基于CUDA GPU的多摄像机场面航空器识别加速算法

摘要

基于摄像机的航空器识别是机场场面监视的重要工具.针对多摄像机场面航空器识别算法存在的计算效率低等缺点,提出基于GPU CUDA的加速算法.利用CUDA线程并行处理能力与GPU计算能力,对算法进行了重新设计与优化.通过实地对多路场面视频监视数据进行了多次实验,验证了在NVIDIA Geforce 8800GTS显卡上可实现10倍以上的加速性能,提高了航空器目标识别效率,可以满足机场场面监视中对航空器识别与跟踪的实时性要求.%One of the important surface surveillance tools is aircraft detection using camera in airport. The accelerating algorithm using CUDA technology on graphics processing units (GPU) was proposed to improve the speed of aircraft detection algorithm based on multi-camera. The aircraft detection algorithm is redesigned and optimized using multi-thread parallel processing capability of CUDA and computation capability of GPU. The true surveillance data from airport video is used to test the detection algorithm many times. The results validated that more than 10-fold speedup compared with CPU are obtained by using NVIDIA Geforce 8800 GTS. This approach can improve efficiency of target detection in airport surface, and satisfy with the real-time requirements of airport surface surveillance.