Background Subtraction

import cv2
import numpy as np

def background_subtraction_demo(video_source="video.mp4"):
    """
    Compare MOG2 and KNN background subtractors.
    Both automatically learn and adapt to the scene background.
    """
    cap = cv2.VideoCapture(video_source)
    if not cap.isOpened():
        print("Could not open video")
        return

    # METHOD 1: MOG2 (Mixture of Gaussians v2)
    # Good general-purpose subtractor, handles gradual lighting changes
    mog2 = cv2.createBackgroundSubtractorMOG2(
        history=500,        # Number of frames to build background model
        varThreshold=50,    # Sensitivity threshold
        detectShadows=True  # Detect and mark shadows separately
    )

    # METHOD 2: KNN (K-Nearest Neighbours)
    # Better for scenes with large static objects
    knn = cv2.createBackgroundSubtractorKNN(
        history=500,
        dist2Threshold=400.0,
        detectShadows=True
    )

    frame_count = 0
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))

    while frame_count < 300:
        ret, frame = cap.read()
        if not ret:
            break

        # Apply background subtraction
        fg_mask_mog2 = mog2.apply(frame)
        fg_mask_knn  = knn.apply(frame)

        # Clean masks
        fg_mask_mog2 = cv2.morphologyEx(fg_mask_mog2, cv2.MORPH_OPEN, kernel)
        fg_mask_knn  = cv2.morphologyEx(fg_mask_knn,  cv2.MORPH_OPEN, kernel)

        # Extract moving objects
        foreground = cv2.bitwise_and(frame, frame, mask=fg_mask_mog2)

        # Count moving pixels
        moving_pixels = cv2.countNonZero(fg_mask_mog2)
        total_pixels  = fg_mask_mog2.shape[0] * fg_mask_mog2.shape[1]

        if frame_count % 30 == 0:
            pct = moving_pixels / total_pixels * 100
            print(f"Frame {frame_count}: {moving_pixels:,} moving pixels ({pct:.1f}%)")

        frame_count += 1

    cap.release()
    print("Background subtraction complete")

background_subtraction_demo()