Quaternion deep neural network model method for identifying traffic signs of intelligent automobile

An intelligent automobile can use a sensor to collect road information, assists in driving through computing and analysis, and can help to ensure safe traffic. Automatic identification of traffic signs is one of the key technologies of intelligent automobiles. However, current methods fail to conduct combined mining and study on important colors, contours and time domain information of traffic signs within one uniform framework in an effective manner. To address the aforementioned problem, according to the invention, the method herein is based on a quaternion mathematical representation framework, provides a quaternion deep neural network model method in order to increase the robustness of identifying traffic signs, and provides an accurate model for conducting auxiliary driving research of intelligent automobiles. 智能汽车能利用传感器收集道路信息，通过计算分析进行驾驶辅助，有利于保障交通安全。对交通标志进行自动识别是智能汽车的关键技术之，然而，现有方法仍未在个统框架中有效对交通标志重要的颜色，轮廓及时间域信息进行联合挖掘学习。针对以上问题，本发明基于四元数的数学表示框架，提出了种四元数深度神经网络模型方法以提高交通标志识别的鲁棒性，从而为智能汽车的辅助驾驶研究提供了更准确的模型保障。