无人车自主定位和障碍物感知的视觉主导多传感器融合方法（英文）

发布时间：2018-06-20 05:07

本文选题：视觉感知 + 自主定位　；参考：《Frontiers of Information Technology & Electronic Engineering》2017年01期

【摘要】：人类驾驶与自主驾驶在对交通环境的理解方式上有着明显差别。首先,人主要通过视觉来理解交通场景,而机器感知需要融合多种异构的传感信息才能保证行车安全。其次,一个熟练的驾驶员可以轻松适应各种动态交通环境,但现有的机器感知系统却会经常输出有噪声的感知结果,而自主驾驶要求感知结果近乎100%准确。本文提出了一种用于无人车交通环境感知的视觉主导的多传感器融合计算框架,通过几何和语义约束融合来自相机、激光雷达(LIDAR)及地理信息系统(GIS)的信息,为无人车提供高精度的自主定位和准确鲁棒的障碍物感知,并进一步讨论了已成功集成到上述框架内的鲁棒的视觉算法,主要包括从训练数据收集、传感器数据处理、低级特征提取到障碍物识别和环境地图创建等多个层次的视觉算法。所提出的框架里已用于自主研发的无人车,并在各种真实城区环境中进行了长达八年的实地测试,实验结果验证了视觉主导的多传感融合感知框架的鲁棒性和高效性。
[Abstract]:There are obvious differences between human driving and autonomous driving in understanding the traffic environment. Firstly, people mainly understand traffic scenes through vision, and machine perception needs to fuse various heterogeneous sensing information to ensure traffic safety. Secondly, a skilled driver can easily adapt to a variety of dynamic traffic environments, but the existing machine sensing system often outputs noisy results, and autonomous driving requires that the sensing results are nearly 100% accurate. In this paper, a visual dominant multi-sensor fusion computing framework for traffic environment perception of unmanned vehicles is proposed. The information from camera, lidar and GIS is fused by geometric and semantic constraints. This paper provides high precision autonomous positioning and accurate robust obstacle sensing for unmanned vehicles, and further discusses the robust visual algorithms which have been successfully integrated into the above framework, mainly including training data collection, sensor data processing, and so on. Low-level feature extraction to obstacle recognition and environment map creation and other multi-level visual algorithms. The proposed framework has been used for self-developed unmanned vehicles and has been tested in various real urban environments for up to eight years. The experimental results verify the robustness and efficiency of the vision-led multi-sensor fusion sensing framework.
【作者单位】： Lab
【基金】：supported by the National Key Program Project of China(No.2016YFB1001004) the National Natural Science Foundation of China(Nos.91320301 and 61273252)
【分类号】：TP391.41;TP212

【相似文献】