The customer was AUCHAN Retail Russia, one of the leaders of Russian retail and a long-term partner of ICL Group, which has been using the company’s IT services and products for more than 14 years.
The self-service checkout (SSC) format is actively gaining popularity in the client's stores. It speeds up shopping and reduces queues without adding extra pressure on the staff. However, SSCs also lead to increased losses for stores due to scanning errors and theft.
The proposed solution was a smart video analytics system where a neural network, analyzing camera data, would alert security about anomalies at the SSC.
Our team was chosen for the project. In addition to the excellent test results, the client saw the advantage in the years of flawless collaboration with the company on checkout systems and the team's readiness to transfer rights to the developed software and hardware suite to the client.
Key Challenges
- Maximize the percentage of detection of anomalies when scanning goods and signal this to customers at the checkout thanks to the video analytics system.
- Jointly design customer journeys both for the client at the checkout and for employees in the self-service and security checkout area.
- Implement a system with minimal impact on the overall customer payment experience at self-service checkouts.
The project was launched in December 2022 at two Auchan "pilot" stores in Moscow.
The initial step involved coordinating the solution architecture and IT infrastructure security requirements with the client. In line with the approved architecture, a network firewall was installed to safeguard client data, enabling us to simultaneously use video from the stores' own cameras and SSC data.
Further work on the project went in several directions:
1. elaboration of client and employee paths;
2. receiving and analyzing video recordings from customer cameras;
3. video marking and AI training;
4. integration with self-checkout software.
During the implementation process, the ICL team, together with the security service, determined the optimal parameters for installing video surveillance cameras for the operation of both the video analytics system and the security service.
The system was launched in stages.
First of all, the AI has learned to distinguish the discrepancy between the number of items in the cart and those included in the receipt. Subsequently, it was possible to distinguish the lack of payment from the client, as well as identify specific goods, thanks to which it was possible to track incorrect scanning of goods with the same name, but different types (taste, color).
Based on the results of the pilot implementation, an approach to further development of the system and technical improvements was formulated.
Products and technologies
- Computer vision
- Integration with self-service checkouts
- Firewall
- PostgreSQL database
Results
- The system is now operating in two hypermarkets, analyzing in real time the behavior of customers at self-service checkouts.