AI-based Face Recognition
Mindbox facial recognition software searches an existing database of faces and compares them with the faces detected in the scene to find a match. Face Recognition detects faces in the camera’s field of view – as many as 15 at the same time – and matches them against faces previously stored in the database. Anti-spoofing is provided through “liveness” testing without the need for a stereo or 3D camera. Faces can be enrolled in the database from existing still images or from the video camera itself.
Key Features
Facial recognition accuracy over 99.5% on public standard data sets
Face recognition in real-time, depending on resources
Easily enroll faces from still or video images
Zero gender or racial bias through model training with millions of faces from datasets from around the world
Anti-spoofing technology ensures the system cannot be fooled by a photo or video image
Detect matches with faces in the database and provide alerts
Create a log of faces in the scene for later investigation
REST API for building into applications and devices
Search for similar faces from a single camera or across multiple cameras
In use in thousands of cameras for access control, VIP greeting, shoplifter, and unwanted person applications
Deep learning algorithms
Facial recognition in Face Intellect is powered by deep neural networks (DNN). Algorithms of the new generation are free of recognition issues which were typical for the previous “non-DNN” generation.
Neural network algorithms are basically AI, artificial intelligence — powerful machine learning-based techniques, which emulate how the human brain operates. DNN is trained on a huge dataset with labeled faces to map a face to a numerical vector representation. Once the network has been trained, it can compare any faces, even ones it has never seen before.
Facial recognition with DNN offers top-quality predictions regardless of the camera angle, lighting, hairstyle, facial expression, glasses, or other variations. Actually, modern algorithms work even better than humans can do.
How Face IntelliVision Works
Step 1: Face Intellect automatically picks out faces in the video feed from cameras.
Step 2: It compares them to a database, such as an employee access list or a blacklist.
Step 3: When it determines a given degree of similarity (high or low), it triggers the system to lock or unlock a door, send an alert to security personnel, and so on.
Step 4: When used in access control, facial recognition can also be part of a Time and Attendance system.
Search video footage
You can quickly find faces that match a picture, video image, or photo-fit and jump to event video.
Collect statistics
Use Face Intellect as a people counter to get unique and total visitor numbers, find out their gender and age, and get reports for business analysis.
Using AI and deep learning, Mindbox face recognition has achieved accuracy benchmarks better than industry leaders like Google and Facebook. It scores the following accuracy in the leading public test databases – LFW: 99.6%, YouTube Faces: 96.5%, MegaFace (with 1000 people/distracters): 95.6%.
Recognition is available in both real-time and off-line modes and enrollment is available from both video and still images. Facial recognition is achieved by analyzing multiple images per face and can be achieved in around 0.5 seconds depending on resources.
Face Recognizer is available with a REST API/SDK for OEM partners and application builders. Easy integration of alerts is achieved through HTTP/JSON and open architecture.