本文選題：機(jī)器視覺 + 道路狀況感知��； 參考：《華南理工大學(xué)》2014年碩士論文

【摘要】：近年來(lái)世界范圍內(nèi)尤其在中國(guó)汽車保有量持續(xù)增加，但是越來(lái)越多的道路汽車駕駛安全隱患也在增長(zhǎng)，故對(duì)提高安全駕駛的輔助駕駛系統(tǒng)需求越來(lái)越強(qiáng)烈。本文中基于機(jī)器視覺的汽車行駛道路狀況感知系統(tǒng)正式輔助駕駛系統(tǒng)的一部分。通過(guò)本系統(tǒng)可以對(duì)單目攝像機(jī)采集的圖像進(jìn)行分析處理，并得到道路車道線的位置信息與車輛的位置信息，通過(guò)這些信息對(duì)輔助駕駛系統(tǒng)的反饋，很大程度上可以提高道路汽車駕駛的安全。 首先，本文引入了兩項(xiàng)道路智能感知系統(tǒng)的關(guān)鍵技術(shù)：車道線檢測(cè)和車輛檢測(cè)。并對(duì)機(jī)器視覺的概念與兩項(xiàng)關(guān)鍵技術(shù)前的圖像預(yù)處理進(jìn)行闡述。 其次，研究了用于車道線檢測(cè)的隨機(jī)霍夫變換算法，，考慮到該算法的準(zhǔn)確性與實(shí)時(shí)性問(wèn)題，采用了區(qū)域約束與角度約束的方法提高了車道線檢測(cè)的準(zhǔn)確性與實(shí)時(shí)性。 然后，對(duì)車輛檢測(cè)方法進(jìn)行深入研究，根據(jù)車輛檢測(cè)準(zhǔn)確性與實(shí)時(shí)性的需求，并整理出一套較好解決高準(zhǔn)確性與實(shí)時(shí)性需求的方法。將檢測(cè)過(guò)程分為了假設(shè)產(chǎn)生模塊與假設(shè)驗(yàn)證模塊。針對(duì)假設(shè)產(chǎn)生模塊對(duì)實(shí)時(shí)性的優(yōu)化與假設(shè)驗(yàn)證模塊對(duì)準(zhǔn)確性的優(yōu)化，分別詳細(xì)設(shè)計(jì)了兩個(gè)模塊的各自框架與實(shí)現(xiàn)方法，包括基于AdaBoost和Haar-like特征的假設(shè)產(chǎn)生框架與基于SVM和HOG特征的假設(shè)驗(yàn)證框架，以上兩個(gè)框架分別結(jié)合了級(jí)聯(lián)AdaBoost快速篩選與Haar-like特征快速計(jì)算的速度優(yōu)點(diǎn)以及SVM對(duì)高維特征的分類與HOG特征豐富的梯度信息的準(zhǔn)確性優(yōu)點(diǎn)。 經(jīng)系統(tǒng)實(shí)現(xiàn)與測(cè)試驗(yàn)證，本文研究的汽車行駛道路狀況智能感知系統(tǒng)，在結(jié)構(gòu)化道路情境下，達(dá)到了較高的檢測(cè)率與較優(yōu)的實(shí)時(shí)性。
[Abstract]:In recent years, the number of automobile ownership has been increasing in the world, especially in China, but more and more hidden dangers of driving safety are also increasing, so the demand of auxiliary driving system for improving safe driving is becoming more and more intense. This paper is a part of the vehicle driving condition perception system based on machine vision. Through this system, the image collected by the monocular camera can be analyzed and processed, and the position information of the road lane and the vehicle can be obtained, and the feedback to the auxiliary driving system can be obtained through these information. To a large extent, it can improve the safety of driving on the road. Firstly, this paper introduces two key technologies of road intelligent sensing system: lane detection and vehicle detection. The concept of machine vision and the image preprocessing before two key technologies are expounded. Secondly, the random Hough transform algorithm for lane detection is studied. Considering the accuracy and real-time of the algorithm, the method of region constraint and angle constraint is adopted to improve the accuracy and real-time performance of lane detection. Then, the method of vehicle detection is studied deeply, according to the demand of accuracy and real-time of vehicle detection, and a set of methods to solve the requirement of high accuracy and real-time is put forward. The detection process is divided into hypothesis generation module and hypothesis verification module. Aiming at the optimization of the real-time performance of the hypothesis generation module and the optimization of the accuracy of the hypothesis verification module, the respective frameworks and implementation methods of the two modules are designed in detail. The hypothesis generation framework based on AdaBoost and Haar-like features and the hypothesis verification framework based on SVM and HOG features are included. These two frameworks combine the advantages of cascaded AdaBoost fast filtering and Haar-like feature fast computation and SVM classification of high-dimensional features and the accuracy of hog feature rich gradient information. Through the system realization and test, the intelligent sensing system of vehicle driving road condition is studied in this paper. Under the situation of structured road, the intelligent sensing system achieves higher detection rate and better real-time performance.
【學(xué)位授予單位】：華南理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類號(hào)】：TP391.41;U495

【參考文獻(xiàn)】

相關(guān)期刊論文 前3條

1 謝勰;王輝;張雪鋒;;圖像閾值分割技術(shù)中的部分和算法綜述[J];西安郵電學(xué)院學(xué)報(bào);2011年03期

2 王植,賀賽先;一種基于Canny理論的自適應(yīng)邊緣檢測(cè)方法[J];中國(guó)圖象圖形學(xué)報(bào);2004年08期

3 陳軍;徐友春;趙明;彭永勝;;基于隨機(jī)Hough變換的道路邊界識(shí)別算法研究[J];中國(guó)圖象圖形學(xué)報(bào);2009年05期

本文編號(hào)：2018739