A Method of GPU-CPU Co-processing Rapid Images Mosaicking Based on Valid Areas

A method for rapid mosaicking aerial images is introduced, based on GPU-CPU co-processing and a seamline network based on Voronoi diagrams, taking valid areas of the images into account. A seamline network is created based on a Voronoi diagram with the valid areas of the images, solving the problems occuring when some areas of a mosaiced image are not covered after mosaicking. This approach is based on the traditional Voronoi diagram network under low overlap conditions. Orthorectification is embedded in the mosaicking process during GPU-CPU co-processing, and only the valid area of each image is orthorectified. Meanwhile, the mosaicking efficiency is further improved by optimizing the configuration and the hierarchical memory of the GPU. Experimental results indicate that efficiency was increased by 20 times when mosaicking 237 high-resolution aerial images, and much faster than traditional methods which complete orthorectification of the aerial images before mosaicking. The precision of the mosaicked DOM was very high, thus meeting emergency mapping requirements.