基于卷积神经网络交通标志识别系统的设计与研究

基于卷积神经网络交通标志识别系统的设计与研究(任务书,开题报告,论文10000字)
摘要
在车辆行驶过程中，经常会因为驾驶员的神经松懈或者客观地因为天气原因导致驾驶员看不清交通标志而发生严重的交通事故。交通标志识别系统能在环境光线过明或过暗以及有障碍物部分遮拦的情况准确识别交通标志信息并将其及时地反馈给驾驶员，从而大大提高了道路行车的安全性。本文将针对卷积神经网络模型构建和交通标志图片识别两个方面展开研究，以提高交通标志的识别准确率。本文的主要工作如下：
（1）设计基于卷积神经网络的交通标志识别模型。本文选用德国交通标志识别基准数据库（German Traffic Sign Recognition Bench，GTSRB）作为研究对象，对其43种交通标志图片在不同状态下的形态进行训练及测试，从而构建一个较为有效的交通标志识别系统。
（2）本文使用Caffe平台进行交通标志的识别工作。首先将GTSRB数据集的43组图片分为训练和测试两组，完成相应的样本预处理工作之后生成两组lmdb模型，然后用卷积神经网络模型进行训练，最后统计输出值来判断交通标志识别的准确性。
关键词：卷积神经网络；交通标志识别；图像分类处理
 
Abstract
In the process of driving the vehicle, often because of the driver's nervousness or objectively because of weather causes the driver can not see the traffic signs that causes serious traffic accidents. The traffic sign recognition system, can accurately identify the traffic sign information and give it to the driver in a timely manner when the ambient light is too bright or too dark and the obstacle is partially covered. Greatly improving the safety of road traffic. In this thesis we will study the classification of traffic signs and the construction of convolution neural network model to improve the accuracy of traffic sign recognition. The main work of this thesis is as follows:
(1) Based on the design principle of traffic signs, the selected objects are classified according to their color and shape. Commonly used convolution neural network model has the effect of image enhancement. Contrast adaptive histogram equalization of the data representation of the image characteristics, which makes the model is not sensitive to color, color features will not be practical. In this paper, German Traffic Sign Recognition Bench (GTSRB) is used as the research object to train and test the 43 kinds of traffic signs in different states, so as to construct a more effective traffic identification system.
(2) This thesis uses Caffe to identify traffic signs. First, the GTSRB data set of 43 groups of pictures is divided into two groups of training and testing to complete the corresponding write after the work with the corresponding script to generate two sets of lmdb model, and then use MNIST network model training, and finally observe the console output value To determine the accuracy of traffic sign recognition.
Key words: convolution neural network; traffic sign recognition; image classification processing

目录    II
摘要    IV
Abstract    V
第1章 绪论    1
1.1 课题研究的背景与意义    1
1.1.1课题研究的背景    1
1.1.2课题研究的意义    1
1.2 国内外研究现状    2
1.2.1交通标志图片特征值提取与匹配    2
1.2.2 深度学习    3
1.3论文主要内容简介    3
第2章 基于深度学习的交通标志识别模型    5
2.1 道路交通标志特征分析    5
2.2 卷积神经网络的基本网络结构    6
2.3 卷积神经网络的特点    8
2.4 基于卷积神经网络的交通标志识别模型    8
2.5 本章小结    9
第3章交通标志识别程序实现    10
3.1 交通标志识别程序参数设计    10
3.1.1参数设置    10
3.1.2定义LeNet模型    10
3.2 交通标志数据集预处理    12
3.3 交通标志识别训练过程    14
3.4 本章小结    14
第4章 交通标志识别实验结果与分析    15
4.1 实验数据    15
4.2 实验步骤    16
4.2.1实验环境配置    16
4.2.2进行训练和识别    17
4.3识别过程    18
4.3 实验结果    19
4.4 实验结果分析    19
4.5本章小结    19
第5章 总结与展望    20
5.1 研究总结    20
5.2 研究展望    20
参考文献    22
致谢    24