CHARLES RIVER ANALYTICS, INC. — Department of Defense SBIR Phase I: A19-048

Description

The US Army is updating the capabilities of its combat fleet with its Next Generation Combat Vehicle (NGCV). A core component of the NGCV effort is upgrading the sensing and situational awareness capabilities of combat vehicles, aided by higher resolution electro-optical, infrared, and thermal cameras that can be mounted 360 degrees around the vehicle. Existing Aided and Automatic Target Detection and Recognition (AiTD/AiTR) technology has trouble processing video imagery from on-the-move ground vehicles because these algorithms only support stationary cameras. To mitigate the problems associated with moving cameras and background motion relative to the sensor, we must model both the camera pose and scene structure to enable existing AiTD/AiTR algorithms to work on moving future combat vehicles. Our Low-SWAP-C Background Estimation Labeling System (LABELS) solution is a very fast, accurate, and stable method of visual odometry (VO) that compensates for vehicle motion. By knowing the precise rotation and translation of the vehicle in a scene, LABELS presents a relatively static view of the world as an input to existing AiTD/AiTR algorithms, like a sliding window around a 360-degree view of the vehicle, enabling these algorithms to detect and track targets of interest in real time.