【作者】 王献伟；

【导师】 周毅；

【作者基本信息】 河南大学 ， 控制理论与控制工程， 2020， 硕士

【摘要】 作为智能驾驶和智能交通系统发展道路上的重要一环,驾驶行为辨识受到了广泛地关注和研究。通过研究安全行车场景下的驾驶行为可以实时了解车辆的行车状,从而可以有效的提高行车安全;研究事故场景下驾驶行为的合理性,可以减少纠纷进而加快事故处理。在安全行车场景下,准确估计车辆的跟车行为和变道行为结束时刻有利于及早对车辆发出警告,保障行车安全。同时基于事故数据改善行车驾驶安全的研究日益兴起,研究事故场景下驾驶行为的合理性,可以减少纠纷进而提高事故处理和行车效率。针对以上问题,本文设计了基于生存分析方法估计驾驶行为生存时间的模型;使用深度学习算法门控循环单元(Gated Recurrent Unit,GRU)网络设计了预测车辆意图和轨迹的算法模型;基于隐马尔科夫模型(Hidden Markov Model,HMM),提出了追尾事故下驾驶行为评价策略,进行合理的责任认定。本文的主要研究内容包括:(1)设计了基于生存分析模型的驾驶行为生存时间估计方法。安全行车场景下,使用自然驾驶数据集NGSIM数据集中的车辆轨迹信息,对跟车和变道两种基本的驾驶行为进行分析。引入生存分析模型估计跟车和变道行为生存时间的生存率,进而确定行为结束时间,增强行车安全性。(2)设计了基于GRU算法的车辆意图检测和轨迹预测模型。安全行车场景下,根据自然驾驶数据集NGSIM数据集中的车辆轨迹信息,分析车辆动态特征,提取车辆的显著特征,例如横向速度和横向加速度,使用深度学习方法门控循环单元构建融合驾驶意图的车辆轨迹预测模型,进而对驾驶行为做出更为准确的预测和分类。(3)提出了追尾事故场景下驾驶行为评价策略。事故场景下,构建基于多边缘计算的追尾事故取证系统,使用100-Car数据集对驾驶行为做初步分析,得到追尾事故场景下的驾驶行为特征,将其归结为车辆的轨迹特征,使用隐马尔可夫模型解码可能的安全驾驶行为参数,结合追尾事故场景下的驾驶行为特点,制定合理的责任分配机制,以期减少纠纷从而加快事故处理和行车效率。更多还原

【Abstract】 As an important part of the development path of intelligent driving and intelligent traffic system,driving behavior has been paid much attention and research.By studying the driving behavior under the safe driving scene,we can understand the driving condition of the vehicle in real time,and then improve the driving safety effectively.By studying the reasonableness of driving behavior under the accident scene can reduce the dispute and speed up the accident treatment.In the safe driving scene,accurately estimating the end time for follow-up behavior and lane-change behavior of vehicle is conducive to early warning of vehicles to ensure driving safety.At the same time,the research on improving driving safety based on accident data is also increasing.Research on the rationality of driving behavior in accident scenario can reduce the dispute and improve the accident handling and driving efficiency.In view of the above problems,this paper designs a model to estimate the survival time of driving behavior based on the survival analysis method,the algorithm model that predicts vehicle intention and trajectory by using the deep learning algorithm Gated Recurrent Unit(GRU),and the model of driving behavior evaluation under the rear-end accident Hidden Markov Model(HMM)is proposed.The main research content of this paper includes:(1)A method for estimating the survival time of driving behavior based on the Survival Analysis model is designed.Under the safe driving scene,the vehicle trajectory information in the NGSIM dataset of the naturalistic driving data set is used to analyze the driving behavior of the two basic driving behaviors of the follow-up vehicle and the lane change.The survival analysis model is introduced to estimate the probability distribution of the survival time with the vehicle and the change of behavior,and then determine the behavior deadline to enhance the safety of the driving.(2)A vehicle intent detection and trajectory prediction model based on GRU algorithm is designed.Under the safe driving scene,according to the trajectory information of vehicle in the NGSIM dataset of the naturalistic driving data set,the characteristics of vehicle dynamics are analyzed,the characteristics of the vehicle are extracted,such as lateral speed and lateral acceleration of vehicle,and the vehicle trajectory prediction model of the fusion driving intention is constructed using the GRU of deep learning method,so as to make a more accurate prediction and classification of driving behavior.(3)A driving behavior evaluation strategy under the scene of a rear-end accident is proposed.Under the accident scene,the rear-end accident forensic system based on multi-edge calculation is constructed,the driving behavior is analyzed by using the 100-Car dataset,and the driving behavior characteristics under the rear-end accident scene are obtained,it is attributed to the trajectory characteristics of the vehicle,the possible safe driving behavior parameters can be decoded by the HMM,and the reasonable liability distribution mechanism should be developed with the characteristics of the rear-end accident scene,with a view to reduce the dispute to speed up the handling of accidents and driving efficiency.更多还原