Moving target detection based on temporal-spatial information fusion for infrared image sequences
wuqin toing,jinyu xiong,a zeng,xiaoping wu,haopeng xu
DOI: https://doi.org/10.1117/12.834112
2009-01-01
Abstract:Moving target detection and localization is one of the most fundamental tasks in visual surveillance. In this paper, through analyzing the advantages and disadvantages of the traditional approaches about moving target detection, a novel approach based on temporal-spatial information fusion is proposed for moving target detection. The proposed method combines the spatial feature in single frame and the temporal properties within multiple frames of an image sequence of moving target. First, the method uses the spatial image segmentation for target separation from background and uses the local temporal variance for extracting targets and wiping off the trail artifact. Second, the logical "and" operator is used to fuse the temporal and spatial information. In the end, to the fusion image sequence, the morphological filtering and blob analysis are used to acquire exact moving target. The algorithm not only requires minimal computation and memory but also quickly adapts to the change of background and environment. Comparing with other methods, such as the KDE, the Mixture of K Gaussians, etc., the simulation results show the proposed method has better validity and higher adaptive for moving target detection, especially in infrared image sequences with complex illumination change, noise change, and so on. © 2009 SPIE.