Highly performant algorithm
Argus system provides real-time facial detection; an essential feature in computer vision. Our algorithm is highly performant with noticeable speed and accuracy, which runs on GPU devices produced by NVIDIA. You don't need to have expensive hardware to run these algorithms as many devices that execute real-time facial detection and recognition can also be found in the form of a mobile CPU, GPU, or some neural processing units.
Widespread architecture
Vision Queues AI-powered facial detection system is designed to be distributed with most of the backend services running on the distant edge nodes inside our architecture.
Meet the algorithm
Get familiar with how our products run artificial intelligence algorithms in the background.
Face detection and tracking
First, each frame of the video stream is analyzed. The face of each person present in the current frame is detected and prepared for identification. In all subsequent frames, previously detected individuals are reidentified and tracked without needing the whole identification process. The face detection algorithm can detect an unlimited number of faces in a frame. Each face detection is represented with bounding box coordinates (defined by top left and bottom right corners) and five facial landmarks (eyes, nose, and corners of the mouth) for further work with each one of them.
Head pose estimation and face alignment
Face alignment is the task of identifying the geometric structure of the face and obtaining a canonical alignment of the face based on translation, scale and rotation. These affine transformations are applied based on the five detected face key points. If the face image is aligned, the model for face identification will be more successful because the center of the face will be equal for all the images that are compared.
Face embedding extraction
Vision Queue's neural network extracts biometric features from the aligned face image as a multidimensional vector.
Face identification
An extracted feature vector is compared against the feature vectors of the people in the database to find the best match. The recognition process is not affected by age-related changes in individuals or partial face coverings such as glasses, beards, or medical masks. Each unrecognized person is automatically added to the database. That way, even if a person's identity is unknown, it can be treated as a regular individual - the history and statistics of all its previous detections can be seen and analyzed.
Antispoofing / Liveness detection
Facial anti-spoofing prevents false facial verification by using a photo, video, mask, or a different substitute for an authorized person's face. Our system can detect whether the detected face belongs to the real person or is part of a fake prop.