Video Forensic Analysis
Mark specific areas of the frame and detect the exact moment when there is motion inside them. Each event is labeled with AI to identify what caused it: a person, a vehicle, an animal…
The video is processed in your own browser. It is never uploaded to any server.
Select the video
MP4, WebM, MOV…Outline the areas to watch
Zones 0
No zones yet. Draw a polygon over the video or use “Full frame” to analyze the whole image.
Analyze
Higher sensitivity catches subtle motion, but also shadows, rain or camera noise.
Frames analyzed per second of video. Higher sampling = more temporal precision, but the analysis takes longer. When WebCodecs is available the whole clip is decoded at high speed; at 1 fps only keyframes are decoded, which is fastest.
Preparing… 0%
Detected events 0
- Events
- 0
- Time with motion
- 00:00.000
- Frames analyzed
- 0
- Zones
- 0
How the analysis works
- When your browser supports WebCodecs, the video is decoded straight through at high speed and sampled frame by frame — no seeking. Otherwise it falls back to exact time seeks (
currentTime), so the result stays reproducible. - Each frame is scaled down to 320 px wide and converted to luminance (Rec. 601), which dampens color and compression noise.
- It is compared with the previous frame pixel by pixel, but only inside the masks of your polygons. If the share of changed area exceeds the sensitivity threshold, that instant is flagged as motion.
- At 1 fps sampling only keyframes are decoded (they are self-contained), skipping the in-between frames for a large speed-up on long clips.
- Consecutive instants are grouped into an event; it closes after 1 second without motion, and its moment of peak intensity is stored.
- If you enable AI, the peak frame of each event is run through COCO-SSD and the detections whose box overlaps the zone are kept.