([thresholds[0]], pixels_threshold=global_pixels_threshold, area_threshold=global_area_threshold, merge=True):
        color = RED_TOKEN
        img.draw_rectangle(blob.rect(), color=(255,0,0))

    # Detect Blue Tokens
    if color == NO_TOKEN:
        for blob in img.find_blobs([thresholds[1]], pixels_threshold=global_pixels_threshold, area_threshold=global_area_threshold, merge=True):
            color = BLUE_TOKEN
            img.draw_rectangle(blob.rect(), color=(0,0,255))

    # Detect Green Tokens
    if color == NO_TOKEN:
        for blob in img.find_blobs([thresholds[2]], pixels_threshold=global_pixels_threshold, area_threshold=global_area_threshold, merge=True):
            color = GREEN_TOKEN
            img.draw_rectangle(blob.rect(), color=(0,255,0))

    ch.pulse_width_percent(color)

Would be happy if I could finally finish this little project (Initially I had RED, BLUE and WHITE tokens, but detecting white reliably seems to have been sort of impossilbe, so I painted the white tokens green and now all is good … at least theoretically)

Your help is greatly appreciated.

Chris

2

No not really, there shouldn’t be any difference, not with that script anyway. Maybe the lighting is different where you deploy the cam ? It would change the colors and require different thresholds. Or maybe the script wasn’t fully written and it’s running an older version ? You could also try recording a raw video (see the imageio examples) play it back and run the detection and see what’s going on.

3

I think I’ve solved my problem. Probably the industrial power supply still had enough to not reset the device. After powering off the diver for a longer period of time, the cam started doing it’s job correctly. Sorry for the noise.

4

I bought a really long usb cable because I find it’s too easy to effect the lighting if you don’t have dedicated lighting . This helps me stay away during programming and keep things consistent during production. But nothing beats dedicated lighting.