Dedicated GPU Processing Pipeline

Upload Video for AI Analysis

Turn perspective surveillance footage into calibrated 2D top-down trajectories, entry/exit volumes, and a natural language scene query model.

Select Footage

Drag and drop your camera recording or browse your computer

Drag & drop your video file here Supports MP4, MKV, MOV, WebM (up to 24 hours / 10 GB)
Don't have footage handy? Test our live pre-analyzed plaza camera right now without uploading.
Open Sample Demo →

Expected Video Quality & Specs

Follow these recommendations to achieve 99%+ trajectory detection precision

Resolution & Framerate 1080p / 15-60 fps

Optimal detection requires 1080p (1920×1080) or 4K (3840×2160) at 15 to 60 FPS. Low framerate timelapse (e.g. 2–5 fps) degrades motion velocity vectors and track continuity.

Camera Elevation & Static Position Elevated & Fixed

Mount the camera 3 to 15 meters above ground looking downward at an oblique angle (30° to 60° inclination). The camera must remain stationary (no panning, tilting, or zooming during the recorded segment).

Lens & Optical Distortion Minimal Fisheye

Standard surveillance lenses (2.8mm, 4mm, 6mm, 8mm) work out of the box. Extreme fisheye (>160° Field of View) or heavy barrel distortion requires optical unwarping for homography alignment.

Ground Reference Points Visible Surface

Ensure clear line of sight to sidewalks, roads, or plaza surfaces where targets enter and exit. Visible ground features (paving tiles, road striping, crosswalks) facilitate high-precision 2D homography calibration.

Clip Duration & Encoding Up to 24h

Supports single clips up to 24 hours in length. Common container codecs: H.264 (AVC) or H.265 (HEVC) with constant framerate.