Russian scientists develop AI system to identify people in videos: Report
Russian scientists have developed an artificial intelligence (AI) system that can pinpoint the age and gender of people in videos.
According to a report by Press Trust of India, scientists at Higher School of Economics in Moscow say that the development is already providing the basis for offline detection systems in Android mobile applications.
Facial recognition software typically has separate neural networks responsible for different detections, as in one identifies a person, the second finds out gender and so on.
However, the Russian researchers claim to have developed an effective neural network that has multiple outputs. It performs several tasks like age and gender prediction. It also generates a set of numbers which it attributes to each individual for distinguishing one from the other.
The method involves gender detection using modern neural networks with 90% accuracy.
Detecting the age, the report points out, is a much more complex process.
Traditional neural networks read distinctive features on each frame and associate a certain age probability. The sum total of the probabilities from numerous frames is then considered to arrive at a certain age.
The probability can vary because of the various face angles captured in different frames. A slight head rotation spawns various frames which further leads to variations in the prediction within a range of up to five years, give or take.
Identifying faces has become a common feature in consumer products, but questions have been raised over the use of such technology.
US civil liberties groups in May called on Amazon to stop offering facial recognition services to governments, warning that the software could be used to target immigrants and people of colour unfairly.
Earlier this month, Microsoft announced ethical principles for the use of its facial recognition technology, saying it would bar such technology from being used to engage in unlawful discrimination and would encourage customers to be transparent when deploying such services.