云计算-面向绿色通道检查的高效车型识别方法及其GPUCPU协同加速计算.pdf资源-CSDN文库

版权申诉

190 浏览量 2022-07-03 22:23:02 上传评论收藏 2.2MB PDF 举报

资源推荐

资源详情

资源评论

Abstract

- II -

Abstract

This dissertation is belonged to “JT-G Integration Green Channel Special Inspection

System”， it is a technical development Established-Project of Science and Technology

Department of Guangxi Zhuang Autonomous Region, developing an integrated and

automation equipments for green channel check based on radiation perspective imaging

technology as the goal. In order to avoid the cab of vehicle exposed to radiation, must

realize cab security avoidance, the conventional method based on artificial control

radioactive source shutter is too subjective, poor security, radiation image generation is

not complete. This paper presents a method based on vehicle model recognition，this

method realizes automatic recognition of different vhicle model to avoid the cab and

ensures radiation safety, the main study of this paper is vehicle model recognition

methods based on machine vision, designd and implemented ROI positioning, feature

extraction and vehicle model recognition from the application point of view, the main

tasks are as follows:

(1) Given efficient headstock, Logo, radiator grid positioning methods, lays a

foundation for vehicle model recognition. ROI positioning is difficult due to the

complexity of the outdoor environment. For headstock’s positioning, this paper

combines Gaussian background modeling and morphological gradient edge detection,

presents a sliding window scanning algorithms to extract the headstock accurately; For

Logo positioning, designed a coarse-to-fine Logo positioning algorithm according to the

characteristics of field environment and green channel vehicle models, the key is

utilization texture suppression algorithm, blur detection and normalization algorithm to

suppress noise around Logo such as as radiator grill, windows, wipers, etc. Solve the

problem of previous Logo positioning method with a low accuracy in an outdoor

environment. For radiator grid, this paper can positioning it accurately based location of

plate and Logo.

(2) Proposed and implemented a vehicle model recognition method based on Logo

recognition, the key lies in that it takes the advantage of the recognition of vehicle Logo

in advance to reduce the range of the vehicle model. For vehicle Logo recognition,

extracting Logo DCT low-frequency as the feature vectors, studied the selection of the

feature vector’s dimension and bulit SVM model to identify Logo category. As for

万方数据

- IV -

摘要 .................................................................................................................................................................. I

Abstract ........................................................................................................................................................... II

目录 ............................................................................................................................................................... IV

第一章绪论 ..................................................................................................................................................... 1

§1.1 论文选题背景及意义 ........................................................................................................................... 1

§1.1.1 绿色通道检查技术 ........................................................................................................................ 1

§1.1.2 驾驶员自动检测与安全避让 ........................................................................................................ 2

§1.2 车型识别技术 ....................................................................................................................................... 5

§1.2.1 国内外车型识别技术总体概述 .................................................................................................... 5

§1.2.2 基于机器视觉的车型识别研究 .................................................................................................... 6

§1.3 GPU/CPU 协同加速计算 ..................................................................................................................... 9

§1.3.1 GPU/CPU 协同加速计算 .............................................................................................................. 9

§1.3.2 NVIDIA CUDA 计算平台............................................................................................................. 9

§1.4 本文主要内容 .................................................................................................................................... 10

第二章车型识别框架 .................................................................................................................................... 11

§2.1 车型识别应用环境 .............................................................................................................................. 11

§2.1.1 车型范围 ....................................................................................................................................... 11

§2.1.2 现场环境 ....................................................................................................................................... 11

§2.2 车型识别框架 ..................................................................................................................................... 12

§2.2.1 系统组成 ...................................................................................................................................... 12

§2.2.2 图像采集 ...................................................................................................................................... 13

§2.3 本章小结 ............................................................................................................................................. 15

第三章 ROI 定位和特征提取 ...................................................................................................................... 16

§3.1 ROI 定位 ............................................................................................................................................. 16

§3.1.1 车头作为 ROI 的定位 ................................................................................................................. 16

§3.1.2 Logo 作为 ROI 的定位................................................................................................................ 19

§3.1.3 散热器作为 ROI 的定位 ............................................................................................................. 23

§3.2 特征提取 ............................................................................................................................................. 24

§3.2.1 直方图特征 .................................................................................................................................. 24

§3.2.2 频域特征 ...................................................................................................................................... 28

§3.2.3 纹理特征 ...................................................................................................................................... 31

§3.2.4 角点特征 ...................................................................................................................................... 33

§3.3 实验结果及分析 ................................................................................................................................. 36

§3.4 本章小结 ............................................................................................................................................. 37

万方数据

- V -

第四章基于 Logo 识别的车型识别方法 .................................................................................................... 38

§4.1 Logo 识别 ........................................................................................................................................... 38

§4.1.1 特征提取 ...................................................................................................................................... 38

§4.1.2 维数选择 ...................................................................................................................................... 39

§4.2 车型识别 ............................................................................................................................................. 40

§4.2.1 ROI 定位 ...................................................................................................................................... 40

§4.2.2 特征提取 ...................................................................................................................................... 41

§4.3 实验结果及分析 ................................................................................................................................. 42

§4.3.1 数据集描述及实验环境 .............................................................................................................. 42

§4.3.2 Logo 识别对车型识别的影响 .................................................................................................... 43

§4.3.3 车型识别准确率 .......................................................................................................................... 44

§4.3.4 算法的鲁棒性 .............................................................................................................................. 45

§4.4 本章小结 ............................................................................................................................................. 46

第五章基于自适应 Harris 算法的车型识别方法及其 GPU/CPU 协同加速计算 .................................. 47

§5.1 自适应 Harris 算法 ............................................................................................................................ 47

§5.1.1 经典 Harris 的缺陷 ...................................................................................................................... 47

§5.1.2 自适应 Harris ............................................................................................................................... 49

§5.2 Max-Correlation 匹配角点 ................................................................................................................. 51

§5.2.1 匹配流程 ...................................................................................................................................... 51

§5.2.2 参数选择 ...................................................................................................................................... 53

§5.3 GPU/CPU 协同加速计算 ................................................................................................................... 54

§5.3.1 减小 IO 密集度的优化设计 ........................................................................................................ 54

§5.3.2 自适应 Harris 算法的并行优化设计 .......................................................................................... 56

§5.3.3 Max-Correlation 算法的并行优化设计 ...................................................................................... 58

§5.3.4 识别任务的并行化 ...................................................................................................................... 59

§5.4 总体流程 ............................................................................................................................................. 60

§5.5 实验结果及分析 ................................................................................................................................. 61

§5.5.1 数据集描述及实验环境 .............................................................................................................. 61

§5.5.2 车型识别准确率 .......................................................................................................................... 62

§5.5.3 识别时间 ...................................................................................................................................... 62

§5.5.4 算法的鲁棒性 .............................................................................................................................. 63

§5.6 本章小结 ............................................................................................................................................. 63

第六章总结与展望 ....................................................................................................................................... 64

§6.1 总结 ..................................................................................................................................................... 64

§6.2 后续研究计划 ..................................................................................................................................... 65

§6.2.1 扩大识别范围 .............................................................................................................................. 65

§6.2.2 提高算法的鲁棒性 ...................................................................................................................... 65

参考文献 ......................................................................................................................................................... 66

致谢 ............................................................................................................................................................... 70

万方数据