Search code examples
opencvraspberry-piraspbian

Pi live video color detection

I'm planning to create an ambilight effect behind my TV. I want to achieve this by using a camera pointed at my TV. I think the easiest way is using a simple ip-camera. I need color detection to detect the colors on the screen and translate this to rgb values on the led strip.

I have a Raspberry Pi as hub in the middle of my house. I was thinking about using it like this

Ip camera pointed at my screen Process the video on the pi and translate it to rgb values and send it to mqtt server. Behind my TV receive the colors on my nodeMCU.

How can I detect colors on a live stream (on multiple points) on my pi?

Solution

  • If you can create any background colour the best approach might be calculating k-means or median to get "the most popular" colours. If the ambient light can be different in different places then using ROI at the image edges you can check what colour is dominant in this area (by comparing number of samples of different colours).

    If you have only limited colours (e.g. only R, G and B) then you can simply check which channel has highest intensity in desired region.

    I wrote the code with an assumption that you can create any RGB ambient color.

    As a test image I use this one: Input image

    The code is:

    import cv2
import numpy as np

# Read an input image (in your case this will be an image from the camera)
img = cv2.imread('saul2.png ', cv2.IMREAD_COLOR)

# The block_size defines how big the patches around an image are
# the more LEDs you have and the more segments you want, the lower block_size can be
block_size = 60

# Get dimensions of an image
height, width, chan = img.shape

# Calculate number of patches along height and width
h_steps = height / block_size
w_steps = width / block_size

# In one loop I calculate both: left and right ambient or top and bottom
ambient_patch1 = np.zeros((60, 60, 3))
ambient_patch2 = np.zeros((60, 60, 3))

# Create output image (just for visualization
# there will be an input image in the middle, 10px black border and ambient color)
output = cv2.copyMakeBorder(img, 70, 70, 70, 70, cv2.BORDER_CONSTANT, value = 0)

for i in range(h_steps):
    # Get left and right region of an image
    left_roi = img[i * 60 : (i + 1) * 60, 0 : 60]
    right_roi = img[i * 60 : (i + 1) * 60, -61 : -1]

    left_med = np.median(left_roi, (0, 1)) # This is an actual RGB color for given block (on the left)
    right_med = np.median(right_roi, (0, 1)) # and on the right

    # Create patch having an ambient color - this is just for visualization
    ambient_patch1[:, :] = left_med
    ambient_patch2[:, :] = right_med

    # Put it in the output image (the additional 70 is because input image is in the middle (shifted by 70px)
    output[70 + i * 60 : 70+ (i + 1) * 60, 0 : 60] = ambient_patch1
    output[70 + i * 60 : 70+ (i + 1) * 60, -61: -1] = ambient_patch2


for i in range(w_steps):
    # Get top and bottom region of an image
    top_roi = img[0 : 60, i * 60 : (i + 1) * 60]
    bottom_roi = img[-61 : -1, i * 60: (i + 1) * 60]

    top_med = np.median(top_roi, (0, 1)) # This is an actual RGB color for given block (on top)
    bottom_med = np.median(bottom_roi, (0, 1)) # and bottom

    # Create patch having an ambient color - this is just for visualization
    ambient_patch1[:, :] = top_med
    ambient_patch2[:, :] = bottom_med

    # Put it in the output image (the additional 70 is because input image is in the middle (shifted by 70px)
    output[0 : 60, 70 + i * 60 : 70 + (i + 1) * 60] = ambient_patch1
    output[-61: -1, 70 + i * 60 : 70 + (i + 1) * 60] = ambient_patch2

# Save output image
cv2.imwrite('saul_output.png', output)

    And this gives a result as follows: Output image

    I hope this helps!

    EDIT: And the two more examples: Example1 Example2