Abstract
In this paper, we detect persons in unknown circumstance using HMM that is theoretically stabilized in speech recognition and has high recognition rate in the time-series pattern recognition. To apply HMM to person detection, we propose Pmodel (Person Model) that have the time-series consisting of head, shoulder, arm, and leg. Since images have 2-Dimension, we use PHMM (pseudo 2-D Hidden Markov Models), and obtain its observation vector from DCT. On the other hand, we make use of the skin color to detect persons without studying a background and regardless of stopping or walking people. Then, we propose PD Filter (Person Detection Filter) in order to estimate their skin color. Finally, we recognize persons using Pmodel that applies the regions extracted by PD Filter to PHMM.
