Computationally Efficient Parallax Processing

Automated video processing for airborne surveillance applications requires accurate registration of the video frames. For example, change detection processing for moving target detection requires accurate frame-to-frame alignment of the image plane projections of world points. When the surveillance platform is moving, significant parallax effects are observed due to the varying ranges to and orientations of objects and surfaces in the scene. Thus, advanced algorithms are required to accurately model, estimate, and apply the registration transformations. Toyon’s approach to computationally efficient parallax processing is based on a detailed understanding of projective imaging geometry and robust statistical estimation. Besides frame registration, Toyon’s algorithms can be used to provide additional parallax mitigation capabilities, including identification and, optionally, filling in of missing data due to occlusion effects. All this is achieved with computational complexity that is linear in the number of pixels. When fully developed in commercially available hardware, this technology is expected to provide significant improvements in exploitation of aerial video surveillance data, in real time.