Estimating body related soft biometric traits in video frames

Area of Science:

• Computer Vision
• Biometrics
• Artificial Intelligence

Background:

• Facial recognition is often hindered in surveillance due to varying challenges.
• Soft biometrics offer an unobtrusive alternative for recognition systems.
• Body-related information can be inferred from visual appearance at a distance.

Purpose of the Study:

• To develop a novel approach for estimating body weight using artificial neural networks and body measurements.
• To integrate height estimation using single-view metrology in low frame-rate videos.
• To assess the effectiveness of body-related soft biometrics for subject prediction.

Main Methods:

• Utilized artificial neural networks for body weight prediction based on body measurements.
• Incorporated single-view metrology for height estimation from video data.
• Evaluated the approach on a newly compiled dataset of 1120 frames from 80 subjects.

Main Results:

• Demonstrated adequate prediction of body weight using the proposed artificial neural network model.
• Successfully estimated height from low frame-rate videos.
• Validated the feasibility of predicting soft biometric information from video frames.

Conclusions:

• Body-related soft biometrics can be effectively predicted from video frames.
• The proposed method offers a promising solution for unobtrusive biometric identification in visual surveillance.
• This research contributes to advancing soft biometric techniques for practical applications.