Research And Realization Of Omnidirectional Image Unwrapping Parallel Optimization Algorithm

In recent years, With the development of the microprocessor towards to the multi-core direction, multi-core processors are proliferated across different domains. Nowadays, the development technology of the multi-core processor, bring the new trend in the development of the high performance computing. Therefore, it's no doubt that if multi-core technology is applied to the omnidirectional vision system, it will play a crucial role for its performance optimization.Omnidirectional image unwrapping algorithms are complex in calculation, so their unwrapping time is not satisfied on the single processor. Because of the serial unwrapping algorithms itself, it can't take full advantage of the high performance computing of the multi-core processor. So it can't solve the problem of large calculation. This will limit the application range of the omnidirectional vision technology.To solve this problem, this paper used multiple-core processor, studied a method of omnidirectional image unwrapping algorithm based on the edge-smoothness quadratic interpolation including the table-checking method. Through analysing the lack of the task performance of omnidirectional image unwrapping algorithm on the large calculation part, and the feasibility of its parallelization, focused on studying the parallel unwrapping algorithm for yielding the panorama image. After proposing the parallel strategy, realized it in parallel processing by OpenMP program. Above the execution time, measure the parallelization performance of the unwrapping algorithm with the speedup and parallel efficiency.In order to further the parallel unwrapping algorithm, this paper also improved the serial unwrapping algorithm. Analyzing the unwrapping time, parallel performance and rendering speed with the unwrapping experiment in different resolution, we can get the result that the improved algorithm is better than the unimproved, no matter in serialization or in parallelization. The speedup of the parallel unwrapping algorithm is up to 90.2%, higher 13.045% than the unimproved one. Benefit from the improving, the omnidirectional image with high resolution 1024×768 can realize the processing speed on 26.281 fps. Here, the serial and the parallel unwrapping time is 71.06ms and 38.05ms respectively, while it's 120.4ms and 73.08ms before improving it.Parallel optimization algorithm improved the whole unwrapping performance of the omnidirectional vision software, effectively solved the problem that the large calculated amount of the omnidirectional image unwrapping algorithm affected the image unwrapping time.