基于单目摄像头的主动式驾驶行为分析算法
Active driving behavior analysis algorithm based on monocular camera

为了预防事故发生,提出一种以人体三维姿态估计对驾驶员行为进行识别监测的算法。利用单目摄像头获取运动中驾驶员的视频流,提取每帧图像的二维轮廓特征,与预先建立的三维人体模型的二维投影进行匹配,实时估计驾驶员上半身的姿态。根据获取驾驶员的8个骨骼节点的三维坐标,对驾驶员的行为识别分析。试验模拟驾驶员正常、单手、接听电话和疲劳/醉酒驾驶4种驾驶状态,通过骨骼节点的坐标变化,实现检测和识别驾驶员的姿态行为并给予提醒。在光线较好的情况下,与PRECLOSE(percent eye closure)算法相比,该算法的误检率降低了24.24%。
In order to prevent accidents, an algorithm for recognizing and monitoring the driver′s behavior based on the three-dimensional pose estimation of the human body was proposed. A monocular camera was used to capture the video stream of the driver in motion, the two-dimensional contour features of each frame of the image was extracted, and the two-dimensional projection of the pre-established three-dimensional human body model was matched to estimate the attitude of the driver′s upper body in real time. Based on the three-dimensional coordinates of the driver′s eight skeletal nodes, the driver′s behavior was identified and analyzed. Four driving states of driver′s normal, one-handed, answering calls and fatigue/drunk driving were simulated. Through the coordinate changes of the skeletal nodes, the gesture behavior of the driver could be detected and recognized, and the driver could be given reminders. When the light was enough, the algorithm could reduce the false detection rate by 24.24% compared with the PRECLOSE algorithm.