“I am very grateful for all the hard work our team put into this latest release that enhances speed and accuracy and provides compelling new ways to use our software,” said Vytautas Pranckenas, SentiVeillance product lead for Neurotechnology. “There has never been a more important time for reliable recognition solutions that can adapt to new conditions, such as people wearing masks.”
The new SentiVeillance mode combination (face and VH) allows tracking of the subject even when the face is no longer visible – functionality that is particularly useful in scenarios where tracking a person’s position is important.
BIOMETRIC FACIAL RECOGNITION
The high reliability of SentiVeillance’s biometric facial identification algorithm allows it to be used with large watch-list databases, both for identifying a person who is on a list and adding new faces from video streams to watch-lists. It tracks identified faces as they move around the camera’s field of view and continues tracking even if a person moves behind an object and re-appears. It is effective both close-in and at a distance when using high-resolution cameras. SentiVeillance can perform gender classification, evaluate a person’s age, identify facial expressions (e.g. smile, open mouth, closed eyes) and detect particular attributes, such as if the person is wearing a mask, glasses or sunglasses and whether they have a beard or mustache.
VEHICLE and HUMAN DETECTION AND MOVEMENT TRACKING (VH)
SentiVeillance detects both moving and static vehicles or people in a scene and performs object classification and tracking until the subjects disappear. In addition to pedestrian detection and vehicle classification by type, it provides identification of the specific vehicle make and model as well as its orientation. For pedestrians, SentiVeillance identifies different types of clothing and provides gender prediction. The algorithm also returns an estimation of paint color for vehicles and predominant clothing color for pedestrians, and it determines the vector in which they are moving (e.g. north, south, southwest).
AUTOMATIC LICENSE PLATE RECOGNITION (ALPR)
The new ALPR capability in SentiVeillance automatically detects and reads vehicle license plates, recording the information from both stationary and moving vehicles within the scene. The latest release is up to 2 times faster for CPU uses and up to 5 times faster when a GPU is used. Accuracy tests reach 99.1% for >200 pixel width license plates and 88.7% for license plates captured from far away (<100 pixel width).
The latter two modes (VH and ALPR) can be used together to create larger, more varied solutions. For example, when conventional ALPR is used for road tolls, automatic car washes or paid parking systems, users might try to avoid paying by altering or exchanging license plates. Stolen vehicles might also have their license plates changed. When using multiple analytics in concert, the resulting solution could match and verify plate numbers with other characteristics of the vehicle, such as type and color, through queries of previously stored values or vehicle registration databases.
The new SentiVeillance is designed to run on multi-core processors for fast performance and can process video data from multiple cameras simultaneously using a common PC (current generation i7 CPU with 8 or more cores) and can utilize multiple graphics processing units (GPUs) to achieve even better performance. It can be used with large surveillance systems, incorporating many cameras and data-processing nodes. Developers have many and varied options in the creation of scalable, cost-effective solutions for their customers.
THERMAL FACES SAMPLE
The new SentiVeillance SDK package contains a precompiled sample with source codes and device integration for the Mobotix m16 camera with thermal sensor. It shows functionality for face tracking, mask estimation and temperature readings, and it allows the setting of temperature thresholds or estimating deviations from averages. The sample is created to help with automatic entrance control.
Neurotechnology is a developer of high-precision algorithms and software based on deep neural networks and other AI-related technologies. The company was launched in 1990 in Vilnius, Lithuania, with the key idea of using neural networks for various applications, such as biometric person identification, computer vision, robotics and artificial intelligence. Since the first release of its fingerprint identification system in 1991, the company has delivered more than 200 products and version upgrades. More than 3,000 system integrators, security companies and hardware providers in more than 140 countries integrate Neurotechnology’s algorithms into their products. The company’s algorithms have achieved top results in independent technology evaluations, including NIST MINEX, PFT, FRVT, IREX and FVC-onGoing.
Jennifer Allen Newton
Bluehouse Consulting Group, Inc.
jennifer (at) bluehousecg (dot) com