Facial Recognition

  Home / Products / Facial Recognition

Facial Recognition

Nvidia

Bikal Bio Surveillance Facial Recognition is the right solution to increase security in crowded or security warranted environments.


In combination with a proprietary feature dataset, this technology achieves a 5X speedup in 1080p videos over the fastest known implementations, while slightly improving the accuracy. Additionally, the latency of these algorithms is linearly reduced as the core count of GPUs are increased, thus opening the door to 4Kx2K (QHD) resolutions. Key patents in the area of pattern recognition for multithreaded stream-based microarchitectures have been filed.


The alarms generated by the system are customisable on the fly, allowing the final users to create and maintain blacklists (or whitelists) of targets. The system also allows both live and static training from multiple sources (e.g. images or videos) BioSurveillance pushes the stream computing paradigm of GPUs to the limit by performing multiple stages of the face recognition pipeline in parallel. Algorithms solve the problem of GPU underutilisation by exploiting both coarse-grain and fine-grain thread-level parallelism for reducing the latency of resource-intensive stages, such as face detection or video decoding.


The system is especially robust to pose and expression variability after the template of an individual has been generalised using multiple input images, or a short video sequence of around 5-10 seconds, containing a series of different poses.


The proposed video surveillance solution detects multiple faces in real-time, generating an alarm for every recognised subject. These alarms can be reviewed directly from the application, while the system is running, in order to retrieve information about the subject, camera snapshots and the timestamp of the alarm. The alarms are also stored in a SQL database and can also be sent to remote devices (mobile, PDA, tablet, security control centre). Our solution consists of a master BioSurveillance server which is used for subject management and centralised alarm supervision, and local BioSurveillance stations which separately process several camera video streams. This structure permits to escalate the system to process a high number of cameras simultaneously.

Facial Recognition Technical Configuration

Working conditions:
  • - Normal lighting conditions, not extreme ones. Uniform illumination desirable.
  • - Users passing through a security gate at a normal speed.
  • - Faces near frontal position (+/-35 degrees left/right deviation, +-20 degrees up/down deviation)
  • - Camera in front of the gate, preferably at 2-2.5 meters height.
  • - Good quality images, at least 720p.
  • - Permanent network connection.
  • - Minimum face size for enrolment: 150px x 150px.
  • - Minimum face size for recognition: 70px x 70px.

Facial Recognition

Facial Recognition

Solution Architecture

Current BioSurveillance solution is divided into two different applications. Subject management and alarm supervision will be centralized in a master BioSurveillance (mBS) server, as well as the administration of all the local BioSurveillance (iBS) stations.

Local BioSurveillance

Every local BioSurveillance station performs the face recognition task from several video streams. The number of cameras that each iBS is able to process simultaneously depends on the computing power of the station, the video resolution and the number of faces that appear in the images.

  • - Local alarms can be supervised in real time and they are all sent to the mBS. Therefore, local alarms table has all the alarms of the iBS station.
  • - Each iBS updates its blacklist by communicating to the mBS. Thus, subjects table is an exact replica of the centralised subjects table.
  • - Local settings can be configured to get the best possible performance.

Master BioSurveillance

The main functionality of the server solution is to administrate and configure the whole system in terms of alarms, subjects and iBS stations.

  • - Alarms from all the iBS stations are received and added to a global alarms table, which contains alarm information, iBS id and camera id. Alarms are sent as they are generated from iBS stations to mBS, which receives them on-the-go.
  • - Subject enrolment and management is conducted in this central station and changes are sent to all the stations, so all subject tables are identical.
  • - iBS management tab allows the user to add, edit, delete and configure iBS stations.
  • - Last alarms from all iBS are shown on a new interface, so live alarm supervision turns into an easy and intuitive task.

Optionally, a secondary mBS can be installed with a replica of the primary server database. Thus, in case of a failure in the primary mBS, the secondary mBS acts as a failover server.

For further information please contact us on uk@bikal.co.uk