Design And Implementation Of Fisheye Video Stream Real-time Correction In Android Environment

With the capability improvement of smartphone and the expansion of network bandwidth,the fisheye video corrected system,which can significantly improve the observability of video,is possible to operate on Android platform.Traditional distortion correction algorithm of fisheye camera cannot be processed in real-time on the Android mobile phone platform because their high computational complexity overwhelm smartphone's capability.So Realizing the real-time fisheye video surveillance system on the Android smartphone is a big challenge.In view of the above problems,the dissertation uses the mobile GPU parallel acceleration strategy to efficiently process the decoded video images,realizing the real-time correction of the fisheye video stream of 1280*1280 resolution,and correcting the smooth playback of the video without the stagnation.The main work of this dissertation is as follows:(1)The principle of fisheye camera imaging and camera calibration methods are studied.The performance of the two algorithms is evaluated by performing a double-longitude algorithm and a cubic box mapping algorithm on the Android platform.The experimental results show that compared with the double-longitude algorithm,the cubic box mapping algorithm has faster training speed and higher correction accuracy,and can be used for accurate correction of fisheye camera video stream.(2)Aiming at the problem that the cubic box mapping algorithm is too time-consuming in the single frame correction of the Android platform,the direct mapping method is adopted in this dissertation.According to the mapping relationship between the corrected planar pixel points and the fisheye image coordinates,the mapping is performed to realize the video calibration,and the correction time is reduced.purpose.In order to verify the effectiveness of the method,the fisheye image with a single frame resolution of 1280*1280 was corrected and compared with the traditional cube box mapping algorithm.The results show that the cubic box mapping algorithm based on direct mapping can have faster correction speed while ensuring the same correction accuracy.(3)The problem of serial-based direct mapping based cube box mapping algorithm is time consuming.The programming framework of OpenCL heterogeneous computing is studied.The bilinear interpolation part of the direct mapping method is designed in parallel using the mobile phone's onboard GPU.To further improve the efficiency of the operation,the thread block size is reasonably set according to the actual GPU performance.The performance of the optimized parallel algorithm was tested by correcting the fisheye image with a single frame resolution of 1280*1280 and compared with the serial algorithm using only the CPU.The experimental results show that the parallel bilinear interpolation direct mapping method based on mobile phone CPU-GPU has faster correction speed,lower CPU usage,meets the requirements of real-time correction of fisheye video,and can be used for real-time correction of fisheye video stream.Based on the above work,this dissertation designs and implements a real-time correction software for fisheye video stream based on Android platform,which can realize RTSP video stream decoding and real-time correction.The performance of the softness is verified by the image correction of 1280*1280 fisheye.The results show that the running speed reaches 20 frames/second,and the software has good application prospect and wide application value.