驾驶人换道意图实时识别模型评价及测试

Evaluation and test of real-time identification models of driver′s lane change intention

通过分析驾驶人换道行为和车辆运动状态,研究了意图换道和车道保持阶段的差异性,并基于BP神经网络模型和证据理论识别模型,对意图换道进行了实时识别试验。结果表明:两种模型在换道前3s对意图换道样本识别准确率分别为78.26%、45.22%,在换道时刻的识别准确率分别为99.13%、86.96%;随机选择样本对两种识别模型进行验证,意图换道样本的识别准确率分别为86.00%、96.00%,车道保持样本的识别准确率分别为21.05%、78.95%,同时模型识别出正确样本的最长时间均小于0.5s,表明证据理论识别模型具有较高的优越性和实用性。

The difference between lane change intention stage and lane keeping stage was studied by analyzing the driver lane change performance and the vehicle movement parameters. Based on BP neural networks model and D-S evidence theory model, the real-time identification of lane change intension tests were carried out. The results show that that samp e identification accuracies of the two models in three seconds before the lane change are 78.3 %fro and 45.2 %fro respectively. The identification accuracies just at the moment of lane change are 99.13%and 86.96% respectively. By verifying the two models with random sample data, the identification accuracies for lane change intension are 86% and 96%, while for lane keeping are 21.05% and 78.95% respectively. The maximum time of the models for accurate identification is less than 0. 5 second. It indicates that the evidence theory identification model is more applicable and reliable in both accuracy and timeliness.