This paper presents a novel biometric system for real-time walker recognition using a pyroelectric infrared sensor, a Fresnel lens array and signal processing based on the linear regression of sensor signal spectra. In the model training stage, the maximum likelihood principal components estimation (MLPCE) method is utilized to obtain the regression vector for each registered human subject. Receiver operating characteristic (ROC) curves are also investigated to select a suitable threshold for maximizing subject recognition rate. The experimental results demonstrate the effectiveness of the proposed pyroelectric sensor system in recognizing registered subjects and rejecting unknown subjects.
Biometrics;Maximum likelihood estimation;Object recognition;Principal component analysis;Real time control;