基于深度学习的交通路口自动识别

Automatic recognition of road intersection based on deep learning

针对轨迹数据在交通路口的分布特性,提出一种基于长短期记忆的深度学习转弯轨迹模式自动提取方法,实现自动快速地提取转弯轨迹;然后,针对路口转弯点的稀疏性缺陷,提出一种联合补偿点计算和转弯轨迹航向变化幅度的方法选取路口候选点,并通过聚类路口候选点识别道路交叉口.采用福州市鼓楼区出租车实际轨迹数据进行验证测试,结果表明,该方法对道路交叉口识别准确率达到96.7%,可为电子地图实时更新及无人驾驶自动导航应用等提供关键技术支持.

Based on the distribution characteristics of trajectory data at traffic intersections,a long short term memory mode was applied to extract the turning trajectories,then the trained model could automatically and quickly extract the turning trajectory from the new unknown trajectory data.Then,a method was proposed to decide the intersection candidate points by combining the calculated compensation point and the closest to the intersection point.Finally,the intersection was identified by clustering the intersection candidate points.The taxi trajectory data of Fuzhou were used to conduct experiments.The results showed that the road intersection could be identified by the accuracy of 96.7%.This method can be widely applied to the continuous and rapid update of digital maps.