Accelerating Multi-Sensor Image Fusion Using Graphics Hardware

This paper shows approaches to accelerate pixel-level image fusion speed using graphics hardware. Recently, to improve visibility through maximization of information collected through development of various sensors and improvement of sensing technology, the importance of not only development of new fusion algorithm but speed of fusion process is increasing. Though specialized fusion boards for real time fusion processing are already developed, but they have disadvantages such as expensive price and lack of scalability. These disadvantages can be replaced by GPU (Graphics Processing Unit) that have good price/performance ratio, hardware programmability, enormous computing power and speed. Fifteen fusion methods were used for the tests that give numerical data regarding comparison of GPGPU (general-purpose GPU), CUDA (the latest architecture of GPU) with traditional CPU-based implementations. The evaluation results prove GPU acceleration to be much faster than CPU-based multi-threading.