Hidden Markov Model Parameter Estimation for Multiple Dim Target Detection

This paper presents a modified hidden Markov model (HMM) filtering algorithm for detecting multiple dim targets in image sequence under low SNR condition. The proposed algorithm consists of three steps. As a first step, morphological filtering is applied for extracting features in pre-processing level. The second step is a hidden Markov model filter. To enhance a detecting performance of the filter, state transition probability matrix of HMM filter is updated to the re-defined parameter in each single recursive process. The estimation process uses potential target’s local path from several continuous frames. The last third step is subwindow application. When a target is detected, the target is treated by sub-window to apply individual HMM filtering for detecting multiple targets. Based on numerical results, the proposed algorithm has slightly better detecting performance for multiple targets from a sequence of an image sensor under very low SNR( 2 ; ) conditions