In this paper, a region correspondence approach is proposed for use in image sequence analysis to remove false alarms in target detection. Object segmentation techniques based on thresholding are performed first. A range of threshold values is determined according to different image backgrounds which are classified by gray level contrast and cooccurrence matrices. Hierarchical clustering is then used to construct effective feature-based tracking models. Local least-mean-square-error estimation is finally used to track the targets continuously. High target detection rates and no false alarm prove the feasibility of the proposed approach for real applications.