Reidentification of ground targets in EO/IR- imagery

targets in aerial surveillance data being developed at FOI. The work has been conducted within the project "ARCUS" and "Autonomous Surveillance". A complex signal processing chain containing the following subparts is discussed: detection, multi-target tracking, visual tracker, segmentation, view alignment, feature extraction, reidentification, object database, planner, and operator interaction. The report describes the status of four areas:  The implementation of an initial reidentification method based on CCM (color co-occurance matrices).This work has been verified against simulated sensor data in the visual range.  Integration of the initial reidentification method with modules for multi-target tracking using HLA.  The development of a novel reidentification method based on learning effective combinations of local descriptors of shape and color. This method has been evaluated against an international database containing thousands of color images of hundreds of civilian cars. Based on test results, we believe the performance of this method to represent the state-of-the-art in reidentification.  A field trial has been conducted to systematically register cars entering the parking lot at FOI. Each car has been registred from different orientations using several optical sensors within the thermal and visual ranges. The field trial and an annotation method for collected data are described.