Autonomous Vision Lab (Sourabh Raikwar)

Autonomous Live CCTV Operator is a real-time, multimodal physical security and operational intelligence platform built for restaurants, retail stores, warehouses, and homes. Traditional CCTV systems suffer from two critical flaws: 98% of feeds go unwatched, and legacy pixel-motion detectors flood users with false alarms from swaying trees, passing pedestrians outside target boundaries, or repeated duplicate alerts for the same seated customer. Our platform combines a 10–20ms client-side Edge Spatial Gating Engine with Meta Muse Spark Open (meta-models/muse-spark-open-1026 on Baseten) and Meta's WhatsApp Cloud API v21.0: 1. Strict Polygon Watch Zone Masking: Operators draw freehand Custom Pencil, Box, or Circle watch zones around specific targets (restaurant patio tables, retail cash register drawers, or vehicle gates). Pixels outside the polygon are masked out before AI verification, eliminating background false alarms. 2. 6x4 Spatial Sub-Section State Locking (Section Notified): Subdivides watch zones into 24 independent spatial cells (SEC [col,row] NOTIFIED). When a guest sits at a restaurant table, that specific table section locks after sending one alert while neighboring empty tables remain 100% active for new arrivals. 3. 10-Second HD Pre-Trigger Temporal Memory: Maintains a rolling 1280x720 circular frame buffer (T-10s to T+3s) so Meta Muse Spark analyzes chronological multi-frame sequences to read moving delivery van fleet numbers and verify cash drawer banknote counting. 4. Instant WhatsApp Dispatch: Automatically uploads binary keyframe snapshots to Meta Graph API and delivers rich visual alerts with AI reasoning and confidence scores directly to the operator's phone.