9361907 Ng Field tracking of smoke and obscurant clouds from forest fires or volcanoes requires determination of the two- dimensional cloud extent shown in visible images scanned from various angles about the clouds. At the present time, two-dimensional cloud extent is determined by an operator who manually locates points on the cloud perimeter -- a procedure which has been found to be expensive, time consuming and vulnerable to human misjudgment. To address this problem, this project undertakes a study for the development of computer vision algorithms based on human preattentive visual processes for the automatic recognition of smoke clouds. The objectives of this research include a study of human preattentive vision to determine important visual cues, incorporation of these visual cues into computer vision algorithms for detecting cloud regions, development of a neural network pattern recognition system utilizing these algorithms, and integration of the system with a user-friendly graphical interface. The innovative use of image analysis features based on human preattentive visual processes for automated cloud detection is significant. Since features modeled on human visual preattention are capable of functioning in a scene- independent manner, a system for automatically recognizing obscurant clouds that is superior to present methods will be demonstrated. This is an SBIR Phase I project. ***
