RICK™ Vision turns existing store cameras into AI-assisted operational intelligence — possible concealment alerts, human review, store memory, and physical retail intelligence. Running live below.
MediaPipe pose skeleton + object detection running on real store CCTV footage — in your browser, no cloud. This is exactly what RICK™ Vision sees in your store.
MediaPipe GPU · PoseLandmarker · EfficientDet · Self-hosted · 30fps
Most camera systems record evidence after the problem. RICK™ Vision is designed to turn store footage into reviewable operational signals while keeping the bar high for false-positive control.
Flags sequences such as hand-to-pocket, hand-to-bag, object disappearance after shelf interaction, and unpaid consumption for human review.
Designed for stores that already have cameras, NVRs, or RTSP feeds. The MVP starts with the hardware already in the building.
Events can attach camera location, time of day, product area, review outcome, and future POS or inventory context.
When multiple cameras are available, the review path can follow the same event across entrance, aisle, cooler, and register views.
Built for real stores: low resolution, odd angles, crowded aisles, occlusion, and older camera systems.
The same store awareness foundation can support fall, safety, compliance, and liability review as the platform matures.
The reference videos proved the product does not need perfect cinematic footage. It needs useful extraction, event context, and careful review language.
One suspicious moment is rarely the whole story. The better review workflow follows the event across camera zones so managers can understand before and after.
Retail footage is often grainy, crowded, angled badly, or partially blocked. RICK™ Vision should suppress weak evidence and escalate only what is worth review.
The first version is built around speed, precision, and low infrastructure friction. Local filtering reduces noise. Cloud reasoning generates the operational summary. Humans confirm the outcome.
Connect the camera feed and monitor for motion or person activity in relevant store zones.
Pull a short clip and representative frames instead of sending continuous video upstream.
Analyze the sequence for possible concealment patterns and visible evidence quality.
Generate risk level, false-positive risk, evidence summary, and recommended next action.
Route the event to a human reviewer before action or memory reinforcement.
Confirmed outcomes update store memory so the system gets better inside that environment.
RICK™ Vision’s first job is precision. The language, workflow, and review process are designed around operational awareness and human confirmation.
Alerts use “possible concealment” language. The system does not accuse a customer or employee.
Every meaningful event routes to review before it becomes a confirmed incident or training signal.
Unstable detections, low evidence quality, normal employee activity, and weak sequences are suppressed.
The MVP produces structured event data so alerts, review queues, dashboards, and memory ingestion all speak the same language.
{
"event_type": "possible_concealment",
"suspicious": true,
"confidence": 0.72,
"risk_level": "review",
"false_positive_risk": "medium",
"recommended_action": "send_to_human_review",
"human_review_required": true,
"memory_tags": ["camera_04", "shelf_interaction"]
}
The long-term product is autonomous retail cognition: store awareness that can support merchandising, shelf intelligence, cooler awareness, staffing signals, inventory correlation, and predictive shrink insight.
Know which displays are touched, ignored, blocked, or repeatedly disrupted.
Understand cooler interaction, dwell time, product zones, and restock patterns.
Compare what happened on the floor with what happened at checkout.
Teach the system the layout, patterns, high-risk zones, and normal operating rhythm of each store.
Move from one clip to a complete path across camera angles when the store has enough coverage.
Use the same awareness layer for fall detection, safety events, and compliance review once the concealment workflow is proven.
Start with one location and grow. Every tier includes AI review, human confirmation, and store memory. Larger footprints add deeper awareness and multi-site intelligence.
Pick the tier that matches your footprint. The first deployment is designed to prove operational usefulness before you scale across locations.
Get Started with RICK™ Vision