Detectors

Motion detection is pluggable per camera, so different cameras can use whichever detector fits their scene.

• Census (default) applies a small structural transform to each frame before comparing it to a background model, which makes it resistant to shadow and lighting-driven false triggers that a raw pixel-difference approach is prone to. This is the recommended default for most scenes.
• Frame difference is the simpler pixel-difference approach: fast, but more sensitive to lighting changes and shadows than Census.
• MOG2 is a Gaussian mixture background subtractor, a heavier but more adaptive background model for scenes with more complex, gradually changing backgrounds.
• Optical flow looks at motion vectors between frames rather than static pixel differences, useful for distinguishing genuine directional movement from noise.
• Ensemble combines multiple detectors' judgments rather than relying on one.

All detectors use a decaying-histogram adaptive threshold underneath (see Tuning) that learns each camera's normal background activity level over time, rather than a single fixed sensitivity number that has to be hand-tuned per scene and per time of day.

Choosing one

Census is the right starting point for nearly every camera. Switch only if a specific scene is giving you trouble, a lot of tree or foliage movement might do better with a heavier background model, and a scene where direction of movement matters (a driveway versus a sidewalk, say) might benefit from optical flow. Changing detectors is a per-camera setting, not a stack-wide one.