Retail CCTV footage analytics

Most of the retailers have already implemented CCTV systems, which they use to monitor the property or investigate incidents post-event. But the camera networks can support retailers in building a competitive advantage by leveraging data analytics and AI technologies.

CCTV cameras provide a large amount of information for retailers. However, it is extremely difficult to leverage their business potential, as the data are not structured. The goal of this project was to source data from retail CCTV footage and prepare it for further analytics.

Our solution implements multi-object tracking involving detection and re-identification of customers visible from multiple cameras. The position of customers is mapped to the 2D plane for convenient monitoring and assessment.
The system delivers three major functionalities:

• track customers moving in a shopping environment
• create a 2D map visualization representing customers’ motion
• detect actions “performed” by the moving customers (eg. walking around the shop, waiting in a queue, or paying)

All data are anonymized.

The system has achieved over 94% accuracy in action detection. The retailers’ CCTV cameras footage can be utilised for further analysis and use cases, including:

• Motion detection in particular shop areas
• Customer counting & dwell time monitoring for retail stores
• Recognizing long lines at checkout and sending alerts