부산대학교 도서관

Video dehazing is an important preprocessing task to improve the visibility of various UAV videos for further processing, such as target recognition and tracking. State-of-the-art Guided Filter based method employs dark channel prior to infer the direct scene transmission parameter and refines it using guided image filter. The quality of the dehazing result from this method is satisfactory except some artificial effect in large sky area. Meanwhile, the computational efficiency is not very high and the real-time performance is not reached. In this paper, a parallel framework for video dehazing algorithm is proposed to accelerate the dehazing computation on UAV ground station. Firstly parallel O(1) complexity local minimum filter is employed to get the initial dark channel image, which is further refined by parallel Joint Recursive Bilateral Filter. Combined with the atmosphere parameter which is obtained by histogram based estimation, the dehazing result is finally achieved. The proposed method is evaluated on multi-core UAV ground station using C++ programming language with OpenMP compiler directive. Experimental results show that the proposed method outperforms available Guided Filter based method and has a real-time performance.