路面车辆行为分析与情境理解

From the viewpoint of an ego-vehicle during its on-road naturalistic driving, using its on-board sensors and data processing, motion observations of both the ego- and environmental vehicles are acquired, based on which the methods of behavior analysis, contextual understanding and online reasoning are studied in this project. Major challenges of the research are: 1) moving viewpoint, various behaviors, complex procedure and dynamic interaction; 2) changing road conditions; 3) accurate and efficient detection and tracking of the on-road vehicles from a driving viewpoint in complex traffic. This project consists of studies at three levels, which are behavior analysis and modeling, reasoning, and on-road vehicle detection and tracking, and aims at solving: 1) contextual analysis and modeling on the ego-vehicle’s trajectory; 2) behavior analysis and semantics learning of the environmental vehicles; 3) content- and context-based behavior reasoning; 4) contextual analysis and multi-modal sensor fusion based on-road vehicle detection, tracking and trajectory collection. In this project, algorithms are studied and examined through on-road data collection using a real intelligent vehicle platform, and onboard demonstrations are developed to give an intuitive prototype for potential applications. The research provides analysis and reasoning methods in developing more reasonable motion planning and decision algorithms for autonomous driving systems, and helps in improving processing capabilities of the driving assistance system in complex environments.

项目以路面自然行驶中的主体车为视点，利用其携带的传感器及数据处理，采集主体和环境车辆的运动观测信息，并籍此建立行为分析、情境理解和在线推理方法。项目研究面临的重大挑战包括：1）动态视点、行为多样、过程复杂、动态交互；2）道路交通条件变化；3）复杂交通动态行驶中精确有效的路面车辆检测跟踪。项目从行为分析建模、行为推理、路面车辆检测跟踪等三个层面开展研究，力图解决：1）主体车轨迹的情境分析建模；2）环境车辆行为分析和语义学习；3）融合内容与情境的行为推理；4）基于情境分析和多模态传感器融合的车辆检测跟踪与轨迹提取。项目基于智能车实验平台开展真实公路数据采集、算法研究与实验验证，同时研发在线演示，为应用提供直观的系统原型。项目成果不仅可以为建立更为合理的自动驾驶运动规划与决策提供分析推理方法，也可以为提高安全辅助驾驶系统在复杂环境中的处理能力提供基础手段。

项目以路面自然行驶中的主体车为视点，利用其携带的传感器及数据处理，采集主体和环境车辆的运动观测信息，并籍此建立行为分析、情境理解和在线推理方法。项目制定了四年研究计划，并开展了以下三方面的研究: 1) 行为分析建模，包括主体车轨迹情境分析建模、环境车辆行为分析和语义学习, 2) 融合内容与情境的行为推理，包括轨迹预测与行为规划、行为识别与决策推理, 3) 路面车辆检测跟踪，包括情境分析推理的车辆检测跟踪、多模态传感器融合的车辆轨迹提取。项目基于真实智能车平台的复杂公路交通数据采集，围绕路面车辆行为分析与情境理解开展了上述算法研究，并通过基于数据的算法验证与在线平台演示，验证了算法的有效性。项目成果在在智能车及智能交通领域最有影响力的国际期刊IEEE Trans on Intelligent Transportation Systems（影响因子6.064）上发表6篇论文，在自动化学会A类期刊、自动化学报上发表2篇论文，在智能车、智能交通、机器人领域的国际顶会IROS、ICRA、IV、ITSC上发表12篇论文，其中1篇论文获得IV2019最佳学生论文奖提名，申请中国发明专利4项。项目成果不仅可以为建立更为合理的自动驾驶运动规划与决策提供分析推理方法，也可以为提高安全辅助驾驶系统在复杂环境中的处理能力提供基础手段。

内容获取失败，请点击重试