Combining Image Processing with Signal Processing to Improve Transmitter Geolocation Estimation

This research develops an algorithm which combines image processing with signal processing to improve transmitter geolocation capability. A building extraction algorithm is compiled from current techniques in order to provide the locations of rectangular buildings within an aerial, orthorectified, RGB image to a geolocation algorithm. The geolocation algorithm relies on measured TDOA data from multiple ground sensors to locate a transmitter by searching a grid of possible transmitter locations within the image region. At each evaluated grid point, theoretical TDOA values are computed for comparison to the measured TDOA values. To compute the theoretical values, the shortest path length between the transmitter and each of the sensors is determined. The building locations are used to determine if the LOS path between these two points is obstructed and what would be the shortest reflected path length. The grid location producing theoretical TDOA values closest to the measured TDOA values is the result of the algorithm. Measured TDOA data is simulated in this thesis. The thesis method performance is compared to that of a current geolocation method that uses Taylor series expansion to solve for the intersection of hyperbolic curves created by the TDOA data. The average online runtime of thesis simulations range from around 20 seconds to around 2 minutes, while the Taylor series method only takes about 0.02 seconds. The thesis method also includes an offline runtime of up to 30 minutes for a given image region and sensor configuration. The thesis method improves transmitter geolocation error by an average of 44m, or 53% in the obstructed simulation cases when compared with the current Taylor series method. However, in cases when all sensors have a direct LOS, the current method performs more accurately. Therefore, the thesis method is most applicable to missions requiring tracking of slower-moving targets in an urban environment with stationary sensors