【作者】 郭红健；

【导师】 苏光大；

【作者基本信息】 清华大学 ， 信息与通信工程， 2004， 硕士

【摘要】 随着计算机应用的普及以及图像处理和模式识别领域的研究逐步成熟,人脸识别技术在身份识别、智能监控等领域的应用越来越广泛。在人脸识别过程中,人脸检测与定位是一个重要环节。本论文工作的重点就是设计并实现一个基于DSP的硬件人脸检测与定位系统,对摄像头拍摄的动态视频进行实时人脸检测与定位,给出用于识别的人脸图像区域。本课题实现的系统作为动态人脸识别系统的一个子系统,是进行后续识别工作的基础,它对于提高整个系统的性能有着重要的意义。传统的利用软件处理来实现人脸识别系统中的人脸检测与定位,耗费较多的处理时间,使系统不能达到很好的实时性能,同时也使更复杂的算法开发受到一定的限制。因此,本论文的工作致力于提高图像处理速度方面的研究,利用硬件高速处理的特点,实现实时人脸检测与定位。论文完成了系统的图像采集与人脸检测定位两个部分的软硬件设计开发与集成。设计的基于PCI总线的1:1图像采集卡,实现了1:1方式的图像采集,解决了传统图像采集中数字化图像存在的几何变形问题,有助于提高人脸检测与定位的准确率。1:1图像采集卡作为整个系统的数据采集部分,用于实时显示摄像头拍摄的视频,同时为人脸检测与定位部分提供图像数据。设计的硬件人脸检测与定位卡,采用TI公司的定点DSP芯片TMS320C6201B作为核心处理单元,利用其快速处理数据的能力实现了硬件实时人脸检测与定位。集成的硬件人脸检测与定位系统成功地将人脸检测算法移植到DSP上,每帧图像的处理时间小于30ms,实现了实时人脸检测与定位。此外,论文还对多路并行硬件人脸检测与定位进行了相关的研究与开发。多路并行的人脸检测系统可以同时对多个摄像头输入视频进行实时检测,使系统的性能得到了成倍的提高,减小了系统的体积并降低了成本。更多还原

【Abstract】 Face detection and location is an important step of face recognition process. The emphasis of my dissertation is to design and implement a hardware system of face detection and location based on DSP, which realizes the real-time face detection and location from the dynamic video shot by camera, and gives the region of human face for recognition. This system acts as a sub-system of dynamic face recognition, and serves for the following recognition. It plays an important role in improving the performance of the whole system.It costs much time to realize the face detection and location in face recognition systems using traditional software processing, which results in the low performance of real-time processing and limits the advanced algorithm development. So, my task focuses on the research of speeding the image processing, realizing the real-time face detection and location with high-speed hardware.This project completes the hardware/software development and integration of image acquisition and face detection, the two parts of the system. 1:1 mode image acquisition card based on PCI bus realizes the image acquisition in 1:1 mode, which solves the geometric distortion of digital images in traditional image acquisition. It helps to improve the accuracy of face detection and location. 1:1 mode image acquisition card, as the data acquisition part of the whole system, is used to display the video shot by camera in real time, and to provide image data to the face detection and location part. Hardware face detection and location card with TMS320C6201B as its core processing unit realizes the hardware face detection and location in real time, using the high-speed processing capability of DSP. The integrated system transplants the algorithm of face detection to DSP successfully. The processing time per frame <WP=5>is below 30ms, realizing the real-time face detection and location.Furthermore, the research and development of multi-channel parallel hardware face detection and location are also made in this project. Multi-channel parallel face detection system can realize real-time detection simultaneously with multi-camera inputs. It improves the performance of the system by several times and reduces the size and cost.更多还原