机构地区: 清华大学机械工程学院汽车安全与节能国家重点实验室
出 处: 《汽车工程》 2012年第10期889-893,共5页
摘 要: 通过机器视觉技术对眼睛动作和视线转移特征的分析可实现驾驶人警觉状态的有效估计,但实际行车过程中驾驶人面部姿态的不确定性变化对眼睛定位算法提出了严峻挑战。本文中在采用主动形状模型算法对面部区域进行配准的基础上,提出运用Lucas-Kanade光流进行全局跟踪,并采用基于自商图的Meanshift算法进行局部校准的跟踪策略。实验结果表明,Meanshift算法的局部极值化能力能有效消除Lucas-Kanade光流跟踪中的误差积累,有效提高了人眼定位与跟踪精度。 The analysis on eye movement and the feature of sight diversion by means of machine vision technology can achieve effective estimation of driver's vigilance state. However, the uncertainty change of driver's face gesture in real driving poses a severe challenge to eye positioning algorithm. In this paper, based on face align-ment by using active shape model, an eye tracking strategy is proposed, which uses Lucas-Kanade optical flow method for global tracking experiment show that the the error accumulation in and uses self-quotient-image-based Meanshift algorithm for local alignment. The capability of Meanshift algorithm in Lucas-Kanade optical flow tracking, converging to local extrema can effectively and hence increase the accuracy of human results of eliminate eye posi-tioning and tracking.
领 域: [自动化与计算机技术] [自动化与计算机技术]