基于混合高 斯-隐马尔可夫模型的驾驶意图识别方法研究

Research on Driving Intention Recognition Method Based on Guassian Mixed Model and Hidden Markov Model

驾驶意图的准确识别,能为车辆轨迹预测和行驶风险评估提供有力的理论支撑。首先,提取大量车辆轨迹数据,通过拼接重构方法得到同时包含3类驾驶意图的轨迹数据;其次,以车辆横向速度、加速度以及偏移量作为驾驶意图的表征参数,构建基于混合高斯—隐马尔可夫模型的驾驶意图识别方法,并通过对比不同识别窗口长度得到:窗口长度为2 s时总体精度最高;然后,为避免单点误判对模型精度的影响,设计1种多点识别的修正方法,对驾驶意图结果进一步修正,多次测试发现以连续3帧作为观察窗口时修正效果最好,驾驶意图的识别率高达98.84%。研究成果能应用于轨迹预测和风险评估中,进而为道路交通安全性的提高起到一定的推动作用。

Accurate identification of driving intention can provide strong theoretical support for vehicle trajectory prediction and driving risk assessment.Firstly,a large amount of vehicle trajectory data was extracted,and trajectory data containing three types of driving intentions was obtained through concatenation and reconstruction methods.Secondly,the lateral velocity,acceleration,and offset distance are selected to represent driving intention features,a driving intention recognition method based on Guassian mixed model and hidden Markov model is constructed.By comparing different recognition window lengths,it is found that the overall accuracy is highest when the length is 2 s.Then,a multi-point recognition correction method was designed for avoiding the impact of single point misjudgment on model accuracy.Multiple tests found that the best correction effect was achieved when using three consecutive frames as the observation window,with a recognition rate of up to 98.84%for driving intention.The research results can be applied to trajectory prediction and risk assessment,thereby playing a certain role in promoting the improvement of road traffic safety.