Dense computing task analysis of multi-view matching method and GPU implementation

For the parallel processing of multi-view matching method of high resolution digital imagery, dense computing tasks and crucial parallel computing aspects were analyzed. And then, based on GPU implementation scheme of single stereo matching method, an improved GPU implementation scheme was proposed. Based on practical image datasets, experiments were done and the experimental results proved that the acceleration rate of the improved GPU implementation scheme could reach 20 to 100 for dense computing tasks, and compared with single CPU processing, possessed greater processing superiority.