基于dsp的智能交通运动目标检测与识别技术分析-analysis of intelligent transportation moving target detection and identification technology based on dsp.docxVIP

山东科技大学硕士学位论文摘要 山东科技大学硕士学位论文 摘要 摘 要 在智能交通系统中，运动目标的检测和识别技术已经发展成为计算机机器视觉技术 的重要组成部分。本文的主要研究内容为基于 DSP 的运动目标检测和识别技术研究，其 中主要涉及到静止背景下运动目标的检测和提取，运动目标的识别分类及嵌入式系统软 件设计等方面的内容。 在运动目标检测、提取算法方面。在高斯单模态背景模型算法和对称差分算法的基 础上，基于 blob 分析的思想，提出一种改进的单模态背景模型法。首先通过对称差分法 粗略提取出当前帧中运动目标的形状轮廓；采用基于 blob 分析的设计完成背景模型更 新，较好的解决单模态模型法不能准确描述环境中如树枝晃动的问题；使用分级策略和 光线变化分类思想进行目标精确提取，较好的克服图像中噪声和环境光线变化的影响， 降低系统对光线的敏感度。 运动目标识别方面。为提高系统识别效果，采用一种基于模糊积分的多特征融合目 标分级识别方法，实现大型车、小型车和行人的多目标分类。基本思路如下：首先采用 支持向量机对运动目标的四个仿射不变矩进行一级分类，对于得到的结果，结合面积、 形状复杂度特征，运用模糊积分方法实现运动目标的最终分类，且取得较理想的识别效 果。 系统软件设计。在系统软件设计中，采用了适合 DSP 硬件环境的 DSP/BIOS 操作系统 ， 基于 RF5 框架结构实现了软件程序的构建，给出了系统软件流程图。 关键词：智能交通，DM642，运动目标检测与识别，blob 分析，仿射不变特征，模糊 积分 山东科技大学硕士学位论文AB 山东科技大学硕士学位论文 ABSTRACT ABSTRACT Traffic moving object detection and recognition technology had developed into an important component of the computer machine vision. In this paper, moving object detection, recognization and embedded system’s software designation were mainly presented. In respect of moving object detection, an improved method based on single-mode background model and symmetric difference was proposed which was related to blob analysis. First, the symmetric difference was used to extracte a rough moving object.Second blob analysis is utilized to update background model. Finally, classification strategies were used to extracte foreground accurately and avoid the effect of noise and illumination variance. Experimental results showed that the presented approach worked well in the presence of complex invironment and illumination variance. In respect of object recognization, in order to enhance the effect of system identification, a fuzzy integral based on multi-feature fusion method was used to identify the target classification.It achieved multi-objects classification of large-scale cars, small cars and pedestrians. Firstly, it was utilized to classify the four affine invariant features of moving object by SVM(Support Vector Machin).Secondly, the result of the SVM, area and complexi