Template Generation From A Small Number of Views

"The amount and diversity of data required for meaningful evaluation of automatic target recognition (ATR) systems under extended operating conditions makes data collection efforts prohibitively expensive and time consuming. Generating novel views oftargets from a video segment, or several still images acquired from different viewpoints, will significantly reduce the data collection complexity in evaluation and validation of ATR algorithms. Technical challenges include feature correspondence, targetmodel construction, camera calibration, target model update, and scene synthesis. Promising solutions are based on structure from motion (SfM) and view synthesis using tensor analysis. In Phase I, we propose to investigate and enhance state-of-the-artalgorithms for generating novel views from a video segment or multiple still images. Novel solutions for recovery of relative structure from video and subsequent novel view synthesis, and direct novel views synthesis from still images using trilineartensor approaches will be developed. Emphasis will be on the utility in the design, experimentation and validation of ATR algorithms. During Phase II, a prototype system will be built by integrating the most promising approaches and sensor models. Theprimary military applications for the proposed work are in generating novel data sets for validating ATR systems and building models of targets and scenes for simulation of battlefields. Commercial ap