AI Person Counter & Demographic Detection: Level up Retail Analytics

The Person Counter initiative is a sophisticated Computer Vision solution developed by MOR Software for a Japanese retail leader. Over a 5-month development cycle, our team delivered a 15 Man-Month (MM) project designed to automate foot traffic analysis. By utilizing YOLOv8, BoT-SORT, and PyTorch, we built a real-time system that not only counts individuals with high precision but also classifies demographics to provide actionable business intelligence.

The client is a prominent Japanese retail company operating high-traffic locations, including shops and transit-integrated retail spaces. To optimize their marketing strategies and operational efficiency, they required a deeper understanding of visitor patterns—specifically who is visiting their locations and how they move through the space.

The client faced a significant data gap regarding physical store traffic. Relying on manual counting was labor-intensive and prone to human error, making it impossible to scale across multiple locations. Furthermore, they lacked automated ways to identify the age and gender of pedestrians, a critical requirement for targeted marketing and statistical analysis. They needed a non-intrusive technology that could maintain accuracy in crowded environments where individuals are frequently obscured by other people or objects.

MOR Software engineered a robust AI solution centered on Multi-person Tracking (MOT) and Ensemble Classification. Using a combination of YOLO and BoT-SORT, the system assigns a unique identifier to every person in the camera's field of view. A key innovation in our approach is the system's "re-identification" capability; even when a person is temporarily covered or moves behind an obstacle, the AI maintains the correct object ID.

To provide the demographic insights the client needed, we implemented an Ensemble Model for Gender and Age Recognition. Unlike standard systems that rely solely on facial features, our solution analyzes a combination of face, outfit, and overall appearance (Style Classification). This ensures high accuracy even when a person is not looking directly at the camera. The entire infrastructure is hosted on Amazon Web Services (AWS), ensuring a secure and scalable cloud environment for real-time data processing.

The deployment of the Person Counter system has transformed the client's approach to retail data. They now possess highly accurate, real-time statistics on visitor volume and demographic breakdowns, ranging from children to seniors. These insights have allowed the client to tailor their storefront marketing and staff allocation based on actual traffic trends. With a scalable and automated system now in place, the client has significantly reduced manual labor costs while gaining a level of data granularity that was previously unattainable.