Elastic claims facial recognition searches 100X faster with vector database

The performance requirements for police facial recognition systems vary across use cases and jurisdictions. For a police force in an unnamed large city in Brazil, timely matches are only possible by using a vector database, according to a case study from AI search technology provider Elastic.

The case study from the Dutch-American software firm explains that the Brazilian public security organization in question, which oversees a district of 3 million people, equipped its officers with a custom mobile and web app for easy photo capture and upload.

High search volumes of huge databases, however, can lead to slow results. Facial recognition searches took up to a day before the force switched to the vector database, Elastic says.

“To bridge the gap between this app and the image database, they needed a powerful vector database to enable vector search” of its six terabytes of data. “After evaluating options, the head of technology and their team chose Elasticsearch for its unbeatable combination of blazing-fast speed, user-friendly interface and robust security features.”

Per the document, “the head of technology emphasizes the power of the Elasticsearch Relevance Engine (ESRE) for the organization. It lets the agency use their own, or third-party, machine learning models to transform data into a special format called ‘embeddings.’ These embeddings capture the relationships between different data points. The embeddings are stored efficiently at a massive scale, allowing the agency to search this data extremely fast in real time.”

Clearview VP of Machine Learning and Research Terence Liu explained how his company uses embedding vectors to make its facial recognition run efficiently on billion-scale databases to Biometric Update in a 2023 interview.

Elastic says it has handled over 1 million biometric searches in just the first three months of working with the force. While the company’s materials emphasize how police can use FRT to keep families safe from crime, they also note other use cases beyond criminal prosecution.

“In one instance,” says the case study, “officers found an elderly man with Alzheimer’s disease who had been missing for almost two weeks. The officer took a photo and sent it to the missing person’s database, which returned his name and contact details.”

Article Topics

biometric database  |  biometric matching  |  biometrics  |  Elastic  |  facial recognition