基于CUDA加速的并行人脸检测-计算数学专业论文.docxVIP

华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract As an important issue in computer vision and pattern recognition, face detection has been used widely in information security, video surveillance and biometrics, etc. The Ad- aBoost algorithm based face detection method proposed by Viola et al in 2001 is the first and one of the most popular techniques for real-time face detection. In the present thesis, we parallelize the AdaBoost based face detection method and implement the parallelized method on NVIDA GPU by CUDA. In the framework of the presented parallelized face detection method, the first step is converting the input color image to gray model, then the gray image is resized to a pyramid image at different scales so that all size faces in the original image can be detected efficiently when using fixed-size scanning windows. Both color space converting and image resizing are implemented on GPU with parallelized method, which is hundreds of times faster than the CPU implementation. The next step is creating integral image and squared integral image of the pyramid image on GPU. The parallel prefix sum algorithm and image transposing method are adopted so that the integral image and squared integral image can be computed efficiently. With compared to the CPU-based integral image computation, our method is much faster. When detecting faces with the integral image and squared integral image, the kernels on GPU scan and evaluate all sub-windows in parallel. In this step, the kernels can be accel- erated by storing the classifiers in the constant memory and mapping the integral image to the shared memory on GPU. We also designed a efficient method to processing the detecting result using a bit array. The experiment results showed that the parallel face detection proposed in this thesis is more efficient than the serial face detection method, and could achieve an over 30x speed-up compared with the CPU implementation. Moreover, our method could achieve real-time face detec