SBIR-STTR Award

VSI proposes a flexible sensor estimation framework for airborne calibration processing of image streams over long time periods. The components of the framework will consist of existing software and libraries. The plugin-based framework provides flexibility to switch the components easily. The core of the framework is a belief propagation engine to globally optimize the sensor parameters. The input to the approach is an image stream and its associated navigational data when available. The proposed framework is an online system wherein the sensor estimation is continuously refined as more spatial overlap is discovered. The proposed system can be dynamically tuned for either speed/efficiency or accuracy, depending on the requirements of the operator. The central tenets of the proposed approach are (i) use temporal ordering of the images to increase the computation efficiency, (ii) use spatial overlap of the images which are temporally disjoint to correct for any accumulated drift and (iii) to locally optimize the sensor parameters and use the belief propagation to globally optimize the sensor parameters for hundred thousands of images.

Benefit:
Sensor estimation capabilities for airborne image acquisition systems for Remote Sensing Applications. On-board processing capabilities for updating traffic reports.

VSI proposes the development of the SIGMA (Speedy Imagery Geo-registration and Motion Analysis) system, which will address three key requirements by the end of Phase II: (i) adapt to different type of sensors, sensor-configurations and sensor operating conditions, (ii) handle large amount of data in real-time or close to real-time using limited resources, and (iii) operate under conditions such as lack of metadata in case of GPS-denied areas and linear, unbounded trajecto-ries. The central tenet of the proposed approach will be formation of a graph where the nodes will be a group of images and these nodes will be connected to each other if there is a spatial or temporal connectivity. Furthermore, the development in Phase II will leverage Phase I develop-ments, specifically the robust online SFM system, which was successfully tested on multiple da-tasets such as CLIF 2007, MAMI and synthetic datasets, and (ii) the demonstrated feasibility of the dynamic graph- representation to handle large amounts of image data and variable camera geometry constraints. The proposed system is a plugin-based framework which will allow each of the components to be replaced by different implementations or algorithms to allow the system to work flexibly under different operating conditions.

Benefit:
Real-time mapping of aerial imagery for urban planning, construction and disaster relief and real-time traffic updates.

Keywords:
Graph Optimization, Spatio-Temporal Buindles, Opencl, Gps-Denied Areas, Camera Arrays, Sigma