Case Study

We design and manage the development of an AI-driven solution to authorize premise entry using face recognition in real time that scales horizontally to thousands of requests per second using a state-of-the-art face recognition model and a specialized vector similarity search engine.
Read NowDownload PDF

Face Recognition

Face Recognition is a challenging issue as it requires a method of storing face representations in a compressed, yet comparable way.

We can solve this problem by training a neural network model to reduce a picture of a face into a 512-dimensional vector of floating point numbers. The model is trained to output vectors in such a way that vectors of the same face are similar (close) and vectors of two different faces are different (far away).

This allows us to measure the distance between two face representations. This part of the pipeline is compute-intensive and so runs on our autoscaling neural inference framework.

Vector Search Engine

The next problem is the storing and querying of these vectors. Conventional databases are insufficient for such a task as a single query would require several million distance calculations in high dimensional space.

This can be solved using a specialized database called a vector similarity search engine designed for storing and querying N-dimensional vectors efficiently by using approximate nearest neighbor methods such as locality sensitive hashing or dimensionality reduction.

 

Outcome

Solving these problems enabled us to implement a real-time face recognition pipeline to authorize an individuals access to a particular premises at scale.

  • Delivers a fast and accurate response in less than 200ms for each request.
  • Scales horizontally to thousands of requests per second.
  • Uses a state-of-the-art face recognition model and vector similarity search engine.
  • Adds an extra layer of security to office and laboratory premises.
  • Was developed and deployed on-time with the specified budget.

Need Consulting? Contact Us Now!

If your company is considering using AI, feel free to schedule a consultation.