Face identification system

Face identifier

Face recognition has been the prominent biometric technique for identity authentication. In the 1990s researches achieved significant success in the area of visual-based facial recognition. However only with the emergence of deep learning, face recognition gained real progress and reached tremendous recognition accuracy. Since then all face recognition techniques and approaches leverage neural networks to convert pixels of the digital image into invariant face representation.

BitRefine provides two separate modules: face detector and face identifier. If a user needs to detect faces, track them and count, he adds just face detection module to the video processing pipeline. If face recognition is required, then both modules work one after another. The pipeline looks in this case as follows. After video received from a standard IP-camera, it goes to a face detector. This module extracts all faces and passes these faces to the recognition module. The face recognition module has a convolution neural network in its core that converts an image of a face into numeric representation – so-called biometric template. This template is built in such a way that it preserves key parameters of a face regardless of currently expressed emotions, light conditions, beard shape, and other minor changes. If a user wants to get a real-time notification or trigger an alarm if a camera catches a person from white or blacklists, then we add the next module down our pipeline – the matcher. It compares incoming templates from the neural network with templates that we have in our lists and if templates match, this means that system found required person and it reacts accordingly.

face identifier

The recognition platform supports multiple user lists that can contain images of faces. The user can upload multiple images for a single person to increase the chance of successful identification. If template from recognition pipeline matches one of uploaded face’s template matcher assigns a person’s real name to the found face and saves it in DB for further analysis.

Face identification system may also be of great use even without an active matcher. One of the examples is counting unique visitors within a certain time span. In this case, all the face templates are saved to DB without matching. Only when the user builds a report, the system compares templates to each other and returns the number of unique faces over the total number.

face recognition program

BitRefine’s reports section is the simplest way to analyze collected data. For the face detection and face counting application, the user needs just to specify cameras or counting line names to get a visual chart on a timeline that shows a number of faces registered within each timespan. If a user runs face recognition and wants to find a certain face, he can upload a reference image and the report tool will return all the matched faces from the database in form of a table with corresponding screenshots. In addition, if a person was seen multiple times the system will reflect on a timeline chart.

BitRefine face recognition successfully combines several functions: collecting valuable business insights, security and time attendance. The recognition platform can simultaneously track visitors on one hand and keep an eye on the company’s staff on the other.

Related Solutions

We’ve got more video analytics to share. Let’s be friends. But you first...