Background subtraction is the classical motion detection technique that models static scene appearance and flags pixels that deviate from that model as foreground activity - it is a foundational method for surveillance, traffic analytics, and lightweight video understanding pipelines.

What Is Background Subtraction?

• Definition: Compute difference between current frame and estimated background model to isolate moving objects.
• Core Equation: Pixels with absolute difference above threshold are marked as foreground.
• Model Update: Background is updated gradually to adapt to illumination and long-term scene changes.
• Output: Binary or probabilistic foreground mask per frame.

Why Background Subtraction Matters

• Computational Simplicity: Runs efficiently on edge hardware with low latency.
• Event Triggering: Effective for motion alarms and region-of-interest activation.
• Preprocessing Utility: Provides candidate object regions for heavier detectors.
• Interpretability: Foreground masks are straightforward to inspect and debug.
• Legacy Importance: Still useful in constrained systems and low-compute deployments.

Common Background Models

Running Average:

• Smoothly updates background over time with exponential averaging.
• Good for slowly changing scenes.

Adaptive Median:

• Uses temporal median statistics per pixel.
• More robust to transient motion.

Probabilistic Models:

• Estimate per-pixel distributions for dynamic backgrounds.
• Better for challenging outdoor conditions.

How It Works

Step 1:

• Initialize background model and compute per-pixel difference from current frame.

Step 2:

• Threshold differences to create foreground mask, then refine with morphology and update background model.

Background subtraction is a practical first-line motion isolation tool that transforms raw video into actionable activity masks with minimal compute - it remains valuable whenever speed and interpretability are critical.