Options
2004
Presentation
Titel
Smart recognition assistance for multisensor-image-based reconnaissance
Titel Supplements
Presentation held at the 6th Joint International Military Sensing Symposium MSS. October 18-21, 2004 Dresden, Germany
Abstract
Image-based recognition of objects means selection out of a variety of object hypotheses by exploitation of signals from imaging sensors. Such hypotheses can be expressed by object class, state, and behaviour. In order to support the recognition process, the set of object hypotheses has to be mapped into a set of observable object features (recognition keys) with respect to characteristics of the available set of sensors. In the domain of reconnaissance the task of image-based object recognition is and will be performed primarily by humans with their specific skill for visual pattern recognition. Human performance in this task domain can be significantly increased by computer-based assistance. One crucial starting point to support image analysts is to direct their attention to relevant recognition keys. The observability of those recognition keys highly depends on the selected sensor. So the other crucial point is to support the analyst by recommendation of a sensor or sensor-combination most appropriate to the reconnaissance task. Object recognition in general is an iterative process where the analyst observes a set of recognition keys step-by-step. The appropriateness of the available sensors or sensor combinations may be very different for specific recognition keys. Therefore sensor selection should not stop at the beginning of the recognition chain but has to be repeated at each iterative step. The recognition concept RecceMentor of Fraunhofer IITB provides an interactive catalogue of target objects with dynamic access via a system of elaborated recognition keys. The suitability of each key for the recognition progress is computed with respect to its sensor-dependent observability and is indicated to the analyst. So the analyst sees at a glance to which keys he should direct his attention first and what sensor he should choose best. After each decision about a recognition key resulting in a reduction of the set of matching objects, the suitability of all remaining keys is recomputed in order to get an optimal recommendation for the subsequent step. The concept will be demonstrated using the example of airfield reconnaissance according to STANAG 3596 with SAR, IR and VIS imagery.