【作者】 张祖涛；

【导师】 张家树；

【作者基本信息】 西南交通大学 ， 通信与信息系统， 2010， 博士

【摘要】 研究驾驶员眼动机制及其非线性眼动跟踪技术、探讨眼动与驾驶状态的关系已成为辅助安全驾驶研究的前沿热点之一。本论文以现实驾驶环境下的驾驶员注意力监测为研究对象,以协调非线性跟踪的精确性、鲁棒性以及算法的实时性之间的矛盾为目标,开展驾驶员非线性眼动跟踪研究。具体工作如下：一、提出基于二维正交Log-Gabor滤波的驾驶员人眼检测算法。首先研究基于Harr特征的驾驶员人眼检测方法,虽然该方法检测速度快,但在夜晚驾驶条件下的驾驶员人眼检测精度较低。为此,提出了基于二维正交Log-Gabor滤波的驾驶员人眼检测算法,能有效避免光照对驾驶员人眼检测的影响。二、提出一种基于自适应模糊强跟踪EKF滤波算法的驾驶员眼动跟踪方法。首先基于人眼解剖学的视觉生理基础,结合现有的眼动跟踪研究成果,分析现实驾驶环境下驾驶员眼动模型,提出一种自适应模糊强跟踪EKF滤波算法的眼动跟踪方法。通过模糊逻辑控制动态调整弱化因子,提高算法的滤波精度；将有限差分替代强跟踪EKF滤波算法中的非线性函数的偏导数,使得滤波算法复杂度减低,更简单。最后将该算法用于驾驶员眼动跟踪,并对基于红外光源的驾驶员眼动跟踪中红外光源对人眼伤害进行评估。通过初步的理论推导和现实驾驶环境下的眼动跟踪、基于红外光源的驾驶员眼动跟踪实验表明,基于自适应模糊逻辑有限差分的强跟踪EKF滤波算法具有较高的滤波精度和鲁棒性。三、为提高驾驶员非线性眼动跟踪的实时性,提出一种低复杂度的自适应强跟踪简化UKF滤波算法。该算法通过合理减少强跟踪UKF滤波中Sigma点的采样策略,在保证跟踪精确性和鲁棒性条件下,降低算法复杂度。理论分析和实验结果表明,基于自适应强跟踪简化UKF滤波的驾驶员眼动跟踪算法能有效缓解非线性眼动跟踪的精确性、鲁棒性以及算法的实时性之间的矛盾。四、提出了基于多尺度强跟踪粒子滤波的驾驶员眼动跟踪算法。针对强跟踪粒子滤波对驾驶员眼动跟踪精度不高的问题,结合小波变换多尺度原理和强跟踪粒子滤波,提出了基于多尺度强跟踪粒子滤波的驾驶员眼动跟踪算法,有效地降低强跟踪粒子滤波重要性权值的方差,提高滤波精度。理论分析和实验证明,基于多尺度强跟踪粒子滤波的驾驶员眼动跟踪算法为强跟踪滤波器,能有效地提高驾驶员眼动跟踪精度和鲁棒性。五、基于眼动跟踪算法的驾驶员疲劳检测研究。采用驾驶员疲劳检测评估方法,比较研究了基于眼动跟踪算法的驾驶员疲劳检测性能,给出了一个真实驾驶环境下的驾驶员疲劳检测系统实现方案。总之,本论文围绕着基于采样强跟踪非线性滤波理论的驾驶员眼动跟踪展开研究,提出了一种驾驶员人眼检测算法和三种基于强跟踪非线性滤波的驾驶员眼动跟踪方法。对算法的精确性、鲁棒性和实时性进行理论分析和实验评估,有助于设计出适合现实驾驶环境下基于驾驶员眼动跟踪算法的疲劳检测系统并最终商用。更多还原

【Abstract】 The research of driver eye movement mechanism and nonlinear eye tracking is one of hotspots in the assistant safe driving system based on the relationship between driver eye tracking and driving status. This paper achievements on tacking the conflicts about accuracy, robustness and real time of driver nonlinear eye tracking are classified into the following categories.The first is to propose a new driver eye detection based on 2-D Log-Gabor filter. Research on the algorithm of Harr feature method for eye detection, the accuracy of eye detection is low under night driving condition. We proposed the 2-D Log-Gabor filter to eye detection in night driving condition, which avoids the effect of illumination for image.The second is to propose a novel adaptive fuzzy strong tracking finite-difference extended kalman filter for driver eye tracking. The basic work is a study on the eye movement mechanism of the realistic driving conditions based on the eye anatomical vision physiology. By monitoring the residual mean and standard deviation, the fuzzy logic adaptive controller of this method dynamically adjusts the softening factor according to fuzzy rules. In filtering calculation, strong tracking factor is introduced to modify a priori covariance matrix and improve the accuracy of the filter. The filter uses finite-difference method to calculate partial derivatives of nonlinear functions for eye tracking. The experimental results show the validity of our method for eye tracking under realistic conditions.The third category is low complicated adaptive strong tracking simplified unscented Kalman filter(STS-UKF) for driver eye tracking. The reduced sigma points UT parameterizations can capture distribution information comparable to that of the 2n +1 symmetric UT, and reduce computational resources comparable to the n+2 UT. At the same time, strong tracking filtering (STF) is introduced into simplified UKF to improve the robustness. Suboptimal fading factors of STF are used to time update equations and measurement update equations to improve robustness of algorithm. The theory analysis and simulations show the STS-UKF can improve the computational efficiency and robustness for real-time eye tracking.The fourth is that we proposed a novel eye tracking base on multiresolution decomposition strong tracking particle filter. Nonintrusive methods for eye tracking are important driver fatigue detection. In order to improve the accuracy and stability of eye tracking, we can use the UKF to generate the proposal distribution for the PF (Particle Filter). Then, we can reduce the variance of important weights of above particle filter using wavelet multiresolution decomposition because of the wavelet multiresolution decomposition having a good property of denoising, which can improve the accuracy and robustness of eye tracking under realistic driving condition. At same time,we introduce STF into partice filter to resolve the nonlinear tracking of eye movement. The experimental results and the theoretical analysis show that the proposed method achieves higher estimation accuracy and robustness of eye tracking to head rotation, light variations and non-linear estimation in realistic driving condition.The last is to propose a scheme of drive fatigue detection based eye tracking. After studing the driver fatigue detection technology, we discuss the related driver fatigue detection based eye tracking. proposed a driver fatigue detection system model under realistic driving condition.In summary, the thesis mainly focuses on the research of technology for driver’s eye tracking based on sampling strong tracking nonlinear filter theory. The quantitative evaluation of eye tracking accuracy, robustness and real time performance will be helpful to design the algorithm for driver fatigue detection system.更多还原