面向行人防碰撞預警的駕駛員駕駛意圖辨識方法研究
發(fā)布時間：2018-08-06 19:56
【摘要】：從十九世紀到二十一世紀,汽車已經(jīng)發(fā)展成為交通運輸?shù)闹袌粤α�。與此同時,汽車工業(yè)的發(fā)展也造成了嚴重的環(huán)境和交通安全問題。人們對于交通安全問題的研究也從單純提高車輛自身的安全技術(shù)水平逐步轉(zhuǎn)移到綜合考慮車輛、駕駛員以及環(huán)境等各個因素來提高道路交通安全。在道路交通事故中,車輛和行人的碰撞是最主要的事故形式之一,而行人是事故中最大的受害群體。因此,如何提高汽車的主動安全性能、有效保護行人的安全已經(jīng)越來越被重視。在國家自然科學基金項目(61104165)和中央高�；究蒲袠I(yè)務費專項資金資助項目(DUT13JS02)的資助下,本文開展了面向行人防碰撞預警的駕駛員駕駛意圖辨識方法的研究。在檢測到車輛前方存在行人的基礎上,本文確定了需要辨識的4種駕駛意圖,即加速、減速制動、轉(zhuǎn)彎避讓(正常轉(zhuǎn)向)和急轉(zhuǎn)。通過分析駕駛意圖產(chǎn)生的機理和統(tǒng)計模式識別理論,確定應用隱馬爾科夫模型(Hidden Markov Model, HMM)來辨識駕駛員駕駛意圖。根據(jù)需要辨識的4種駕駛員意圖,通過駕駛模擬器采集實驗所需的傳感器數(shù)據(jù),并對實驗數(shù)據(jù)進行處理。采用羅馬諾夫斯基準則對傳感器數(shù)據(jù)的異常值進行剔除,使用改進的K-means算法對駕駛員正常轉(zhuǎn)向和緊急轉(zhuǎn)向的界限值進行確定。根據(jù)隱馬爾科夫HMM模型的Baum-Welch算法和前向算法,使用MATLAB結(jié)合Baum-Welch算法和前向算法編寫m文件進行駕駛員意圖辨識。使用Baum-Welch算法進行駕駛意圖隱馬爾科夫HMM模型的離線訓練。由于不同模型間觀察序列的長度不同,為了使實驗得到的模型更加精確,對Baum-Welch算法進行了改進。訓練得到表征駕駛意圖隱馬爾科夫HMM模型的參數(shù)。最后,使用處理后的觀察序列進行駕駛員意圖在線辨識。將觀察序列輸入到搭建好的隱馬爾科夫HMM模型中,用MATLAB結(jié)合前向算法編寫m文件得到觀察序列和不同模型間的匹配值,匹配值最大的模型視為駕駛員意圖隱馬爾科夫HMM模型�；谲囕v前方行人檢測結(jié)果,在辨識得到駕駛員意圖后,進行行人防碰撞預警機制的確定。對錯誤的駕駛員操作,如誤踩加速踏板,通過制定的對駕駛員和行人同時預警的機制,可以有效保護行人安全。
[Abstract]:From the nineteenth century to the 21 century, the automobile has developed into the backbone of transportation. At the same time, the development of automobile industry has also caused serious environmental and traffic safety problems. The research on traffic safety has been gradually transferred from simply improving the safety technology level of vehicles to taking into account various factors such as vehicles drivers and environment to improve road traffic safety. In road traffic accidents, the collision between vehicles and pedestrians is one of the most important accident forms, and pedestrians are the largest victims of accidents. Therefore, more and more attention has been paid to how to improve the active safety of vehicles and effectively protect the safety of pedestrians. With the support of the National Natural Science Foundation of China (61104165) and the DUT13JS02 (Central University basic Scientific Research Business Fund) project, this paper studies the identification method of driver's driving intention for pedestrian anti-collision warning. On the basis of detecting the presence of pedestrians in front of the vehicle, this paper determines four driving intentions that need to be identified, that is, acceleration, deceleration and braking, turning and avoiding (normal steering) and sharp turning. By analyzing the mechanism of driving intention and the theory of statistical pattern recognition, the hidden Markov model (Hidden Markov Model, HMM) is used to identify driver's driving intention. According to the four kinds of driver's intention which need to be identified, the sensor data needed in the experiment are collected by driving simulator, and the experimental data are processed. The outliers of sensor data are eliminated by Romonovsky criterion, and the limit values of normal steering and emergency steering of drivers are determined by the improved K-means algorithm. According to the Baum-Welch algorithm and forward algorithm of Hidden Markov HMM model, the author uses MATLAB combined with Baum-Welch algorithm and forward algorithm to write m file for driver intention identification. The Baum-Welch algorithm is used to train the hidden Markov HMM model of driving intention. In order to make the model more accurate, the Baum-Welch algorithm is improved because of the different length of observation sequence among different models. The parameters of the hidden Markov HMM model are obtained. Finally, the treated observation sequence is used to identify the driver's intention online. The observation sequence is input into the constructed hidden Markov HMM model, and the m file is compiled by MATLAB combined with the forward algorithm to get the matching value between the observation sequence and the different models. The model with the largest matching value is regarded as the hidden Markov model of the driver's intention. Based on the results of pedestrian detection in front of the vehicle, the pedestrian collision prevention warning mechanism is determined after the driver's intention is identified. The safety of pedestrians can be effectively protected by the wrong driver operation such as stepping on the accelerator pedal by establishing a warning mechanism for both the driver and the pedestrian.
