CAJ | 학술논문

针对实际人脸识别系统需要满足实时性的应用需要，探讨了在图形处理器（ GPU）硬件架构基础上的基于主成分分析（ PCA）人脸识别系统设计与实现。结合统一计算设备架构（ CUDA）的计算平台，通过将算法中耗时长、适合并行的部分过程映射到GPU上并行执行改进系统的加速实现。实验结果表明：相对于基于CPU平台的串行实现，基于GPU的实现在整体上能够获得约5倍的加速，而两个执行并行的模块能分别获得最大20倍和30倍的加速。
침대실제인검식별계통수요만족실시성적응용수요，탐토료재도형처리기（ GPU）경건가구기출상적기우주성분분석（ PCA）인검식별계통설계여실현。결합통일계산설비가구（ CUDA）적계산평태，통과장산법중모시장、괄합병행적부분과정영사도GPU상병행집행개진계통적가속실현。실험결과표명：상대우기우CPU평태적천행실현，기우GPU적실현재정체상능구획득약5배적가속，이량개집행병행적모괴능분별획득최대20배화30배적가속。
Aiming to the application requirement of the real-time facial recognition systems, the implementation of principle component analysis ( PCA ) based face recognition in graphics processing unit ( GPU ) was exploited.With Compute Unified Device Architecture ( CUDA) computational platform, some time-consuming modules are mapped to GPUs for parallel processing for acceleration.Experimental results demonstrate that, when comparing to the CPU-based serial implementation, the GPU-based realization could achieve about 5 times speedup in all with an utmost 20-30 times speedup for the two parallel processing modules.