Research on Fast Image Mosaic Based on CUDA

To get sufficient environmental information for a teleoperated unmanned vehicle, a matched image with wide field and high quality image is necessary. Image matching is a key point in image mosaic. And the vast amounts of data and complex calculations make it bottlenecked to get a high speed on mosaicing images. Considering the requirements of real-time image mosaic, a self-adaptive image matching method, considering the priori information on the spatial relationship between images, is proposed in this paper. The overlapping region is used as one of the constraint to reduce the search range during the image matching process. And using the General Purpose Graphic Process Unit (GPGPU) to accelerate complex computations, is becoming a research focus. In this paper, the algorithm of image matching is parallelized based on Compute Unified Device Architecture (CUDA), which is a platform of GPGPU programming. Experiment results show that, compared with the serial scheme on CPU, the efficiency of image mosaicing, implemented with the parallel scheme on Graphic Process Unit (GPU), is improved more than 30 times, with 12.8 frames per second.