Human identification system based on moment invariant features

Video surveillance is an active research topic in computer vision. Recent research in video surveillance system has shown an increasing focus on creating reliable systems utilizing non-computationally expensive technique for detecting and observing humans' appearance, movements and activities. In this paper, we present a human identification technique suitable for video surveillance. The technique we propose includes background subtraction, foreground segmentation, feature extraction and classification. First of all, we extract all foreground objects from the background. Then, we perform a morphological reconstruction algorithm to recover the distorted foreground objects. The feature extraction is done using affine moment invariants of full body and head-shoulder of the extracted foreground objects and these were used to identify human. When the partial occlusion occurs, although feature of full body cannot be extracted, still the features of head shoulder can be extracted. Thus, it has a better classification on solving the issue of the loss of property arising from human occluded easily in practical applications. The experiment results show that this method is effective, and it has strong robustness.