本文選題：交通標(biāo)識(shí)識(shí)別　切入點(diǎn)：特征融合　出處：《廣東工業(yè)大學(xué)》2017年碩士論文　論文類型：學(xué)位論文

【摘要】：汽車在人們生活中扮演著越來越重要的角色,安全暢通的駕駛環(huán)境是交通系統(tǒng)的理想狀態(tài)。交通標(biāo)識(shí)識(shí)別是智能交通系統(tǒng)的重要組成部分,它主要包括交通標(biāo)識(shí)的目標(biāo)定位和目標(biāo)識(shí)別兩部分。以交通標(biāo)識(shí)為研究對(duì)象,提出了基于特征融合和深度卷積神經(jīng)網(wǎng)絡(luò)的交通標(biāo)識(shí)識(shí)別方法。首先介紹了國內(nèi)外交通標(biāo)識(shí)識(shí)別的研究現(xiàn)狀,對(duì)比了過去研究中目標(biāo)定位與目標(biāo)識(shí)別方法的優(yōu)劣,提出了基于特征融合的目標(biāo)定位方法和基于深度卷積神經(jīng)網(wǎng)絡(luò)的目標(biāo)識(shí)別方法。在目標(biāo)定位問題上,通過提取HOG特征和LBP特征,串行融合后使用支持向量機(jī)作為分類器。實(shí)驗(yàn)證明該方法可以對(duì)含交通標(biāo)識(shí)的圖片進(jìn)行有效定位,并能夠排除不含交通標(biāo)識(shí)的圖片干擾。深度卷積神經(jīng)網(wǎng)絡(luò)是近年來提出的區(qū)別于淺層神經(jīng)網(wǎng)絡(luò)的機(jī)器學(xué)習(xí)方法模型,因其優(yōu)秀的學(xué)習(xí)能力和應(yīng)用效果受到廣泛重視。介紹了自動(dòng)編碼機(jī)、稀疏編碼、受限玻爾茲曼機(jī)、深度信念網(wǎng)絡(luò)和卷積神經(jīng)網(wǎng)絡(luò)等原理和訓(xùn)練方法,重點(diǎn)介紹了ALex Net和Google Net等深度卷積神經(jīng)網(wǎng)絡(luò)模型。根據(jù)研究對(duì)象和應(yīng)用場(chǎng)景,提出了針對(duì)交通標(biāo)識(shí)識(shí)別的深度卷積神經(jīng)網(wǎng)絡(luò)模型TSR9L-Net,并建立了相應(yīng)的訓(xùn)練圖像數(shù)據(jù)庫。通過平衡識(shí)別率和識(shí)別速度,提出一個(gè)含9層的輕量級(jí)參數(shù)數(shù)量模型,其中權(quán)重層為6層。分別對(duì)含7類警告標(biāo)識(shí)和15類禁令標(biāo)識(shí)的樣本訓(xùn)練集進(jìn)行訓(xùn)練,同時(shí)對(duì)比Le Net-5、Alex Net和TSR9L-Net三種模型的訓(xùn)練效果。其中TSR9L-Net能夠在保證準(zhǔn)確率的前提下,提升識(shí)別速度。GPU硬件平臺(tái)下,7類標(biāo)識(shí)每批40張識(shí)別速度達(dá)29.3ms,準(zhǔn)確率99.09%;15類標(biāo)識(shí)每批40張識(shí)別速度32.0ms,準(zhǔn)確率99.29%。無論是識(shí)別率還是識(shí)別速度,都優(yōu)于Alex Net。
[Abstract]:Automobile plays a more and more important role in people's life. Safe and smooth driving environment is the ideal state of traffic system. Traffic identification is an important part of intelligent transportation system. It mainly includes two parts: target location and target recognition of traffic signs. A traffic identification method based on feature fusion and deep convolution neural network is proposed. Firstly, the research status of traffic sign recognition at home and abroad is introduced, and the advantages and disadvantages of target location and target recognition methods in previous research are compared. In this paper, a target location method based on feature fusion and a target recognition method based on deep convolution neural network are proposed. In the problem of target location, HOG features and LBP features are extracted. After serial fusion, support vector machine is used as classifier. Experimental results show that this method can effectively locate images with traffic signs. The deep convolution neural network is a machine learning method model proposed in recent years, which is different from the shallow neural network. The principles and training methods of automatic coding machine, sparse coding, constrained Boltzmann machine, depth belief network and convolution neural network are introduced. The deep convolution neural network models such as ALex Net and Google Net are introduced in detail. In this paper, a deep convolution neural network model TSR9L-Netfor traffic identification is proposed, and the corresponding training image database is established. By balancing recognition rate and recognition speed, a 9-layer lightweight parameter quantity model is proposed. The weight layer is six layers. The training sets with 7 warning marks and 15 ban marks are trained, and the training effects of Le Net-5N Alex Net and TSR9L-Net are compared. TSR9L-Net can ensure the accuracy of the training. On the hardware platform of GPU, the recognition speed of each batch of 40 marks is up to 29.3 ms.The accuracy rate of 99.0915 class marks is 32.0ms. the accuracy is 99.290.The recognition rate and speed are better than that of Alex NetNet.
【學(xué)位授予單位】：廣東工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41;TP18

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 高燦;;一種基于CNN手寫字符識(shí)別的改進(jìn)方法[J];黑龍江科技信息;2017年03期

2 張俊俊;何良華;;動(dòng)態(tài)深度信念網(wǎng)絡(luò)模型構(gòu)建[J];微型機(jī)與應(yīng)用;2017年01期

3 康麗萍;許光鑾;孫顯;;受限玻爾茲曼機(jī)的稀疏化特征學(xué)習(xí)[J];計(jì)算機(jī)科學(xué);2016年12期

4 劉成云;常發(fā)亮;陳振學(xué);;基于感興趣區(qū)域和HOG-MBLBP特征的交通標(biāo)識(shí)檢測(cè)[J];電子與信息學(xué)報(bào);2016年05期

5 劉凱;張立民;孫永威;林雪原;;利用深度玻爾茲曼機(jī)與典型相關(guān)分析的自動(dòng)圖像標(biāo)注算法[J];西安交通大學(xué)學(xué)報(bào);2015年06期

6 楊杰;孫亞東;張良俊;劉海波;;基于弱監(jiān)督學(xué)習(xí)的去噪受限玻爾茲曼機(jī)特征提取算法[J];電子學(xué)報(bào);2014年12期

7 周佳俊;歐智堅(jiān);;深層神經(jīng)網(wǎng)絡(luò)預(yù)訓(xùn)練的改進(jìn)初始化方法[J];電訊技術(shù);2013年07期

8 孫光民;王晶;于光宇;李罡;許磊;;自然背景中交通標(biāo)志的檢測(cè)與識(shí)別[J];北京工業(yè)大學(xué)學(xué)報(bào);2010年10期

9 王楠;劉威;陳雪;袁淮;劉積仁;;基于多線索混合的交通標(biāo)志檢測(cè)與跟蹤[J];控制與決策;2010年08期

10 朱雙東;劉蘭蘭;;基于顏色信息與SVM網(wǎng)絡(luò)的交通標(biāo)志檢測(cè)[J];自動(dòng)化儀表;2009年03期

，

本文編號(hào)：1570662