光学 精密工程, 2016, 24 (11): 2863, 网络出版: 2016-12-26
计算机图形处理器加速的光学航空影像正射校正
Orthorectification of optical aerial images by GPU acceleration
航空影像 正射校正 计算机图形处理器(GPU) 并行算法 有效像素区域 orthorectification Graphic Processing Unit(GPU) parallel algorithm effective pixel region
摘要
提出了计算机图形处理器(GPU)加速的光学航空影像正射校正并行算法, 以满足获取光学航空影像对实时性的要求并提高对海量影像数据在CPU上串行正射校正的效率。介绍了光学影像正射校正算法原理以及正射校正算法的并行化处理。为减少GPU执行的计算负载, 引入“有效像素区域”概念, 设计了改进的GPU并行校正算法。通过配置选择以及存储器访问优化进一步提高了算法的执行效率。最后, 分析了GPU并行算法的精度, 并验证了噪声干扰对算法的影响。实验结果表明, 优化的改进GPU并行算法显著提高了正射校正的速度, 影像大小为5 000×5 000时, 加速比最高可达CPU串行算法的223倍以上。虽然GPU单精度计算和噪声干扰会使影像校正精度有所下降, 但尚在误差允许范围之内。该算法能够快速实现光学航空影像的正射校正, 校正后的影像满足实际应用需要。
Abstract
An optical aerial image orthorectification parallel algorithm by Graphic Processing Unit(GPU) acceleration was presented to improve the image real-time processing ability and the serial orthorectification efficiency for massive image data on a CPU. The principle of optical image orthorectification algorithm was introduced, and the parallel processing of orthorectification algorithm was described. To reduce the computational load of GPU execution, the concept of “effective pixel region” was introduced and an improved GPU parallel correction algorithm was designed. Then, the efficiency of the algorithm was improved through configuration options and a memory access optimization. Finally, the algorithm precision was analyzed, and impact of the noise on the algorithm was verified. The experimental results show that the optimized improved-GPU parallel algorithm significantly improves the speed of the correction. When the image size is 5 000×5 000, the speed up is 223 times as compared with the CPU serial algorithm. Although the GPU single precision calculation method and noise interference will cause the serious decline of correction precision, it is still in an allowable error range. As a result, the GPU algorithm implements the orthorectification of optical aerial images rapidly and the corrected images satisfy the need of practical applications.
全吉成, 王平, 王宏伟. 计算机图形处理器加速的光学航空影像正射校正[J]. 光学 精密工程, 2016, 24(11): 2863. QUAN Ji-cheng, WANG Ping, WANG Hong-wei. Orthorectification of optical aerial images by GPU acceleration[J]. Optics and Precision Engineering, 2016, 24(11): 2863.