智能汽车环境精细感知、深度融合与动态建模方法

Environmental perception and modeling is the core content of intelligent vehicle. Whether the environmental information can be processed and analyzed correctly and promptly is directly related to the safety and efficiency of intelligent vehicle operation. It plays an important role in the path planning and the effect of vehicle control. At present, the level of information perception of intelligent vehicle can’t meet the requirements of automatic driving under full situation and complex environment. This project aims at the major needs of intelligent vehicle’s environmental perception and map construction. It will comprehensively use pattern recognition and real vehicle experiment methods. The research focuses on the key issues such as intelligent vehicle perception and map construction under inadequate information condition, the integration of multi-source sensing information, and comprehensive safety risk assessment considering human, vehicle and road factors. The focal point of this project is to explore the high precision dynamic positioning technology based on geographic visual labeling, precise perception and deep fusion of multi-source information under information missing condition, modeling method of driving scenario based on driving behavior understanding and person-attendance, and collision risk integrated situation sensing method based on "driving intention-vehicle state-dynamic environment". Finally the methods will be verified by field experiment and driving simulation experiment. The research results can complement and improve the existing environmental perception and map scene construction theory. It will provide the development and application of intelligent vehicles with strong support.

环境感知与建模是智能汽车的核心内容，环境信息能否正确及时的处理、分析直接关系到智能车辆运行的安全和效率，对路径规划和车体控制的效果具有决定性作用。目前，智能汽车信息感知水平还不能满足全工况、复杂环境下全自动驾驶的要求。本项目面向智能汽车环境感知和地图构建的重大需求，综合运用模式识别、实车实验等方法，围绕智能汽车在信息缺失条件下的感知和场景构建、多源传感信息的融合、考虑人车路因素的综合安全风险态势评估等关键问题开展研究，重点探索基于地理视觉标签的高精度动态定位技术、信息缺失条件下的多源信息精细感知与深度融合方法、基于驾驶行为理解与拟人认知的行车驾驶场景地图建模方法、基于“驾乘意图-车辆状态-动态环境”的碰撞风险综合态势感知方法，并通过实车和驾驶仿真实验进行测试验证，课题研究成果可补充和完善现有的环境感知和地图场景构建相关理论，为智能汽车的发展与应用提供理论支撑。

环境精细感知、深度融合与动态建模是智能汽车的核心内容，环境信息能否正确及时的处理、分析直接关系到智能车辆运行的安全和效率，对路径规划和车体控制的效果具有决定性作用。目前，智能汽车信息感知水平还不能满足全工况、复杂环境下全自动驾驶的要求。本项目面向智能汽车环境感知和驾驶地图构建的重大需求，提出了基于地理标签的低成本、高精度、车道级相对定位方法，构建了部分传感器信息缺失下智能车感知信息补偿方法，提出了基于驾驶行为理解与场景信息拟人认知的行车驾驶场景地图建模方法，建立了智能车行车风险量化和评估模型，并在城市复杂环境下进行测试验证。本项目的研究成果补充和完善了现有的环境感知和地图场景构建相关理论，为智能汽车的发展与应用提供理论支撑，能够促进智能汽车的应用和推广。