Visionlytics

Visionlytics is an end-to-end AI-powered shopping mall and retail intelligence platform designed to eliminate operational blind spots, safety risks, and revenue losses caused by a lack of real-time physical retail data. By synchronizing live multi-camera edge computer vision with gradient-boosted machine learning, Visionlytics empowers security teams to proactively prevent crowd surges, guides facility managers in optimizing spatial floorplans, and integrates Point-of-Sale (POS) sales streams to prevent inventory understocking and overstocking.

1. People Counting & Real-Time Occupancy Monitoring

Utilizes fine-tuned YOLOv11 object detection alongside ByteTrack multi-object tracking on entrance camera feeds to calculate real-time visitor entry, exit, and net active occupancy. Operating without intrusive facial identification, this computer vision pipeline achieved a verified 79% counting accuracy under dense footfall conditions.

2. Spatial Footfall Heatmaps & Zone Intelligence

Processes multi-camera surveillance feeds to visually project spatial density heatmaps onto shopping mall architectural floorplans. Mall management can instantly distinguish between high-velocity corridors, prime commercial displays, and underutilized dead zones to negotiate data-backed commercial leases.

3. Precision Dwell Time & Engagement Analytics

Continuously calculates visitor linger durations across general seating, promotional pavilions, and billing checkout counters. Delivering a low Mean Absolute Percentage Error (MAPE) of 9.14%, this module enables store operators to correlate customer dwell times with actual transaction conversion rates.

4. Short-Term Crowd Surge Forecasting

Combines historical foot traffic curves, day-of-week seasonality, and weather factors within a high-speed LightGBM regressor to forecast future crowd volumes hours ahead. Achieving a tight Mean Absolute Error (MAE) of 3.37, security supervisors can dispatch personnel proactively before bottlenecks develop.

5. Product-Level SKU Demand Forecasting

Bridges physical footfall volume with automated Point-of-Sale (POS) transaction datasets to forecast retail demand at the individual SKU level. With a Weighted Root Mean Squared Scaled Error (WRMSSE) of 0.62, retail tenants optimize reorder cycles and mitigate stockout shrinkage.

6. Unified Analytics Dashboard & Role-Based Reporting

Delivers intuitive, role-based dashboards tailored for mall executives, security dispatchers, and individual retail tenants to inspect live occupancy, interactive heatmaps, historical footfall trends, and automated operational alerts in real time.