【摘要】：統(tǒng)計數(shù)據(jù)表明,25%~30%的交通事故與駕駛員的警覺狀態(tài)直接相關(guān),其中車輛轉(zhuǎn)彎、并線、變更車道等轉(zhuǎn)向操控過程是交通事故的主要發(fā)生場合之一,尤其常見于駕駛員未注意車輛轉(zhuǎn)向側(cè)后方交通信息的情況。駕駛員后視鏡查看行為的實時檢測和必要提醒有利于降低此類交通事故的發(fā)生概率。本文基于機器視覺和圖像處理技術(shù)對該行為進行探索和研究,提出了與之相關(guān)的檢測方法和技術(shù)方案,并采用大量的實驗和算例進行驗證,最終開發(fā)出駕駛員后視鏡查看行為檢測系統(tǒng)。主要工作和研究成果包括:(1)在對大量國內(nèi)外相關(guān)技術(shù)和文獻調(diào)研的基礎(chǔ)上,針對課題中可能遇到的問題,提出了一種僅以駕駛員臉頸部輪廓作為處理目標(biāo)的駕駛員后視鏡查看行為檢測方法。算法任務(wù)量小、實時性好、魯棒性高。(2)基于駕駛員行為特性,首先提出一種駕駛員臉頸區(qū)域靜態(tài)識別定位算法,用于在車輛啟動時,完成駕駛員臉頸皮膚在當(dāng)前光源條件下的灰度均值學(xué)習(xí)和臉頸區(qū)域靜態(tài)搜索識別。然后提出一種駕駛員臉頸區(qū)域動態(tài)識別定位算法,用于在車輛行駛時,完成駕駛員臉頸皮膚灰度均值學(xué)習(xí)和臉頸區(qū)域快速跟蹤識別。最后在此基礎(chǔ)上提取駕駛員臉頸可見皮膚輪廓并定義了以頸部基點垂線劃分的左右面積比特征參數(shù)。圖像處理結(jié)果表明,算法具有較好的自適應(yīng)學(xué)習(xí)能力和抗干擾能力。(3)針對駕駛員臉型差異、攝像頭安裝位置不同,以及駕駛員發(fā)型、佩戴物等等干擾情況而導(dǎo)致基準特征參數(shù)的不同,論文結(jié)合駕駛眼動凝視數(shù)據(jù)分析揭示了特征參數(shù)的累積概率局部峰值定律,據(jù)此提出一種后視鏡查看行為的閾值判定原理。當(dāng)后視鏡觀察行為獲得確認后,及時利用本次查看過程中的所有參數(shù)值更新累積概率并重新進行參數(shù)估計。實驗數(shù)據(jù)表明,有效的后視鏡查看數(shù)據(jù)更新累積概率可完成參數(shù)的自適應(yīng)調(diào)整。(4)檢測系統(tǒng)體積小,易于推廣。系統(tǒng)在基于樹莓派3代微處理器和嵌入式Linux系統(tǒng)平臺上,用戶界面設(shè)計使用Qt,圖像接口函數(shù)使用圖像處理開源庫OpenCV。檢測結(jié)果表明,系統(tǒng)具有良好的實時性和普適能力。
[Abstract]:The statistical data show that 25% of the traffic accidents are directly related to the alert state of the driver, in which the turning of the vehicle, parallel lane, lane change and other steering and control processes are one of the main situations of traffic accidents. It is especially common that the driver does not pay attention to the traffic information behind the steering side of the vehicle. The real-time detection and necessary reminder of driver's rearview mirror can reduce the probability of this kind of traffic accident. In this paper, the behavior is explored and studied based on machine vision and image processing technology, and the related detection methods and technical schemes are proposed, which are verified by a large number of experiments and examples. Finally, a driver's rearview mirror viewing behavior detection system is developed. The main work and research results include: (1) on the basis of a large number of relevant technologies and literature at home and abroad, aiming at the problems that may be encountered in the subject, In this paper, a method for detecting the viewing behavior of drivers with rearview mirror is proposed, which only takes the outline of driver's face and neck as the target. The algorithm has the advantages of small task quantity, good real-time performance and high robustness. (2) based on the behavior characteristics of the driver, a static recognition and location algorithm for the driver's face and neck region is proposed, which is used to identify and locate the driver's face and neck area when the vehicle starts. The gray mean learning and static search and recognition of the driver's face neck skin under the current light source condition are completed. Then, a dynamic recognition and location algorithm for driver's face and neck region is proposed, which can be used to complete the learning of the gray mean value of the driver's face neck skin and the fast tracking and recognition of the face neck area when the vehicle is driving. Finally, the visible skin outline of the driver's face and neck is extracted and the characteristic parameters of the left and right area ratio are defined according to the vertical line of the neck base point. The image processing results show that the algorithm has good adaptive learning ability and anti-interference ability. (3) according to the difference of driver's face type, the installation position of camera, and the driver's hairstyle, The interference of wearing objects and so on leads to the difference of reference characteristic parameters. Combined with the analysis of driving eye movement gaze data, this paper reveals the local peak law of cumulative probability of characteristic parameters, and puts forward a threshold determination principle of rearview mirror viewing behavior. When the rearview mirror observation behavior is confirmed, all the parameter values in the process of viewing are used to update the cumulative probability and reestimate the parameters. The experimental data show that the cumulative probability of effective rearview mirror viewing data update can be adjusted adaptively. (4) the detection system is small in size and easy to popularize. On the platform of raspberry send 3 generation microprocessor and embedded Linux system, the user interface design uses Qt, image interface function to use image processing open source library OpenCV.. The test results show that the system has good real-time and universal ability.
【學(xué)位授予單位】：廈門理工學(xué)院
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2016
【分類號】：U463.6;U495;TP391.41

【參考文獻】

相關(guān)期刊論文 前10條

1 張波;王文軍;成波;;基于人臉3D模型的駕駛?cè)祟^部姿態(tài)檢測[J];汽車工程;2016年01期

2 胡平;周文洪;;基于幀間差分和邊緣差分的遺留物檢測算法[J];計算機系統(tǒng)應(yīng)用;2015年03期

3 付先平;王亞飛;袁國良;管瀟;PELI Eli;羅罡;;基于粒子濾波的駕駛員視線自動校準算法[J];計算機學(xué)報;2015年12期

4 李勇達;張超;孟令君;;基于頭部姿態(tài)特征的列車司機疲勞駕駛檢測系統(tǒng)研究[J];交通信息與安全;2014年05期

5 李文勝;;基于樹莓派的嵌入式Linux開發(fā)教學(xué)探索[J];電子技術(shù)與軟件工程;2014年09期

6 李榮;劉坤;高文鵬;;基于視覺的目標(biāo)姿態(tài)估計算法[J];黑龍江科技大學(xué)學(xué)報;2014年01期

7 張萬枝;王增才;徐俊凱;;基于面部特征三角形的機車駕駛員頭部姿態(tài)參數(shù)估計[J];鐵道學(xué)報;2013年11期

8 鄒奇敏;辛樂;陳陽舟;;基于3D人臉模型的駕駛員頭部姿態(tài)魯棒跟蹤算法[J];計算機測量與控制;2011年12期

9 陳振學(xué);常發(fā)亮;劉春生;徐建光;;基于Adaboost算法和人臉特征三角形的姿態(tài)參數(shù)估計[J];武漢大學(xué)學(xué)報(信息科學(xué)版);2011年10期

10 初秀民;萬劍;嚴新平;毛U，

本文編號：2485642