基于深度视觉学习的汽车辅助驾驶安全关键方法研究

Aiming at the key and difficult questions of moving objects recognition for the Driver Assistance Systems such as feature extraction and classifier design, Pedestrian captured by vehicle camera as the target will be used in this project to carry out our research of pedestrian detection and pedestrian’s walking direction estimation based on deep neural networks. The sparse data for deep conventional neural networks will be obtained based on the pedestrian characteristic view. By means of correlation analysis and hyperparameter optimization, the effect of the deep neural networks learning algorithm on image feature extraction ability, generalization ability and convergence performance of networks will be discussed. At the same time, a parallel optimization scheme based on BSP model and multiple GPU will be proposed for the fast realization of deep convolutional neural networks. Through the investigation of the project, the relationship between the mechanism of the moving objects recognition and realization of deep neural networks algorithm can be established. The research is of significant importance for opening out the key and basic question, such as feature extraction of moving object recognition for the Driver Assistance Systems, and promoting the research of moving object recognition theory and method in-depth. The research results have important application reference value in military and civil fields.

针对汽车辅助驾驶安全系统中运动目标识别的特征挖掘和识别分类等关键难点问题，本项目以车载视频中行人为对象，开展基于深度神经网络的汽车辅助驾驶安全技术中行人识别及行人行走方向分析的研究，项目拟在构建车载视频中行人特性视图的基础上，获得面向深度卷积神经网络的稀疏样本；以相关性分析和参数优化等为手段，阐明深度神经网络学习算法对图像特征挖掘能力、网络泛化能力及收敛性能的影响规律；为达到汽车辅助驾驶安全技术的实时性，以深度卷积神经网络的快速实现为目的，提出BSP模型和多GPU相结合的深度学习并行优化方案。通过研究，建立运动目标识别机理与深度神经网络实施算法的本征关系，提出结合行人行走方向信息的深度卷积神经网络耦合新模型。本项目对于揭示汽车辅助驾驶安全技术中的运动目标识别中的特征提取等关键基础问题，发展运动目标识别理论与方法具有重要意义，研究成果在军事和民用领域具有重要的应用参考价值。

项目组重点围绕国家和首都智慧城市、智能交通的重大战略需求，以基于深度视觉学习的汽车辅助驾驶安全系统科学与关键技术为主要研究方向，持续聚焦于智能交通中智能计算、计算机视觉、模式识别以及人工智能相关理论、方法、技术的系统研究，在智慧城市与智能交通领域取得了系列关键创新性技术成果。