Automated Entity Classification in Video Using Soft Biometrics

SET will identify and evaluate several gait- and face- based soft biometrics acquisition methods that are invariant to view points, range and illumination. This Phase 1 effort will focus on evaluating in-house software and newly developed algorithms to extract a number of soft biometrics from surveillance and other videos, demonstrate the potentials for reliable and fast retrieval and recognition of humans with matching soft biometrics signatures in reasonably unconstrained scenarios, and identifying technology enhancements for Phase II. The focus will be on evaluating the effectiveness of soft biometric signatures for different scenarios, depending on whether live visible or infrared video is available or a recorded video is being processed. We will analyze algorithms for determining the gender, height, cadence and stride parameters of the subject using gait analysis. When the subject’s face images are available, we will verify gender determined using gait, and determine the hair and eye color. By evaluating various soft biometrics extraction techniques using publicly available datasets, we will determine those that are most informative and better understand their working conditions. In Phase II, we will integrate a human authentication and verification system prototype for real time and automatic soft biometrics acquisition, indexing and retrieval functionalities.