Lazy question! Is anyome here using simple "video capture" from a webcam?

[villares]

It was great fun and huge impact to use in class the Processing video library in class, and then with very few clicks on the Processing IDE, the OpenCV for Processing by Greg Borenstein & Florian Bruggisser.

Sadly, video & video capture in Processing Python mode started to fail in different platforms (much like Sound and Serial*), and, well, I don't want to use Processing Python mode in class anymore.

* I was able to re-write my Serial/Firmata projects with pyFirmata and it was ok.
I know there are people working on Sound.

So, before I put energy on this, has anyone tried something on this direction?

• reference: ASCII camera example with OpenCV capture + PySympleGUI

[villares]

This seems to work...

import cv2
import numpy as np
#from PIL import Image
from py5 import create_image_from_numpy

cap = cv2.VideoCapture(0)

def setup():
    size(640, 480)

def draw():
    ret, frame = cap.read()
    #pil_img = Image.fromarray(frame)     # from npa to PIL.Image
    ## if you want to use PIL manipulation here
    #img_as_np_array = np.array(pil_img)  # back to npa
    img_as_np_array = frame
    gray = cv2.cvtColor(img_as_np_array, cv2.COLOR_BGR2GRAY)
    edges = cv2.Canny(gray, 100, 80)
    # blended_gray = cv2.addWeighted(gray, 0.6, edges, 0.4, 0)
    edges_rgb_npa = cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR)
    blended_rgb_npa = cv2.addWeighted(img_as_np_array, 0.6,
                                      edges_rgb_npa, 0.4, 0)
    py5_img = create_image_from_numpy(blended_rgb_npa, 'RGB')
    image(py5_img, 0, 0)
     
def exiting():
    print('over and out')
    cap.release()

[villares]

Works great, but I don't see that much difference (both ways about 30 FPS), but then maybe my computer & camera are slow...

# Updated using hx2A's tip from:
# https://github.com/py5coding/py5generator/discussions/206#discussioncomment-4329841

import cv2
import numpy as np
from py5 import create_image_from_numpy

cap = cv2.VideoCapture(0)
py5_img = None

def setup():
    size(640, 480)

def draw():
    global py5_img
    if frame_count % 30 == 0:
        window_title(f'FR: {get_frame_rate()}')
    ret, frame = cap.read()  # frame is a numpy array
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    edges = cv2.Canny(gray, 100, 80)
    #blended_gray = cv2.addWeighted(gray, 0.6, edges, 0.4, 0)
    edges_bgr_npa = cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR)
    blended_bgr_npa = cv2.addWeighted(frame, 0.6,
                                      edges_bgr_npa , 0.4, 0)
    blended_rgb_npa = cv2.cvtColor(blended_bgr_npa , cv2.COLOR_BGR2RGBA)
    py5_img = create_image_from_numpy(blended_rgb_npa, 'RGB', dst=py5_img)
    # display image
    image(py5_img, 0, 0)
     
def exiting():
    print('over and out')
    cap.release()

Updated with: cv2.cvtColor(blended_bgr_npa , cv2.COLOR_BGR2RGBA)

[hx2A]

Works great, but I don't see that much difference (both ways about 30 FPS), but then maybe my computer & camera are slow...

Your webcam must be 30 fps. If you were using threading and running the Sketch at 60 fps, you might see a small difference.

In any case, that performance feature is there for when you need it.

[villares]

Cheers @hx2A. I now learned that OpenCV captured frames (cv2) are always BGR. So I had to add a BGR to RGB conversion. Something that might be a py5 issue: when I tried using "BGR" as an argument in create_image_from_numpy() it failed silently returning a blank image. Could you please add a warning? (Or maybe add these alternative modes...?)

[hx2A]

Something that might be a py5 issue: when I tried using "BGR" as an argument in create_image_from_numpy() it failed silently returning a blank image. Could you please add a warning? (Or maybe add these alternative modes...?)

This issue came up a month ago in a conversation with @marziabil and I recall I added support for BGR ordering. What version of py5 are you using? Are you using the test version of py5 I gave you or the latest release of py5?

[villares]

It was at my Sesc job's computer, so it must have been an older version. I'll test on the dev/test version then! Very cool!