ETU “LETI” Developer’s Neural Network System To Evaluate Customer Activity

The project of an intelligent system or controlling the movement of people, presented by Sergey Antonov, a master’s student of the Faculty of Computer Science and Technology of ETU “LETI“, has won a grant in the UMNIK competition.

The developed software can be used in retail stores providing functional to divide customers into groups for determining the most popular products for each group of customers and evaluating the effectiveness of advertising, promotions, and other activities.

“Companies can use the project to analyze customer activity in the store and improve the marketing strategy based on the received data. Nowadays, there are many different analogs: from ordinary sensors for counting the number of visitors to analytical systems that use machine learning methods. The system under development will be superior to cheap sensors and comparable to expensive analytical systems,” the developer says.

An intelligent system will identify the most popular products for different groups of visitors and compile statistics to assess the effectiveness of advertising according to the data obtained. Also, the software will have functional to determine the optimal number of consultants and cash registers to ensure quality service and work distribution of staff: the system will help to draw up a schedule of cleaning, maintenance, equipment configuration, and merchandising. The project will also make stores safer for visitors by collecting data on the number of people who were inside after the store was closed or evacuated.

The innovativeness of the development lies in the use of machine learning methods with modern neural network architectures, which allows solving tasks with high accuracy. Specially trained neural networks will identify people on video and recognize their faces. The system uses methods of projective geometry to determine the location of a person: it transforms the image received from CCTV cameras into a top view.

“The proposed solution will find people using CCTV cameras, track and analyze their movements in real-time. With the help of a facial recognition system, the software can identify employees of a store among customers and divide customers into groups by gender and age. Also, now we are working to apply the system for social distance control in the fight against COVID-19,” Sergei Antonov said.

The idea of using such a system in stores appeared after Sergey wrote his bachelor’s degree in Automated System for Controlling the Movement of Employees. After that, the student decided to develop a ready-made system under the guidance of Alexander Sinitsa, a postgraduate student of the Department of Automation and Control Processes.

Sergey assumes that companies in any sphere can use the project to analyze how long employees stay in various departments, are late or absent from work. Museums can use the software to control attendance and analyze capacity.

The developer will use the funds of the received grant to purchase the necessary hardware and software. The project will take two years to implement. During the first year, the researcher plans to study various models of neural networks and train the suitable ones to find people on video and to develop software for building heat maps with trajectories of found people.

During the second year, the student will research different architectures of neural network models for finding and recognizing faces, compare different classifiers and choose the optimal one for finding the same faces, create a prototype of the user interface, and test the system on real cameras.