Automated Smoke and Obscurant Cloud Recognition Using Machine Vision

The objective of this program is to develop advanced image processing/analysis technology for the automatic determination of smoke and obscurant cloud spatial extent and the extraction of relevent features from video images produced by multiple visible and infrared sensors viewing a cloud from different angles. The 2D information is used to quantify the dimensions, volume, centroid, and location of the cloud. Phase I tasks will mainly include: a review of previous work particularly within DoD and in related applications such as meteorology; an analysis of application requirements; an investigation and limited prototyping of required image processing/analysis approaches. This will include proposing and evaluating image segmentation schemes that can reliably identify image regions corresponding to cloud cover, approaches for region or boundary representation and description, and extraction of geometric/topological features or other 2D measures such as cloud density. The Phase I prototyping will be performed on state-of-the-art image processing/analysis workstations implemented on standard computer platforms. Videotaped images will be used for testing the prototypes and evaluating the applicability and feasibility of the proposed schemes. On the basis of the Phase I results, needs for future R&D will be identified and the Phase II effort will be planned.