Histogram-Based Segmentation For Stationary Target Detection In Urban Environments

Detection of stationary targets in urban sensing and through-the-wall radar images using likelihood ratio test (LRT) detectors has recently been considered in the literature. A shortcoming of the LRT detectors is that appropriate probability density functions of target and clutter images need to be predefined. In most practical scenarios, this information is not available a priori, and the mismatch of the assumed distribution functions degrades the performance of the LRT. In this paper, we apply image segmentation techniques to radar images of scenes associated with urban sensing. More specifically, the Otsu's method and maximum entropy segmentation are considered to aid in removing the clutter, resulting in enhanced radar images with target regions only. Performance of the segmentation schemes is evaluated and compared to that of the assumed LRT detector using real-data collected with Defence Research and Development Canada's vehicle-borne through-the-wall radar imaging system. The results show that, although the principles of segmentation and detection are different and serve disparate objectives, the segmentation techniques outperform the LRT detector for the considered cases.