A novel approach based on spectral-temporal information fusion for multi-target detection

This paper describes a novel particle filter-based track-before-detect (TBD) technique for multiple dim moving targets with fluctuating intensity in multi-spectral IR image sequence. The proposed approach consists of a combination of a dimension reduction of multi-spectral and a particle filter. In our novel algorithm, a generallized likelihood ratio test (GLRT) method is first applied to the original multi-spectral IR image to reduce dimension and provide the priori of targets. A TBD method based on particle filtering is then applied to the sequence obtained from dimension reduction of raw multi-spectral IR image. Experimental results are presented for the proposed algorithm for detecting and tracking three dim moving targets with varying intensity in multi-spectral IR image sequence of 20 bands and 30 frames. The results show that the proposed algorithm can detect and track multi-targets with a signal-to-noise ratio (SNR) as low as 3dB.