基于确定学习的步态识别的并行计算实现

发布时间：2018-06-07 02:41

本文选题：步态识别 + 确定学习　；参考：《华南理工大学》2012年硕士论文

【摘要】：随着信息技术的不断发展，计算机可以更加快速、更加有效的为我们服务，尤其是在处理一些高阶次、计算量很大的科学问题上，计算机的性能直接影响问题解决的方式。当今计算机的处理水平已经不局限于用于物理上的内存扩展和处理器速度的提高，，类型多样的并行处理体系得到应用。而并行计算的应用可以在处理器能力没有质的飞跃的前提下，大幅度提高计算机的性能。其原理是将一个计算任务分配到不同的进程中去，同时处理，从而加快计算的速度，提高解决问题的效率，与很多科学问题有很好的结合能力。步态识别技术可以根据人体的步态特征，在远距离、具有隐蔽性的前提下，识别出个体的身份，这在当今存在越来越多需要对人的身份进行识别的场合的背景下，具有重要意义。现在的步态识别技术是基于行为特征的，对包含个体特征的图像序列进行预处理、检测和特征提取，然后才可以进行识别。由于人体的行为特征非常复杂，进行建模时包含很多未知部分。这时，确定学习理论的应用很好的解决了这个问题。确定学习理论是基于RBF神经网络的在未知动态环境下，可以对系统局部动态进行准确逼近的新理论。在辨识出来步态的动态特征之后存储于常值神经网络，然后利用已存储的模式对步态特征进行快速准确识别。本文主要介绍了确定学习理论在步态特征识别中的实际应用。为了加速识别过程，提高识别效率，在步态识别的具体应用中，应用了并行计算的实现方式。在并行的硬件平台上，构造并行的识别算法，并且在多核的平台上实现了步态识别的并行程序设计，分析了程序的性能。最后为了处理过程的直观性和方便性，分别用基于C++的MFC框架和matlab GUI建立了步态特征提取和识别的系统界面，展示了识别的快速性和有效性。
[Abstract]:With the continuous development of information technology, computers can serve us more quickly and effectively, especially in dealing with some scientific problems of high order and large amount of computation. The performance of computers directly affects the way to solve the problems. Nowadays, the processing level of computers is no longer limited to physical memory expansion and processor speed improvement. Parallel processing systems of various types have been applied. Parallel computing applications can greatly improve the performance of computers without a qualitative leap in processor power. The principle is to assign a computing task to different processes and deal with it at the same time, so as to speed up the calculation, improve the efficiency of solving problems, and have a good ability to combine with many scientific problems. Gait recognition technology can recognize the identity of individuals on the premise of distance and concealment according to the gait characteristics of human body. It is of great significance. The current gait recognition technology is based on behavioral features. The image sequences containing individual features are preprocessed, detected and feature extracted before recognition can be carried out. Due to the complexity of human behavior, modeling contains many unknown parts. At this time, the application of deterministic learning theory solves this problem very well. Deterministic learning theory is a new theory based on RBF neural network which can accurately approximate the local dynamics of the system in unknown dynamic environment. After the gait dynamic features are identified, they are stored in the constant neural network, and the stored patterns are used to identify the gait features quickly and accurately. This paper mainly introduces the practical application of deterministic learning theory in gait feature recognition. In order to accelerate the recognition process and improve the recognition efficiency, parallel computing is applied in gait recognition. The parallel recognition algorithm is constructed on the parallel hardware platform, and the parallel program design for gait recognition is implemented on the multi-core platform, and the performance of the program is analyzed. Finally, in order to deal with the intuitiveness and convenience of the process, the system interface of gait feature extraction and recognition is established by using MFC framework based on C and matlab GUI, respectively, which shows the rapidity and validity of recognition.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41;TP338.6

【参考文献】