Face recognition solution: 80% accuracy rate in identifying faces

Innowise has developed an advanced face recognition solution to enhance security and streamline identification processes.

Customer

Industry

Retail industry

Region

Client since

2020

Our client is an Australian software development and IT consulting company specializing in building IT solutions for the retail industry. This company has an impressive track record of conceptualizing, designing, developing, and launching a range of digital retail solutions catering to various product categories, including general merchandise, apparel, and grocery.

Operating on a global scale, this company offers a suite of highly scalable products and serves a diverse client base, ranging from multinational retail giants to individual store owners.

Challenge

Enhancing facial recognition accuracy for the retail industry

Our team was tasked with the development of a cutting-edge facial recognition system for retail. The project’s primary objective was to create a scalable and distributed architecture that employed various algorithms for accurate face recognition.

A significant hurdle we encountered was the inconsistency in the quality and characteristics of the video inputs and their respective images. This inconsistency primarily stemmed from varying lighting conditions and the disparate quality of the input frames, which hampered the system’s ability to identify and analyze anthropometric points and their adjacent features accurately. Overcoming this challenge was crucial for ensuring the reliability and effectiveness of the facial recognition solution.

The presented below images are either non-homogeneously lighted or blurred or “two-in-one” – blurred and non-homogeneously at once. It is complicated to get a satisfactory recognition result on the basis of such frames.

Solution

Enhancing retail security and efficiency with a face recognition system

Innowise embarked on a project to develop a custom facial recognition software tailored for the retail industry.

Diverse face recognition algorithms

We implemented a collection of face-recognition algorithms, such as unmanaged face recognition PCA, managed face recognition PCA, and managed face recognition eigenfaces. These algorithms are seamlessly interchangeable, offering flexibility and adaptability to meet the specific requirements.

Unmanaged face recognition PCA

This approach utilizes principal component analysis (PCA) to efficiently identify and extract key facial features, enhancing the system’s capability to recognize faces under diverse conditions.

Managed face recognition PCA

Enhancing the basic PCA framework, this method introduces precision management to optimize feature extraction, ensuring reliable recognition even when image quality varies.

Managed face recognition eigenfaces

Leveraging the eigenfaces technique, the system employs a sophisticated selection of eigenvectors for improved recognition efficiency, particularly useful for processing large image volumes.

Enhancements for image accuracy and system performance

We focused on improving the accuracy of images and performance of the system. By implementing two OpenCV algorithms for face detection and eye localization, we achieved stable and reliable face recognition.

However, we encountered difficulties with the eye localization algorithm’s accuracy. That’s why we configured the system to detect the centers of the eye pupils, which significantly enhanced the system’s stability. This adjustment facilitated more accurate image stabilization, rotation, and scale normalization while filtering out images captured at incorrect angles.

Efficient batch image processing

To simplify image processing tasks, Innowise developed a batch image processing module. We then integrated this module into the system to allow extracting images from image series, videos, or cameras efficiently. It significantly saves time and effort, enabling smooth operations even when handling large volumes of data.

Integration with a closed-circuit television (CCTV) system

The integration of a face-recognition solution with a closed-circuit television (CCTV) system involves combining advanced facial recognition technology with the existing surveillance infrastructure. This integration transforms surveillance capabilities, allowing for precise real-time identification of individuals within stores or warehouses. Such a system fortifies security measures against unauthorized access and optimizes employee management by monitoring attendance and behavior. This comprehensive approach ensures a safer and more efficient operational environment for retail outlets.

Now the face-recognition solution can access live video feeds from the CCTV cameras. It contributes to the analysis of individuals’ faces within the video stream in real time. The face-recognition solution employs sophisticated algorithms to detect and extract facial features from the video footage. These algorithms analyze the unique characteristics of each face, such as the shape of the eyes, nose, and mouth.

Additionally, the integration includes features such as facial tracking, which allows the system to follow an individual’s movements across different camera views. This feature enhances situational awareness and provides a comprehensive overview of their employees’ activities.

Technologies

Back-end

.NET 3.5 SP1, C# 3.0 and Platform SDK

Cloud

AWS (Kinesis Video Streaming, EC2, EKS, ECR, S3, Glue)

DevOps

Jenkins, Nginx, Docker, Docker Compose

Machine learning

OpenCV, ONNX Runtime, Armadillo, Scikit-learn, numpy, pandas

VCS

Git, GitHub

Process

Our facial recognition development project commenced with a thorough evaluation of the client’s requirements. To streamline the development process, we adopted the Scrum methodology. This approach involved daily stand-up meetings for real-time progress updates and monthly demonstrations to present advancements and solicit client feedback.

We organized our workflow and documentation using Jira and Confluence, ensuring efficient task tracking and knowledge sharing, while Microsoft Teams served as our main channel for client communication.

At the heart of our technical strategy was integrating cutting-edge algorithms for precise face and eye detection. A key innovation was refining the system’s ability to detect pupil centers and improving image quality through better stabilization and normalization, which is crucial for dealing with the issue of inconsistent video input quality.

This comprehensive Agile approach allowed us to deliver a bespoke, high-performing facial recognition system that met the client’s specific needs, demonstrating our commitment to innovation and client satisfaction.

Team

Business Analyst

Project Manager

Data Scientist

Back-End Developer

Front-End Developer

Results

Enhanced security and streamlined face identification

We achieved a high level of accuracy in identifying and distinguishing individuals, even when working with inferior sources. This accuracy significantly enhanced security measures by providing robust authentication, allowing authorized individuals to securely access restricted areas and systems, preventing unauthorized entry. Additionally, the system enabled real-time monitoring through video surveillance cameras, promptly detecting and alerting security personnel about unauthorized or suspicious individuals attempting to enter restricted areas.

Overall, the face recognition system proved to be a highly reliable, efficient, and secure solution for identification and authentication. The solution offers benefits across various sectors, including access control, attendance management, and improved customer experience.

Project duration

May 2020 - October 2020

80%

accuracy rate in identifying faces

75%

time saving for employees’ verification

Related cases

Contact us!

Book a call or fill out the form below and we’ll get back to you once we’ve processed your request.

What happens next?

Having received and processed your request, we will get back to you shortly to detail your project needs and sign an NDA to ensure the confidentiality of information.

After examining requirements, our analysts and developers devise a project proposal with the scope of works, team size, time, and cost estimates.

We arrange a meeting with you to discuss the offer and come to an agreement.

We sign a contract and start working on your project as quickly as possible.

Face recognition solution: 80% accuracy rate in identifying faces

Customer

Challenge

Enhancing facial recognition accuracy for the retail industry

Solution