🎩 Background Subtraction

Background subtraction is the process of separating moving foreground objects from a static background in a video.

“What’s new or different in this frame compared to what’s usually there?”

---

🧠 What Is Background Subtraction?

It’s often used in:

• 🚗 Traffic monitoring (detect vehicles)
• 🚶 Security cameras (detect people)
• 🎯 Motion tracking (track movement without full object recognition)

---

⚙️ How It Works (Conceptually)

1. Build a background model over time from the video.
2. Compare each new frame to that model.
3. Difference = moving foreground (e.g. a person walking in front of a static wall).

---

✅ Built-in Background Subtraction Algorithms in OpenCV

OpenCV provides ready-to-use models:

Method	Description
cv2.createBackgroundSubtractorMOG2()	Gaussian mixture model; handles shadows
cv2.createBackgroundSubtractorKNN()	Uses k-Nearest Neighbours for segmentation

---

🧪 Example Code: Background Subtraction on baby.mp4

import cv2
import matplotlib.pyplot as plt

# Open the video
cap = cv2.VideoCapture("baby.mp4")

# Create background subtractor
backSub = cv2.createBackgroundSubtractorMOG2()

# Step 1: Warm up the model with initial frames
for _ in range(30):        # feed first 30 frames into the model
    ret, frame = cap.read()
    if not ret:
        break
    backSub.apply(frame)   # update model, don’t use mask yet

# Step 2: Seek to the target frame (frame 40)
cap.set(cv2.CAP_PROP_POS_FRAMES, 40)
ret, frame = cap.read()
if not ret:
    print("❌ Couldn’t read target frame")
    cap.release()
    exit()

# Step 3: Apply background subtraction on this frame
fg_mask = backSub.apply(frame)

# Step 4: Show result
plt.figure(figsize=(10, 4))

plt.subplot(1, 2, 1)
plt.imshow(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
plt.title("Original Frame")
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(fg_mask, cmap="gray")
plt.title("Foreground Mask (MOG2)")
plt.axis("off")

plt.tight_layout()
plt.show()

cap.release()

📋 What This Code Does

Step	Purpose
1	Feeds 30 initial frames to learn the “background”
2	Seeks to a later frame with potential motion
3	Applies apply() — now it knows what’s “background”
4	Displays the result — moving objects appear as white