【學位授予單位】：大連理工大學
【學位級別】：碩士
【學位授予年份】：2015
【分類號】：U463.6;U495
本文編號：2168823
[Abstract]:From the nineteenth century to the 21 century, the automobile has developed into the backbone of transportation. At the same time, the development of automobile industry has also caused serious environmental and traffic safety problems. The research on traffic safety has been gradually transferred from simply improving the safety technology level of vehicles to taking into account various factors such as vehicles drivers and environment to improve road traffic safety. In road traffic accidents, the collision between vehicles and pedestrians is one of the most important accident forms, and pedestrians are the largest victims of accidents. Therefore, more and more attention has been paid to how to improve the active safety of vehicles and effectively protect the safety of pedestrians. With the support of the National Natural Science Foundation of China (61104165) and the DUT13JS02 (Central University basic Scientific Research Business Fund) project, this paper studies the identification method of driver's driving intention for pedestrian anti-collision warning. On the basis of detecting the presence of pedestrians in front of the vehicle, this paper determines four driving intentions that need to be identified, that is, acceleration, deceleration and braking, turning and avoiding (normal steering) and sharp turning. By analyzing the mechanism of driving intention and the theory of statistical pattern recognition, the hidden Markov model (Hidden Markov Model, HMM) is used to identify driver's driving intention. According to the four kinds of driver's intention which need to be identified, the sensor data needed in the experiment are collected by driving simulator, and the experimental data are processed. The outliers of sensor data are eliminated by Romonovsky criterion, and the limit values of normal steering and emergency steering of drivers are determined by the improved K-means algorithm. According to the Baum-Welch algorithm and forward algorithm of Hidden Markov HMM model, the author uses MATLAB combined with Baum-Welch algorithm and forward algorithm to write m file for driver intention identification. The Baum-Welch algorithm is used to train the hidden Markov HMM model of driving intention. In order to make the model more accurate, the Baum-Welch algorithm is improved because of the different length of observation sequence among different models. The parameters of the hidden Markov HMM model are obtained. Finally, the treated observation sequence is used to identify the driver's intention online. The observation sequence is input into the constructed hidden Markov HMM model, and the m file is compiled by MATLAB combined with the forward algorithm to get the matching value between the observation sequence and the different models. The model with the largest matching value is regarded as the hidden Markov model of the driver's intention. Based on the results of pedestrian detection in front of the vehicle, the pedestrian collision prevention warning mechanism is determined after the driver's intention is identified. The safety of pedestrians can be effectively protected by the wrong driver operation such as stepping on the accelerator pedal by establishing a warning mechanism for both the driver and the pedestrian.
【學位授予單位】：大連理工大學
【學位級別】：碩士
【學位授予年份】：2015
【分類號】：U463.6;U495
【參考文獻】
相關(guān)期刊論文 前1條
1 張鳳;李永娟;蔣麗;;駕駛行為理論模型研究概述[J];中國安全科學學報;2010年02期
相關(guān)碩士學位論文 前1條
1 李守衛(wèi);基于輪速信號的側(cè)向加速度估算與彎道制動控制算法研究[D];吉林大學;2008年
，本文編號：2168823
本文鏈接：http://sikaile.net/kejilunwen/daoluqiaoliang/2168823.html
