Digital reconnaissance images gathered by low-altitude over-flights with resolutions on the order of a few feet and fields of view up to 120 degrees can generate millions of pixels per second. Storing this data in-flight, transmitting it to the ground, and analyzing it presents significant problems to the tactical community. One potential solution is in-flight preview and pruning of the data where an operator keeps or transmits only those image segments which on first view contain potential intelligence data. To do this, the images must be presented to the operator in a geometrically correct form. Wide-angle dis-tortion, distortions induced by yaw, pitch, roll and altitude variations, and distortions due to non-ideal alignment of the focal plane array must be removed so the operator can quickly assess the scene content and make decisions on which image segments to keep. When multiple sensors are used with a common field of view, they must be mutually coregistered to permit multispectral or multimode processing to exploit these rich data dimensions. In addition, the operator should be able to alter the apparent point of view of the image, i.e., be able to zoom in and out, rotate, and roam through the displayed field of view while maintaining geometric and radiometric precision. These disparate requirements have a common feature in the ability to perform real-time image geometry manipulation. The role of image geometry manipulation, or image warping, is reviewed and a "strawman" system dis-cussed which incorporates the Pipelined Resampling Processor (PRP). The PRP is a real-time image warping processor discussed at this conference in previous years"2'3". Actual results from the PRP prototype are presented. In addition, other image processing aids such as image enhancement and object classification are discussed as they apply to reconnaissance applications.
|