Abstract: Telepresence with wide field of vision is one of the key technologies for teleoperated unmanned vehicle. To meet the real-time requirement, a fast image mosaic method was proposed, which selects evenly distributed points and calculates their similarity by rotation-invariant NCC (Normalized Cross Correlation) operator, calculates the projective transformation model by RANSAC (Random Sample Consensus) method,and fuses images by linear fade in and fade out. Finally, the algorithm was accelerated with a parallelized self-adaptive image matching method using GPU (Graphic Process Unit). Experiment results show that the efficiency of the parallclized image mosaicing method, compared with that of the serial scheme on CPU, is dramatically improved with more than 60 times on frame rate.

Key words: Image mosaic, Parallel image mosaic, RANSAC, GPU, CUDA