SecurOS Auto NN license plate capturer and new SecurOS FaceX features in SecurOS 10.2 R1
Despite the fact that in three weeks a new SecurOS release – SecurOS 10.3 will be launched, ISS would like to present you an intermediate release that includes crucial improvements of the License Plate and Face Recognition video analytic modules.
From now on the License Plate Recognition module includes license plate capturer based on Convolutional Neural Networks (CNN) – NN capturer. This new component significantly increases recognition accuracy in difficult conditions:
- very few symbols on the license plate
- multiple text signs or symbols on a vehicle
Close partnership between ISS NN Laboratory and Intel R&D center in Nizhny Novgorod allowed us to maintain high-level support of Intel OpenVINO technology. As a result most of ISS neural network components and modules including NN-capturer are released with built-in support of the Intel OpenVINO. OpenVINO allows to efficiently utilize both Intel CPU resources and resources of Intel HD Graphics (integrated graphics processor), providing computing power for NN calculations even when a discrete graphics card is not available.
At the same time, we provide a separate version of our NN-modules with support of NVIDIA graphics accelerators. NVIDIA version of modules is used when heavy calculation power is required.
ISS NN Laboratory in the GitHub community
Vast practical experience allows the NN Laboratory to create its own approaches to meet the challenges of neural network development. Some useful tools created by the Laboratory are published on the GitHub at ISSResearch web-page. GitHub is the world’s leading OpenSource development platform where developers host and review code, manage projects, and build software alongside other developers.
The Laboratory staff created a convenient toolset to convert data between different formats of “datasets”– sets of annotated images. The conversion tools allow developers to use the majority of existing datasets in order to train neural networks in various frameworks. Developers can convert annotated images in the most popular dataset formats (ADE20K, CVAT, CITYSCAPES, Open Images Dataset, VOC) into the COCO format that is primarily used by the Laboratory. From COCO the dataset can be converted into formats used for NN training in Caffe, TensorFlow (Tensorflow Object Detection API), MXNet (Gluon), Caffe2 (Detectron) frameworks and also into VOCCALIB format supported by Intel OpenVINO.
Dataset converter is available in the ISS open software repository at the GitHub, Dataset Converters section.
Follow ISSResearch account on the GitHub for the latest news.
Now SecurOS FaceX module is capable to detect when someone is trying to provide a printed photo or photo displayed on phone or tablet screen for face recognition. When spoofing is detected FaceX operator sees a warning message.