Detection Modes

Background Subtraction

Best when animals move on stable backgrounds.

What It Means

Foreground is inferred by frame-to-background difference, then morphology/refinement is applied.

Key Controls and Tradeoffs

• SUBTRACTION_THRESHOLD
• Lower: more sensitive, more noise.
• Higher: cleaner, may miss faint targets.
• ENABLE_ADAPTIVE_BACKGROUND
• Helps with slow lighting drift.
• Can absorb stationary animals if too aggressive.
• Morphology (MORPH_KERNEL_SIZE, split/dilation toggles)
• Larger kernels smooth noise but can merge close animals.

Use When

• Arena is static.
• Lighting is controlled or slowly changing.
• Target count is moderate and movement is visible.

YOLO OBB

Best for complex backgrounds or weak motion contrast.

What It Means

A model predicts oriented boxes per frame; detections feed the same tracking pipeline.

Key Controls and Tradeoffs

• YOLO_CONFIDENCE_THRESHOLD
• Lower: catches more objects, includes more false positives.
• Higher: precision improves, recall may drop.
• YOLO_IOU_THRESHOLD
• Controls suppression overlap behavior.
• YOLO_DEVICE, TensorRT options
• Throughput and startup complexity vary by platform.

Use When

• Targets can be stationary.
• Background subtraction is unstable.
• You have a suitable OBB model.

Practical Selection Matrix

Scenario	Preferred Mode	Why
Static arena, moving insects	Background subtraction	Simple and fast
Stationary animals / cluttered scene	YOLO OBB	Learned visual cues
Very large videos with limited GPU	Background subtraction + resize	Better throughput control
Heterogeneous data across setups	YOLO OBB	More robust across conditions