The Goal

A big Bulgarian retail chain wanted to improve their marketing and merchandising by unobtrusively tracking customers in the store and analyzing their behavior based on general demographic patterns (like age, gender, race). The required solution had to be able to model and assess the count, profile and shopping behavior of all customers who entered the store (including peak and off-peak hours) with speed and precision. The main goal was to use this data when making strategic decisions on how to best adapt promotional offers and adjust inventory layouts to retain customers.

The Challenges

There were various challenges in achieving this goal. Some of the most significant ones were:

• finding suitable locations to install the cameras for the best field of view (FOV), optimal lighting conditions, etc.;
• ensuring the minimal quality requirements for face detection to enable reliable face recognition and demographic classification.
• counting the same visitor only once by re-assigning the visitor appearances on different cameras to the same virtual identity;
• allowing for different face orientation and taking into account different hairstyles, glasses, beards and other visual biasing;
• excluding the staff from the data collection.

The Solution: An Intelligent Customer Demographic Analysis System

By analyzing the video streams captured by the cameras, the solution identifies different profiles and what attracts shoppers and what doesn’t. To account for the place-specific requirements and ensure the collection of high quality data, Ontotext installed cameras that provided the necessary optics and embedded functionality.

At the back-end, the system consists of several modules, the most important and computation intensive of which is the stream analyzer. It is responsible for human detection and tracking, face detection, characteristics calculation and demographic classification. The system features several novel trained ML models and efficiently handles different face orientations and visual biasing.

Using a human head detector and an innovative tracking algorithm, the system determines the moments when a specific person appears and disappears, which enables logging these events. The collected metadata is sent to server processes, where the data from different cameras is combined, all the appearances of a person are matched and some sophisticated aggregations (such as excluding the staff from the data collection) are calculated.

After that, the shoppers are classified using Google Analytics demographic data categories and the data collected from visits to both online and physical store is analyzed. This enables the store management to adapt to both the physical and digital demands of their customers and to improve customer segmentation.

Finally, the web UI enables users to easily manage locations and specific sensors, to define different zones for collecting data and to access various reports.

Why Choose Ontotext?

Empowered by the intelligent customer demographic analysis, now the retail chain store can take advantage of the vast amount of data that also existed previously, but was difficult to collect and analyze. The result was a more efficient supply chain, with better use of available resources and, ultimately, increased profits.

With Ontotext’s technology, the retail chain store could also:

• have a reliable tool for validating running marketing campaigns;
• calculate the demographic distribution of sales per product (in conjunction with the cash register sales system);
• increase up-selling and complementary purchases.

Do you think this case resembles your particular needs?