Russian Video Services Learn to Identify People by Silhouette
Researchers at Moscow State University have developed a new re-identification method that allows people to be recognized by their silhouette, even when facial data is unavailable.
Scientists at the Artificial Intelligence Research Center of Moscow State University have developed a new method of re-identification—recognizing individuals based on their silhouette, according to the outlet Scientific Russia. The approach relies on training neural networks on heterogeneous datasets and is designed to identify people across different locations and time periods. The technology is used in Smart City systems and other forms of intelligent video analytics.
Facial recognition does not always work, but extracting silhouettes is often easier. Existing silhouette-based methods, however, have performed poorly due to a lack of data—specifically, the need to capture the same person’s silhouette from multiple angles.
Heterogeneous Data Makes the Difference
Russian researchers, working together with specialists from the company Tevian, developed a method that can analyze heterogeneous data and accurately identify a person in an image. During model training, researchers supplemented standard multi-camera images with simpler images of people. According to Timur Mamedov, a researcher at MSU’s AI Center and head of silhouette recognition at Tevian, this stylistic diversity significantly improves the quality of the core task.
Smart video analytics systems are already widely used in Smart City projects, analyzing video streams from city streets and public spaces.
