Real-Time Image Mosaic Based On CUDA

Navigation and control method is one of the key technologies for mobile robots. To navigate and control robots sufficiently, how to get environment images with big eyeshot in real time is an important problem to solve. One of the effective way is image mosaic. During the image mosaic process, there are vast amounts of data access and complex calculations, so it is a difficult task to math the real time requirement using serial algorithm only. Thus, with the background of teleoperated ground mobile robots, a real time image mosaic technology based on CUDA (Compute Unified Device Architecture) is researched in this paper, including a priori information based self-adaptive image matching algorithm, an improved RANSAC (Random Sample Consensus) algorithm, and its application on the navigation and control of mobile robots.Three primary works are finishd in this paper:1) Considering the practice application of image mosaic, a priori information based self-adaptive image matching method is propose in this paper. It selects matching points in the overlap region of left image symmetrically, and searches the corresponding points in right image by rotation unalterable similarity calculation. Experiment results show that the priori information based self-adaptive image matching method can improve the matching speed and accuracy effectively.2) Considering the existence of the miss matching points, which can't be avoid absolutely, a improved RANSAC algorithm is propose in this paper to calculate image transform model, in which random sample restrictions and the optimum swatch set select rule are introduced. Experiment results show that the improved RANSAC method can get a more exactive image transform model.3) Based on the comprehender of Compute Unified Device Architecture, a CUDA based real time image mosaic algorithm (CUDA-RTIM) is proposed in this paper, in which the matching points searching and improved RANSAC based image transform are implemented parellelly to make use of GPU's parallel process ability. Experiment results show that CUDA-RTIM is a real time algorithm with robustness, and it is applicable to various invironment images.Finally, the mobile robot teleoperate system, with real-time and big eyeshot image information is implemented based on Pioneer-III robot. To control the robot accurately, a target points orientate method in matched image is propose.