SUMMARY

Camera systems in motion are subject to significant blurring effects that lead to a loss of information during the image capture. This is especially damaging for optical character recognition for which edge preservation is critical to achieving a high recognition rate. Using non-blind motion deblurring, a trajectory and point spread function can be designed to maximize the recognition rate while meeting endpoint constraints. Optimization through the use of radial basis function networks can therefore be used as a way to find ideal trajectories to reduce blurring effects and preserve text sharpness. This work investigates this problem using simulation of a blurred image capture process. The simulation is automated using radial basis function network optimization and a genetic algorithm to determine trajectories with the best recognition rate. Optimized trajectories yielded recognition scores with up to 57.3% improvement in simulation compared to an analogous linear profile. These results were then verified through physical experimentation with a real-world, controlled-blur image capture process that yielded up to 29.4% improvement across the same comparison. Results were then analyzed using spectral analysis to understand why the chosen trajectories preserve text edges. These findings can be applied to a wide variety of controlled mobile camera platforms, such as autonomous automobiles or unmanned aerial vehicles, to improve their ability to gather information from their environment.