Sensor Integration for Autonomous Vehicle Self-Localization in Urban City

城市环境下的车辆自定位是自动驾驶和驾驶辅助领域的一个具有挑战性但意义重大的课题。运动规划和车辆协作都需要准确的位置信息。本研究主要针对被动式与主动式两种车辆定位系统进行研究，首先提出将3D地图导航系统与惯性测量单元（IMU）、车速表、车载摄像头等被动式传感器相结合。我们在东京市的不同地方进行了一系列测试。实验结果表明，所提出的基于被动传感器的定位系统可以实现亚米级的定位精度相对于定位误差的平均值。在此基础上，重点研究了基于主动传感器的抽象地图车辆自定位问题。在东京的一个城市地区进行的实验表明，即使我们极大地缩小了地图的大小，我们可以保持约50厘米的定位的平均误差。在过去的一年里，我们将研究扩展到基于智能手机的行人定位和导航系统。研究结果表明，利用地图和上下文信息可以提高城市环境中行人定位的准确性。

Vehicle self-localization in urban environment is a challenging but significant topic for autonomous driving and driving assistance. Both motion planning and vehicle cooperation need the accurate position information. This research focused on both passive sensor-based and active sensor-based vehicle self-localization systems.At the beginning of this research project, we firstly proposed to integrate 3D map based GNSS with other passive sensors: Inertial Measurement Unit (IMU), vehicle speedometer and an onboard camera. We conducted a series of tests in different places of Tokyo city. The experiment results demonstrate that the proposed passive sensor-based localization system can achieve sub-meter accuracy with respect to positioning error mean. After that, we focused on active sensor-based vehicle self-localization with the abstract map. Experiments conducted in one of the urban areas of Tokyo show that even though we extremely shrank the map size, we could preserve the mean error of the localization about 50 centimeters. In the past year, we extend our research to smartphone based pedestrian positioning and navigation system. Research result indicated that using map and context information can improve pedestrian positioning accuracy in the city urban environment.