Motion Detection

Pause inference while the scene is still

On a camera that is mostly watching nothing happen, motion detection pauses inference until the scene changes. Frames are divided into a grid of cells; a cell counts as moved when enough of its pixels change, and the source counts as moving when enough cells have moved.

Field	Meaning	Default	Range
motionDetect	Turn motion gating on	false	—
motionSensitivity	How readily a cell counts as moved	0.5	0.0–1.0
motionThreshold	Fraction of cells that must have moved to count as motion	0.01	0.0–1.0
motionGap	Seconds of stillness before motion is considered over	1	1–60
motionGridX	Grid cells across	10	8–32
motionGridY	Grid cells down	10	8–32

Python

from eyepop.worker.worker_types import MotionDetectConfig

job = endpoint.load_from(
    "rtsp://camera.example.com/stream1",
    motion_detect=MotionDetectConfig(
        motionDetect=True,
        motionSensitivity=0.6,
        motionGap=5,
    ),
)

Node

const results = await endpoint.process({
    source: { url: 'rtsp://camera.example.com/stream1' },
    motionDetect: {
        motionDetect: true,
        motionSensitivity: 0.6,
        motionGap: 5,
    },
})

Raising motionGap keeps inference running through short pauses — someone standing still mid-frame — at the cost of processing more quiet frames. It is the only control over the end of motion: the same value decides when motion counts as finished and when the stillness is reported, with no way to set the two apart. At the start of motion there is no knob at all — movement has to show up in two consecutive frames before it counts, so a single-frame flicker does not wake the source. Motion detection composes with fps: gate on motion, then throttle what gets through.