基于帧差法和AdaBoost的运动车辆视频检测
发布时间：2019-01-27 19:30
【摘要】：当今社会的发展对于智能交通系统的管理和服务功能需求越来越高,国内外对此方面的研究均相当重视。而车辆检测技术是智能交通系统的根基。在对传统的目标检测技术和相关模型系统进行分析总结的基础上,本文提出了一种将帧差法和AdaBoost相结合的车辆检测技术。主要包含了以下几方面的工作： 一.与传统帧差法相异,本文帧差法的主要目的是将原帧图像中的车辆灰度特征进行简化显示,没有采取二值化操作,而是将带有丰富灰度信息的帧差图像进入下面步骤的处理。 二.通过图像的滤波处理进一步去除图像中的颗粒型噪声,将运动信息从其中进一步突出出来。 三.采用AdaBoost的方法,以帧差图像中提取的车辆和非车辆样本作为训练样本训练弱分类器,并通过一定步骤的迭代操作后得出最终的强分类器。由于有前面的处理方法突出了运动目标,因此需要进行的学习大大简化。 最后经过实验,本文中的识别系统对于道路车辆的识别取得了较好的识别效果和可以接受的识别速度。
[Abstract]:With the development of the society, more and more attention is paid to the management and service function of intelligent transportation system. Vehicle detection technology is the foundation of intelligent transportation system. Based on the analysis and summary of traditional target detection technology and related model system, this paper proposes a vehicle detection technology which combines frame difference method and AdaBoost. Mainly includes the following aspects of work: 1. Different from the traditional frame difference method, the main purpose of the frame difference method is to simplify the display of the vehicle gray level features in the original frame image, instead of taking a binary operation, but to process the frame difference image with rich gray level information into the following steps. II. The granular noise in the image is further removed by filtering the image, and the motion information is further highlighted from it. Three The method of AdaBoost is used to train the weak classifier with the vehicle and non-vehicle samples extracted from the frame difference image as the training samples, and the final strong classifier is obtained after a certain iterative operation. The learning needs to be greatly simplified because the previous processing method highlights the moving target. Finally, through the experiment, the recognition system in this paper has achieved better recognition effect and acceptable recognition speed for road vehicle recognition.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495
本文编号：2416617
[Abstract]:With the development of the society, more and more attention is paid to the management and service function of intelligent transportation system. Vehicle detection technology is the foundation of intelligent transportation system. Based on the analysis and summary of traditional target detection technology and related model system, this paper proposes a vehicle detection technology which combines frame difference method and AdaBoost. Mainly includes the following aspects of work: 1. Different from the traditional frame difference method, the main purpose of the frame difference method is to simplify the display of the vehicle gray level features in the original frame image, instead of taking a binary operation, but to process the frame difference image with rich gray level information into the following steps. II. The granular noise in the image is further removed by filtering the image, and the motion information is further highlighted from it. Three The method of AdaBoost is used to train the weak classifier with the vehicle and non-vehicle samples extracted from the frame difference image as the training samples, and the final strong classifier is obtained after a certain iterative operation. The learning needs to be greatly simplified because the previous processing method highlights the moving target. Finally, through the experiment, the recognition system in this paper has achieved better recognition effect and acceptable recognition speed for road vehicle recognition.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495
【参考文献】
相关期刊论文 前2条
1 张玲,陈丽敏,何伟,郭磊民;基于视频的改进帧差法在车流量检测中的应用[J];重庆大学学报(自然科学版);2004年05期
2 郁梅,蒋刚毅,郁伯康;智能交通系统中的计算机视觉技术应用[J];计算机工程与应用;2001年10期
，本文编号：2416617
本文链接：https://www.wllwen.com/kejilunwen/jiaotonggongchenglunwen/2416617.html
