Obstacle Representation Database From Sensor Data

PercepTek proposes the development of a software package that will process 3D point data from a range sensors such as LADAR or stereo to produce formats more suitable for navigation algorithms. The approach is based on an octree data structure used by PercepTek on its 3D from Video project. The representation serves as an efficient access mechanism for an extensible set of model primitives ranging in complexity from the input 3D points to planar patches or to more complex structures such as buildings or trees. Because this representation is 3D, it will support 3D planners as well as the 2D and 2.5D planners in common use today. This octree data structure easily handles hierarchical descriptions of the model as well as simplified representations of objects (e.g., bounding boxes) useful for rapid intersection testing. The Phase I effort will include a study of existing planning algorithms to determine their requirements, but will focus on the development of a prototype allowing evaluation of the accuracy, memory requirements, and computational requirements of the modeling process and the query response process. Evaluation will be based on experiments using simulated and real data of the McKenna MOUT site from the ACO data collection.