基于多尺度动态分析和环境匹配的疲劳预警干预技术研究

Drowsy drive is a major threats of traffic safety, this makes the detecting driver drowsiness accurately and implement warning intervention has very important significance. However, the normal drowsiness warning system has several technical problems such as high false detection rate and low acceptance due to the complexity of the drowsiness and the actual road environment. For this case, the research of this project are as follows: the robust detection method of the drowsiness characteristics under the traffic situation of various illuminations and driver poses, the dynamic identification method of drowsiness state based on the temporal information of facial expression, the design of warning intervention strategies based on real road environment and drivers’ perceptual mechanism. This research intends to explore the inherent nature of the instability of facial landmark location in real traffic environment, the dynamic model to express driver drowsiness, and the inherent mechanism of interactions between driver and the warning when drowsiness changes. This research also intends to find the drowsiness warning system design methodology for using it in the real road finally. In summary, the purpose of this research is to promote the expansion of applications of the drowsiness warning system in real traffic environment, and to provide some new technical means and theoretical supports for vehicle active safety technology research and development in China.

疲劳驾驶是造成交通事故的重大安全隐患，准确检测驾驶人疲劳状态并对其实施适当的预警干预对改善交通安全具有重要的意义。但疲劳状态以及实际道路的复杂性使得疲劳预警干预技术在实际应用时面临疲劳状态误检率高，信息交互接受度低等关键问题。本课题拟围绕光照、姿态鲁棒的疲劳特征检测、基于面部表情时空特征的疲劳状态判别、符合驾驶人感知特性的信息干预这几项关键科学问题,研究复杂光照与姿态条件下的面部特征点定位理论与方法，探索疲劳状态的动态变化规律与表达模型，以及多因素约束下的人车信息交互机理，进而揭示驾驶人的疲劳状态和客观交通环境对疲劳预警干预技术实际应用的制约机制，为车辆先进安全技术的研究和发展提供新的技术手段和理论支持。

疲劳驾驶是造成交通事故的重大安全隐患，准确检测驾驶人疲劳状态并对其实施适当的预警干预对改善交通安全具有重要的意义。但疲劳状态以及实际道路的复杂性使得疲劳预警干预技术在实际应用时面临疲劳状态误检率高，信息交互接受度低等关键问题。.针对这些问题，本课题首先深入分析了实际行车环境中光照条件、驾驶人姿态变化对眼睛定位算法适应性的影响，建立了基于层叠式形状模型和自商图局部纹理模型的主动形状模型算法，实现了眼睛局部邻域的有效分割和可靠跟踪。同时，充分利用自商图、色度以及梯度的统计分布等光照不变特征，建立了光照不变量约束下的参数化模板算法，实现了眼睛轮廓的精确定位。在此基础上，采用统计学方法分析论证了不同疲劳水平下眼睛动作参数差异的显著性，建立了基于眼睛动作特征的疲劳特征空间。.本研究还从危险感知机理的角度出发，深入研究了不同疲劳状态的驾驶人对交通危险状况的感知特性，分析了适应驾驶员疲劳状态的碰撞预警系统在预警时机和预警方式上所应遵循的设计要求，建立适应不同疲劳状态的综合预警策略，提出以驾驶人的危险感为控制指标的预警时机调节方法，实现了根据驾驶人不同疲劳状态自适应调节危险状况预警时机的目的。