A novel method for pedestrian tracking for infrared image sequences

The process of enhancement consists of three component elements: de-noising based on wavelet, image fusion, contrast enhancement based on multi-scale wavelet transform. Because the wavelet transform has character of multi-scale, multi-resolution analysis and time-domain localize, image enhancement methods based on wavelet transform can improve conventional methods of image enhancement. This paper presents a method for accurate pedestrian tracking for infrared image sequences. It makes use of the wavelet entropy features of pedestrian regions based on the dual-tree complex wavelet transform and is based on a particle filter framework. The method constructs the regions of interest's (ROI) representation using wavelet entropy, so as to hurdle the disadvantage of insufficient information when only intensity feature is considered. Furthermore, it correctly updates the above mentioned pedestrian representation model which is embedded in the particle filter framework and propagate sample distributions over time. The experimental results show that the method is effective.