摘要
在手机通话(MPC)行为中,驾驶人极易陷入认知分心(DCD)状态,对此提出了一种基于头-眼行为特性的DCD图像识别方法。为适应自然驾驶中的波动光照和复杂背景,首先建立基于YCbCr色彩空间的在线肤色模型,提取待检肤色区域的PCA-HOG特征并建立支持向量机分类器来识别MPC手势;与此同时,采用多尺度局部模极大值方法检测嘴部显著边缘,并通过边缘活跃度来识别驾驶人说话行为,综合MPC手势和说话行为建立MPC行为的判别逻辑。最后,以5 s为时间窗口获取驾驶人的眼球活跃度、眨眼指数、头部横摆和俯仰运动活跃度,采用D-S证据理论建立融合头-眼行为特性的DCD识别方法。试验结果表明:融合手势和说话行为图像检测的MPC识别率为92.8%;对于不佩戴眼镜的驾驶人,眼球活跃度是DCD识别率最高的单一指标,"眼球活跃度-头部横摆活跃度-头部俯仰活跃度"融合证据的DCD识别率最高,为86.2%;对于佩戴眼镜的驾驶人,"头部横摆活跃度-头部俯仰活跃度"融合证据的DCD识别率最高,为83.2%;算法对熟练驾驶人的DCD识别率略高于非熟练驾驶人。
During mobile phone call(MPC) behavior, drivers are prone to falling into a cognitive distraction(DCD) state. In this regard, a DCD image recognition method based on head and eye behavioral characteristics is proposed in this paper. An online skin color model based on the YCbCr color space was established to adapt to the fluctuating illumination and complex backgrounds during natural driving. The principal component analysis and histogram of oriented gradients(HOG) feature of the candidate skin region was extracted and a support vector machine classifier was used for MPC hand gesture recognition. Additionally, the multi-scale local modulus maximum method was used to detect significant mouth edges, and the driver’s speech behavior was identified by calculating the activity of these edges. The discrimination logic of MPC behavior was then established by the integration of MPC hand gestures and speech behavior. On this basis, eyeball activity, blink index, and head rotation activity were obtained within a 5-second-time window. Finally, a DCD recognition method was established using the Dempster-Shafer evidence theory based on the feature fusion of head-eye movements. The results of driving experiments show that the MPC recognition rate based on the fusion of hand gesture and speech behavior is 92.8%. For drivers without glasses, eye activity is the single index with the highest recognition rate of DCD, and the fusion evidence of "eye activity-head yaw activity-head pitch activity" reaches the highest DCD recognition rate of 86.2%. For drivers with glasses, the fusion of "head sway activity-head pitch activity" reaches the highest DCD recognition rate of 83.2%. In addition, the DCD recognition rate of skilled drivers is slightly higher than that of unskilled drivers.
作者
程文冬
付锐
马勇
周扬
刘京凯
CHENG Wen-dong;FU Rui;MA Yong;ZHOU Yang;LIU Jing-kai(School of Mechatronic Engineering,Xi'an Technological University,Xi'an 710032,Shaanxi,China;Key Laboratory of Automobile Transportation Safety Technology,Ministry of Transport,Chang'an University,Xi'an 710064,Shaanxi,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2021年第5期168-181,共14页
China Journal of Highway and Transport
基金
国家自然科学基金项目(51775053,52072046)
陕西省自然科学基础研究计划项目(2018JM5158)。
关键词
交通工程
认知分心
机器视觉
智能辅助驾驶
行为特性
D-S证据理论
驾驶人
traffic engineering
cognitive distraction
machine vision
intelligent assisted driving
behavioral characteristics
D-S evidence theory
driver
