| The normal operation of outdoor vision systems has been influenced seriously by severe weather and air pollution, particularly, in the industrial area and the city center. Thus, improving the operation of the visual system under heavy weather is imminent. With the comprehensive usage of the outdoor shooting and the video surveillance equipment, the video data needs to be dehazed become bigger and bigger. Aiming at the phenomenon, a higher requirement of the speed of video haze removal algorithms is needed.There are two state-of-art methods among these existing video haze removal algorithms. The first method is to calculate the transmission map, and then we remove each frame of the video haze based on the transmission map. The second method focus on separating the video into foreground and background, and both of them are achieved after removing their haze respectively. The first method is dependent on calculating the transmission map so that the algorithm is difficult to implement in engineering applications. The second method will lead the video to be uneven and distorted after remove video haze. To solve the advantages of these two methods, we use Single Image Haze Removal Using Dark Channel Prior to dehaze the video data. The effect of this algorithm is excellent. This algorithm not only enhances the contrast of the video data but also retains the mist generated visual depth of field. Owing to the low rate of the algorithm, the defog algorithm is optimized and implemented through the TILE_GX36 multicore platform. This system takes advantage of the defog algorithm and the high rate of the parallel computing. The experimental results demonstrate the speed of haze removal is improved greatly. The detailed steps are as follows:Firstly, we use TILE_GX36 multicore platform as video haze removal parallel processing module and then two mainframes is employed as the transmitter and the receiver. Finally, we use gigabit network cable to connect them to build the hardware structures of the video haze removal system. We design the parallel software architecture, based on the system's hardware architecture. The design is as follow:According to realistic requirements, we design a reasonable transport protocol. In order to ensure the reliability of data transmission, we use the Transmission Control Protocol between the mainframes and the multicore platform. Besides, there is also use of TCP protocol between the two TILE_GX36 chips.Using Pthread library which is C language standard library, we make each task parallel.We create two arrays and two queues to manage memory and task scheduling. The system maintains these two arrays which are used to privatize the process space for parallel tasks for reducing the competition of parallel tasks. Using these two queues as the buffer between the parallel processing tasks and data receiving tasks, the system can improve greatly the degree of parallelism by reducing tasks obstruction which is due to the different speed of the different tasks.After testing the speed of video haze removal, experimental results demonstrate that this system has greatly enhanced the video haze removal speed. The video of the resolutions of 600*400 and 720 P has been dehazed in real time. With expanding the hardware architecture, the video of the resolutions of is 1080 P can be defogged timely. Meanwhile, the system has good versatility for different programs, and it can be implemented parallelism after appropriate modifications. In summary, the presented system improves the speed of the video haze removal effectively, and it is of high practically for multicore parallel programs. |
PDF Full Text Request