Foreground mask
Per-frame binary foreground mask separating the animal from the adapted background — usable for downstream analysis.
PNG seriesNPY
← Capabilities
Background Subtraction
Robust animal detection through adaptive background modeling
Three background models — static, MOG2, and KNN — adapt to lighting changes, reject shadows, and maintain detection through gradual environmental shifts without manual threshold adjustment.
3
Background model options
Auto
Lighting adaptation
Yes
Shadow rejection
0
Manual threshold adjustments
The problem
Fixed detection thresholds fail when lighting changes
Standard tracking uses a brightness threshold to separate the animal from the arena floor. When room lights change (door opening, circadian cycle), shadows move, or bedding shifts, the threshold becomes invalid and detection is lost.
- Circadian light/dark transitions invalidate fixed brightness thresholds
- Shadows cast by the animal or experimenter create false detection regions
- Bedding redistribution during the session changes the background appearance
The solution
Continuously adapting background model
ConductVision maintains a background model that updates continuously, separating the static (or slowly changing) arena from the moving animal. Shadow detection classifies shadow pixels separately from animal pixels.
- Three models available: static (fastest), MOG2 (Gaussian mixture), KNN (non-parametric) — choose based on conditions
- Shadow rejection prevents shadows from being classified as part of the animal
- Gradual adaptation tracks slow environmental changes without losing the animal
Endpoints
Detection quality outputs
Detection confidence trace
Per-frame detection confidence score indicating how clearly the animal is separated from background — flags potential tracking issues.
CSV
Background model snapshot
Current background model image for visual verification that adaptation is tracking environmental changes correctly.
PNG
Applications
Variable-condition tracking scenarios
24-hour home cage monitoring
Light/dark cycle transitions change arena illumination dramatically. Adaptive background maintains tracking through the transition.
Measures
- Detection continuity
- Cross-cycle tracking
- Transition recovery time
Light/dark box and CPP
Moving between light and dark compartments changes local illumination. Background subtraction adapts per-region.
Measures
- Compartment detection rate
- Transition accuracy
- Shadow rejection rate
Extended continuous tracking
Bedding shifts, food/water changes, and gradual soiling change the arena appearance over days. Adaptive models maintain detection.
Measures
- Multi-day detection rate
- Background model drift
- Adaptation response time
Variable natural lighting
Field stations and semi-outdoor enclosures experience cloud cover changes and sun angle shifts. KNN model handles non-stationary backgrounds.
Measures
- Outdoor detection rate
- Cloud-transition robustness
- Sun-angle adaptation
Compared to typical systems
How ConductVision differs
| Feature | ConductVision | Typical systems |
|---|---|---|
| Background model | 3 adaptive models (static/MOG2/KNN) | Fixed threshold |
| Shadow handling | Automatic shadow rejection | Shadows detected as animal |
| Lighting change tolerance | Continuous adaptation | Re-calibrate manually |
| Threshold tuning | Not required | Per-session adjustment |
| 24h recording support | Adapts through light/dark cycle | Fails at light transitions |
Related capabilities
Precision Tracking
High-resolution 30 fps tracking that captures sub-second behavioral events conventional systems miss.
Circadian Home Cage Monitoring
Continuous 24h+ home cage recording with automatic light-cycle binning and activity profiles.
Segmentation-Based Detection
Deep segmentation for pixel-precise detection in any contrast condition.
Track through any lighting condition
Upload a recording with changing lighting and see adaptive background maintain detection.
