考虑车辆间交互作用的驾驶意图预测方法

A Driving Intention Prediction Method Considering Vehicle Interaction

摘要:

准确的意图预测可以帮助智能车辆更好地了解周围环境并做出更加安全的决策，从而提高自动驾驶的安全性，促进人机协同驾驶。为了对驾驶员未来的意图做出更加精准的预测，提出了一种交互式意图预测方法。首先，通过将隐马尔可夫模型（HMM）与高斯混合模型（GMM）相结合，在充分考虑周围场景信息后建立了行为识别模型，用于对当前的驾驶行为做出准确的判断。然后，考虑到交通场景复杂多变的特点，提出基于意图的轨迹预测方法规划出一条最佳的行驶轨迹，并采用最大期望效用理论对未来的驾驶行为进行推理。由于行为识别和意图推理模型综合考虑了交通态势的演变过程和车辆之间的交互作用，所以将两个模型得到的结果相结合可得到车辆最终预测出的驾驶意图。最后，在NGSIM数据集对所提出的方法进行验证，结果表明提出的行为识别模型能够提前0.2~0.3 s识别出车辆的换道意图，结合未来意图推理模型，能够更加准确地预测出车辆未来的驾驶行为，由此可提高车辆驾驶的安全性。

Abstract:

Accurate intention prediction can help intelligent vehicles better understand the environment and make safe decisions， thus improving the safety of automatic driving and promoting cooperative driving. This paper proposes an interactive intention prediction method， which makes a more accurate prediction of the driver's future intention. First， the Hidden Markov Model （HMM） and Gaussian Mixture Model （GMM） are combined to establish a behavior recognition model， which fully considers the surrounding scene information and accurately judges the current driving behavior. Then， an intention-based trajectory prediction method is proposed to plan the best driving trajectory considering the complex and changeable characteristics of the traffic scene， and the maximum expected utility theory is used to infer the future driving behavior. The behavior recognition and intention reasoning models comprehensively consider the evolution process of the traffic situations and the interaction between vehicles， and the final predicted driving intention of vehicles is obtained by combining the results of the above two models. Finally， the proposed method is verified in the NGSIM dataset， and the results show that the proposed behavior recognition model can recognize the lane changing intention of the vehicle 0.2 to 0.3 seconds in advance. Combined with the future intention reasoning model， it can more accurately predict the future driving behavior of the vehicle and improve driving safety.