RealNetworks

Speed Matters: Video Facial Recognition

Current Facial Recognition (FR) evaluations performed by the National Institute of Standards and Technology (NIST) use still photos, not from video. However, successive frames of a video stream can provide a single context of and individual's presence such that consecutive recognitions of the same individual could be made with increasing confidence. The student team worked to redefine FR accuracy for real-time video and evaluate the hypothesis that accuracy compounds with successive frames of a live video feed or develop an alternate formula.