阿里云机器学习PAI平台在驾驶行为识别上的应用

Application of Aliyun PAI Platform in Driving Behavior Recognition

随着具有车内外双拍摄功能行车记录仪的普及,如何通过分析车内驾驶人员实时影像,识别驾驶者状态,对分心驾驶、疲劳驾驶等危险行为进行及时预警已成为近年来研究的热点问题之一.该文基于阿里云机器学习PAI平台,利用Auto Learning自动学习模块,对1 100张含10种驾驶状态的车内影像数据构建了线上深度学习模型,模型训练耗时约11分钟,识别准确率可达99.08%,并通过与传统本地线下模型训练方式的对比,验证了基于PAI平台方案的可行性,为基于图像识别的驾驶行为检测提供了一种新的参考方法 .

With the popularization of driving recorders with dual camera function,how to analyze the realtime images of the driver in the car to identify the driver’s state and provide timely warning of dangerous behaviors such as distracted driving and fatigue driving has become a hot research topic in recent years.Based on the PAI platform,this article uses the Auto Learning module to construct an online deep learning model on 1 100 in-car image data containing 10 driving states. The model training takes about 11 min,and the recognition accuracy is up to 99.08%.In addition,this paper verifies the feasibility of the PAI platform solution based on the comparison with the traditional local offline model training method,and provides a new reference method for driving behavior detection based on image recognition.