基于混合弱势道路使用者行为认知的行车避撞决策方法

我国城市混合交通中的弱势道路使用者种类繁杂、数量众多，且其行为具有高度的动态不确定性。现有面向弱势道路使用者保护的汽车主动安全系统难以实现准确快速的多目标感知和行为预测，且决策过程未考虑弱势道路使用者的风险特性，因此难以为中国混行的交通环境中弱势道路使用者提供有效保护。本项目针对弱势道路使用者保护的汽车主动安全系统的感知、认知和决策三个方面，重点研究：1）行人和骑车人的联合检测框架构建与参数自学习的启发式深度神经网络建模；2）融合环境、人体姿态和运动学等多元信息的行人和骑车人的意图识别与轨迹预测；3）基于人车路环境特性和相互作用机理的行人和骑车人的风险评估与最优避撞决策建模。以期提高行人和骑车人多目标检测的精度和效率，准确识别行人和骑车人的行为意图并进行轨迹预测，探究行人和骑车人的风险机理与主动安全系统的最优避撞方法。为针对弱势道路使用者的汽车主动安全系统的发展提供理论基础和技术支撑。

Vulnerable Road Users (VRUs) in China's urban mixed traffic are numerous in terms of type and quantity and their behavior is highly-dynamic and uncertain. It remains challengeable for the existing automotive active safety systems on VRUs' protection to detect multiple VRUs targets and recognize their future behaviors accurately and efficiently. Moreover, VRUs’ risky behavior characteristics are rarely considered in the decision making process. Therefore, the current active safety systems cannot provide effective VRUs protections in China mixed traffic scenarios yet. This project focuses on the perception, cognition and decision making of the automotive active safety systems for VRUs' protection, including: 1) construction of the joint detection framework for pedestrians and cyclists, and modeling based on a heuristic-based parameter self-learning deep neural network; 2) intention recognition and trajectory prediction of pedestrians and cyclists integrating multiple information, e.g., environment cues, human postures and kinematics information and 3) driving risk assessment and the optimal collision-avoidance decision-making modeling for pedestrians and cyclists based on human, vehicle and road environment attributes and interaction mechanism. This study expects to improve the accuracy and efficiency of multi-target detection for both pedestrians and cyclists, to accurately identify the behavioral intentions of pedestrians and cyclists and predict their trajectories, and to explore the risk causation of pedestrians and cyclists and generate optimal collision avoidance methods for the automotive active safety systems. Theoretical works and technical supports are provided for the development of automotive active safety systems with effective VRUs protections.