您的位置:首页  > 论文页面

基于融合多头注意力机制和门控循环单元的 驾驶员意图识别方法

发表时间:2023-03-30  浏览量:1780  下载量:106
全部作者: 阮钦,杨为
作者单位: 重庆大学机械与运载工程学院;重庆大学机械传动国家重点实验室
摘 要: 为识别自动驾驶条件下驾驶员的细分行为意图,通过搭建驾驶模拟器以及虚拟场景,并采用角传感器、踏板位移、力传感器等,进行加速、制动、转弯、巡航及驻车制动等多种工况的驾驶员在环实验。利用所得车辆运行以及驾驶员驾驶数据,进行数据归一化处理,并提出一种基于融合多头注意力机制(Multi-Head Attention)和门控循环单元(gate recurrent unit,GRU)的驾驶员意图识别模型,将处理数据作为模型输入,识别驾驶者多种细分驾驶意图。研究结果表明:相较于长短期记忆神经网络、单一GRU等方法,融合多头注意力机制和GRU网络模型,其精准率、模型收敛速度更优;利用多头注意力机制进行输入权重分配有助于提高驾驶意图识别模型的分类性能。本研究有助于识别多种细分驾驶意图,进而为智能驾驶辅助系统决策、控制提供参考。
关 键 词: 机械学;驾驶意图;GRU神经网络;多头注意力机制;驾驶模拟器
Title: Driver intention recognition method based on Multi-Head Attention and GRU model
Author: RUAN Qin, YANG Wei
Organization: College of Mechanical and Vehicle Engineering, Chongqing University; State Key Laboratory of Mechanical Transmissions, Chongqing University
Abstract: In order to identify the subdivision behavior intention of the driver under the condition of automatic driving, the driver in the loop experiments under various working conditions such as acceleration, braking, turning, cruise and parking braking were carried out by building a driving simulator and virtual scene, and using angle sensors, pedal displacement and force sensors. Using the obtained vehicle operation and driver driving data, the data was normalized, and a driver intention recognition model based on the fusion of Multi-Head Attention mechanism and gate recurrence unit (GRU) was proposed. The processed data was used as the model input to identify the driver’s multiple subdivision driving intentions. The results show that compared with long-term and short-term memory neural network and single GRU, the fusion of Multi-Head Attention and GRU network model has better accuracy and convergence speed, and that using Multi-Head Attention mechanism for input weight allocation helps to improve the classification performance of driving intention recognition model. This study helps to identify a variety of subdivision driving intentions, and then provides a reference for the decision-making and control of intelligent driving assistance system.
Key words: mechanics; driving intention; GRU neural networks; Multi-Head Attention; driving simulators
发表期数: 2023年3月第1期
引用格式: 阮钦,杨为. 基于融合多头注意力机制和门控循环单元的 驾驶员意图识别方法[J]. 中国科技论文在线精品论文,2023,16(1):8-20.
 
9 评论数 0
暂无评论
友情链接