Developed software to help prevent theft in SPAR stores. The software alerts security staff if a person who has been caught stealing has entered the store. The project was developed by specialists who performed similar tasks for the Ministry of Internal Affairs.
About the customer:
The largest food retailer with 12,700 stores in 48 countries, SPAR.
The client owns a chain of retail chain, and most of the stores are experiencing thefts that significantly reduce profits. Security officers occasionally manages to catch thieves and include them in the photo database. However, to identify a person from the database of at least a hundred criminals is a very difficult task for a person. Especially if the store traffic is dense. We set the following tasks:
Develop a service that alerts the security team when a visitor who looks like one of the already caught thieves enters the store.
Design that system so that security officers could easily operate the interface and add images of new thieves to the base.
It was necessary for the system to work on the existing technical base and require a minimum of effort from the security team.
Following the discussion, we understood that the most effective solution for the customer's task would be neural networks that can recognize objects in pictures. To implement the project:
Together with the legal office of store security we agreed to the task, taking into account possible nuances.
Trained a neural network to instantly recognize thieves in the store traffic.
Developed software and equipped the workplace of security personnel in the store.
The service works as a web service, so security personnel does not need to update or configure it.
The customer was completely satisfied with the results achieved:
The system recognizes known thieves at the entrance and notifies the security team.
The security officers keep track of them during the whole time they are in the store.
To achieve a high quality of neural networks in pattern recognition, we brought in specialists with 8 years of experience working on government projects for the Ministry of Internal Affairs and the Ministry of Education.
Strengthening the team with such experts helped complete the service faster and avoid a number of issues upon deployment.
The system consists of two parts - a client part and a server part. The client part analyzes the video stream and recognizes faces, then sends them to the server. The server part compares the face of a store visitor with thieves from the database. If a match is found, the system notifies the security team.
The service does not need wide access to the Internet, and the server does not need large computing power.
The server and client parts are written in cross-platform Python and Python QT. Customized OpenCV kit is responsible for facial recognition.
To train the neural network we used Tensorflow, on an architecture that recognizes human faces better than humans themselves. We tested the product on a third-party similar neural network to save the project budget at the technology validation stage.
Just add a cheap computer with a camera to the security team, connect it to the Internet to plug in the store to the system.
Just add a new person to the database in a minute, and the system will immediately begin to recognize him throughout the chain of stores.
The first working prototype of the system was presented in 25 days.
All visitor information is preprocessed and then stored only in depersonalized form.
We are ready to develop a similar system with neural networks or other complex IT solutions for you. Contact us, and we will tell you how you can benefit from hi-tech solutions.
Thank you. The message was sent. We'll contact you. Please, wait.
