基于深度学习的驾驶员打电话行为检测方法

A Driver’s Calling Behavior Detection Method Based on Deep Learning

为避免驾驶员因使用手机而无法对突发事故做出及时处理现象的发生,通过视频分析技术对驾驶员行为进行实时监控变得尤为重要。针对目前已有检测方法因存在异物遮掩、图像旋转、光照变化及难以提取图像深层特征等缺点而导致检测精度较低的问题,文章提出了一种基于深度学习的驾驶员打电话行为检测方法:首先采用渐进校准网络(progressive calibration networks,PCN)算法实现人脸检测及实时跟踪,从而确定打电话检测候选区域;然后采用基于卷积神经网络算法在候选区域实现驾驶员打电话行为检测。实际场景驾驶检测结果表明,本文所提方法不仅鲁棒性高,而且精度达到96.56%,误检率为1.52%,处理速度达到25帧/s,可以有效地进行驾驶员打电话行为检查监测。

In order to prevent the driver from being distracted by the cell phone call,real-time monitoring of drivers'behavior through video analysis is especially important.At present,driver's calling behavior detection methods are prone to object occlusion,image rotation,illumination change and are difficult to extract deep features of the image,which degrade the detection accuracy.This paper proposed a driver's cell phone calling behavior detection algorithm based on deep learning.The algorithm comprises two steps.Firstly,face detection and face tracking is supported by PCN(progressive calibration networks)to determine the calling detection area.Secondly,the driver's cell phone calling behavior detection method based on convolutional neural network is used to detect the cell phone in the candidate area.Experimental test results show that the proposed algorithm has high robutness,and its accuracy reaches 96.56%,the false positive rate reaches 1.52%,and the processing speed reaches 25 frames per second.It can effectively detect the driver's calling behavion.