Customer Overview
Customer is a growing Information Technology company with offices spread across multiple geographies.
Business Challenge
- Client has over 1,000 employees with high mobility across multiple sites.
- Problem 1: Accurate attendance data is crucial due to business nature and clientele.
- Problem 2: Coworking space locations pose security risks due to high foot traffic from external parties.
- Issue with Current System: ID card verification is unreliable (forgotten IDs) and prone to tailgating (inaccurate data).
- Limitations of Existing Solutions: Turnstiles are too expensive and impractical due to building limitations.
Our Solution
The solution had to be fast, accurate, hassle free and should run on a low footprint device that will be positioned at all entry/exit points. The solution should not be dependent on always on internet connection and all meta data needed to validate and employee must be available locally, on device.
Face Recognition -Inference Pipeline
Optimizations and Sub-Problems Involved
Business Value Delivered
- Achieved an inference speed of 200 milliseconds from In-camera frame to authenticated
- Accuracy of 99.21% on over a 10,000 validation points and over 100 distinct identities
- Deployed on small footprint devices such as Intel NUC
