摘要
为了在多车道行驶场景中准确判断出旁车并线意图,提出了一种基于贝叶斯网络的旁车并线意图识别模型。首先,从运动轨迹数据集中筛选样本构建训练集和测试集,选取特征参数,并利用卡方分裂算法对其进行离散化预处理。其次,在评分搜索的网络结构学习基础上,加入随机抽样来避免局部最优,通过构造接近目标平稳分布的马尔可夫链,多次迭代直到收敛到平稳状态来选取最佳网络结构。然后,使用贝叶斯估计法学习网络参数,搭建基于贝叶斯网络的旁车并线意图识别模型。最后,以测试样本进行验证。验证结果表明:基于贝叶斯网络的旁车并线意图识别模型,对并线样本和直行样本的识别准确率分别高达94.76%和96.38%,具有良好的识别精度。减少任意一个纵向参数的模型识别准确率都大于87%,而减少任意一个横向参数的模型识别准确率都小于76%,横向参数对识别效果的影响力大于纵向参数。
To accurately identify the cut-in intention of a side-lane vehicle in a multi-lane driving scene,a side-lane vehicle cut-in identification model was proposed based on the Bayesian Network.First,samples were screened from the motion data set to construct a training set and a test set,and characteristic parameters were selected and discretized by using the chi-square splitting algorithm.Second,on the basis of the score searchbased network structure learning,the random sampling was added to avoid local optimum,and the optimal network structure was selected by constructing a Markov chain close to the target’s stationary distribution and iterating it repeatedly until it converged to a stationary state.Then,the Bayesian estimation method was used to learn network parameters and build a side-lane vehicle cut-in identification model based on the Bayesian Network.Finally,the test samples were used for validation.The results showed that the Bayesian network-based side-lane vehicle cut-in identification model exhibits an accuracy of 94.76%and 96.38%for cut-in samples and go-straight samples,respectively.The accuracies are greater than 87%when a vertical parameter is deducted and the accuracies are less than 76%when a horizontal parameter is deducted from the model.It can be seen that horizontal parameters have a greater impact on the identification effect than vertical parameters.
作者
姜顺明
匡志豪
王奕轩
吴朋朋
JIANG Shunming;KUANG Zhihao;WANG Yixuan;WU Pengpeng(School of Automotive&Traffic Engineering,Jiangsu University,Zhenjiang 212013,China)
出处
《河南科技大学学报（自然科学版）》
CAS
北大核心
2021年第5期39-44,M0004,共7页
Journal of Henan University of Science And Technology:Natural Science
基金
国家自然科学基金项目(51575239)。
关键词
贝叶斯网络
意图识别
随机抽样
卡方分裂
Bayesian network
intention recognition
random sampling
Chi-square split
