Congratulations to the winner of the “UMNIK-Sberbank” contest Arseny Zorin!
Arseny’s development proposed for the competition in the nomination “Data Science in the Banking Sector” solves an essential problem of identifying masked persons in video content analysis.
Despite the high accuracy of object recognition achieved using machine-learning technologies, the majority of existing recognition systems are ineffective in detecting partially closed faces.
Arseny Zorin’s method, which solves this problem, is based on the sequential application of two neural networks: outcomes of the first network serve as the source material for the analysis performed by the second one.
Face detection is carried out in stages: the system detects all faces in the video stream, selects each visible part of the face and then examines the presence of all parts (eyes, nose, and mouth). If any parts of the standard set of the face are “missing,” the system determines the face as disguised and sends a signal to the operator.
The funding received through winning the competition, in the next two years will allow the researcher to improve the project in such areas as enhancing the accuracy of detecting persons in the presence of interference (size, lighting, accessories, etc.) and detecting deviant human behavior.
Thanks to resource-efficient algorithms, the method involves no complicated energy-intensive calculations, which makes it possible to use embedded low-power computing devices of the bank card size for its implementation.
Therefore, the proposed approach has high chances of widespread implementation and can be used to solve a broad range of security tasks, primarily for analytics of video surveillance in public places aimed at preventing crime.