
【中文摘要】随着人类社会的快速发展,车辆日益增多,道路交通事故频频发生,给人类社会造成了巨大的损失。因此,如何减少交通事故的发生和降低损失成为了全世界关注的热点。智能辅助驾驶技术是减少交通事故的发生和降低交通事故损失的有效方法之一。行人检测是其中的关键技术,它是利用传感器技术、图像处理技术、计算机技术等多种技术融合检测目标区域中是否存在行人。论文对智能辅助驾驶系统中基于机器视觉技术进行研究。首先分析了运动模糊图像的退化模型,研究了物体运动模糊图像成像原理,得到用于行人检测的运动模糊图像成像方法,确定了造成运动图像模糊的主要因素,然后对运动模糊图像进行处理,重点研究维纳滤波方法,得到较好的图像恢复效果。其次,研究了行人的轮廓特征,采用多种方法对行人图像进行边缘检测,得出基于Canny算子的边缘检测方法具有较好图像边缘检测效果,用此边缘检测方法结合灰度形态学进一步进行图像处理,试验结果显示该方法能很好地消除行人轮廓的干扰因素。最后,详细分析了支持向量机算法,并构造了支持向量机分类器,分别选择纹理特征和不变矩特征作为行人的特征点,从样本图像中提取试验数据,选取部分样本对支持向量机分类器进行训练,然后用训练后的分类器对测试样本进行识别试验。试验结果表明,这两种特征点都能够对行人进行有效地识别。论文的图像试验样本采用摄像机的形式模拟车载视觉传感器获取,对采集到的图像进行预处理,建立行人和非行人图像数据库。
【英文摘要】With the rapid development of human society and the increasing vehicles, the road traffic accidents which occurred frequently cause great damages. How to reduce the traffic accidents and the corresponding loss has become a hot topic. The intelligent assisted driving is one effective way to solve the problem. Pedestrian detection is the key technique for intelligent assisted driving. It detects the target area for the presence of pedestrian by the method. It is combining the technology of sensors, image processing, computer, and so on.This paper is study on machine vision of pedestrians detection technology. Firstly, the degradation model of motion-blurred image was analysed. The method of imaging the motion-blurred image for pedestrian detection and the main factors about image blurring were derived. The wiener filter has better image processing effect by the experiments. Secondly, the pedestrian profile was researched. The method of edge detection based on Canny has a good image edge detection effect by the experiments. Then combining with grayscale morphological for image processing. The results show that the method can be used to eliminate the interference factors of pedestrian profile. Finally, support vector machine was analysed, and support vector machine classifier was structured. Extracting test data from sample image by the texture feature and invariant moments feature. The test sample was recognized by the SVM classifier that was trained by the part of sample image. The experimental results show that these two kinds of features of pedestrians could achieve effectively recognition of the pedestrians.In the experiments, the images were obtained by using hand-held video camera to replace the vehicle-mounted CCD. The image database of pedestrians and non-pedestrians was established by the collected images that were processed simply.
【关键词】行人检测 支持向量机 机器视觉 纹理特征 不变矩
【英文关键词】pedestrian detection machine vision support vector machine texture feature invariant moments
【目录】道路行人特征视觉检测方法研究摘要4-5Abstract51 绪论8-191.1 课题背景及意义8-91.2 行人检测研究现状9-141.2.1 基于传感器的方法9-111.2.2 基于特征提取的方法11-121.2.3 基于人体模型的方法12-131.2.4 基于统计分类的方法13-141.2.5 基于模板匹配的方法141.3 存在的问题14-171.4 本文的创新点171.5 本文的章节安排17-192 图像基础知识及预处理19-332.1 数字图像基本知识19-212.2 图像恢复21-322.2.1 图像退化模型21-252.2.2 模糊图像成像原理25-282.2.3 图像恢复方法28-302.2.4 图像恢复试验30-322.3 本章小结32-333 形态学与图像分割33-463.1 灰度形态学33-343.1.1 膨胀运算和腐蚀运算33-343.1.2 开运算和闭运算343.2 图像分割34-413.2.1 阈值法35-373.2.2 边缘检测法37-413.3 行人图像边缘检测试验41-453.4 本章小结45-4 支持向量机算法46-524.1 最优分类超平面46-484.2 支持向量机算法48-504.3 构造支持向量机分类方法50-514.4 本章小结51-525 行人识别52-655.1 纹理特征52-575.1.1 纹理特征提取52-535.1.2 试验分析与结论53-575.2 不变矩57-5.2.1 不变矩提取58-605.2.2 试验分析与结论60-5.3 本章小结-656 总结和展望65-676.1 主要研究成果656.2 展望65-67参考文献67-74个人简历、在学期间发表的学术论文与研究成果74-75致谢75
