基于HOG与IOELM的人体检测方法研究及系统实现

发布时间：2018-12-20 15:25

【摘要】：图像目标检测是计算机视觉技术中的一个研究热点,其中人体检测是图像目标检测的重要内容,旨在利用计算机模拟人脑的思维方式从图像或视频中找出人体所在区域。目前人体检测技术广泛应用于智能视频监控、车辆辅助驾驶系统和虚拟现实等领域。本文主要从基于统计学习的角度出发,对人体检测方法展开了系统研究,具体的工作介绍如下:1、提出一种基于梯度方向直方图(Histogram of Oriented Gradient,HOG)与优化极限学习机(Optimization Extreme Learning Machine,OELM)相结合的人体检测方法。首先,采用梯度方向直方图方法提取图像的特征值;然后,利用OELM算法对提取图像的特征值进行分类训练;最后,利用非极大值抑制方法准确标记出目标人体区域。实验表明,此方法相对于传统的HOG与ELM方法在训练精度上有显著的提高,在算法运行时间上相对于经典HOG与SVM方法更是快了近20倍。2、针对分类问题,提出一种基于EAS(Efficient Active Set)算法优化OELM的方法,即 IOELM(Improved Optimization Extreme Learning Machine)。首先,利用有效集算法在迭代求解优化问题最优解的过程中,找出符合条件的最大搜索步长来保证函数值严格下降;然后,设置临时迭代步长找到最优步长使目标优化问题的函数值较有效集法进一步下降,并通过推测赋值法来减少迭代过程中产生的误迭代;最后,提出基于HOG与IOELM相结合的人体检测方法。通过实验证明,此方法不仅减少了人体检测中训练过程的计算代价,同时降低了样本的训练时间。3、基于上述提出的方法设计并实现人体检测系统。该系统主要包括三个模块:图像界面功能模块,人体检测模块和性能评价模块。图像界面功能包括输入和保存图像,人体检测包含几种经典的人体检测方法,性能评价模块是记录几种方法的评价指标。
[Abstract]:Image target detection is a hot topic in computer vision technology, in which human body detection is an important part of image target detection. It aims to find out the region of human body from image or video by computer simulation of human brain. At present, human detection technology is widely used in intelligent video surveillance, vehicle-assisted driving system and virtual reality. In this paper, the human body detection method is studied systematically from the point of view of statistical learning. The specific work is as follows: 1. A gradient direction histogram (Histogram of Oriented Gradient,) based method is proposed. HOG) and optimized extreme learning machine (Optimization Extreme Learning Machine,OELM). Firstly, the gradient histogram method is used to extract the image eigenvalues; then, the OELM algorithm is used to classify the extracted image eigenvalues. Finally, the non-maximum suppression method is used to accurately mark the target human body region. The experimental results show that the training accuracy of this method is significantly higher than that of the traditional HOG and ELM methods, and the running time of the algorithm is nearly 20 times faster than that of the classical HOG and SVM methods. A method of optimizing OELM based on EAS (Efficient Active Set) algorithm, that is, IOELM (Improved Optimization Extreme Learning Machine)., is proposed. Firstly, in the process of iterative solving the optimal solution of the optimization problem, the maximum search step size is found by using the effective set algorithm to ensure the strict decline of the function value. Then, the function value of the objective optimization problem is further reduced by setting the temporary step size to find the optimal step size, and the error iteration in the iterative process is reduced by the method of inferred assignment. Finally, a human detection method based on HOG and IOELM is proposed. It is proved by experiments that this method not only reduces the computational cost of training process in human body detection, but also reduces the training time of samples. 3. Based on the method mentioned above, a human body detection system is designed and implemented. The system consists of three modules: image interface function module, human detection module and performance evaluation module. The function of image interface includes input and preservation of image, human body detection includes several classical human detection methods, and performance evaluation module is the evaluation index of recording several methods.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】