基于LBP和级联XGBoost的驾驶员打电话行为检测算法

Driver Call Behavior Detection Algorithm Based on LBP and Cascade XGBoost

基于机器视觉检测手机位置是检测驾驶员打电话行为的重要方法和依据。为解决实际应用中光照、姿态等因素对检测效果的影响,笔者提出了一种基于LBP特征和级联XGBoost的打电话行为检测算法,用于筛选滑窗采集到的手持电话样本,同时其应用基于MLP(Multi-LayerPerceptron)神经网络进行回归校准,进而得到更准确的位置。LBP的特征提高了打电话行为的辨识度,级联XGBoost和MLP网络回归提高了检测效率和定位准确度。实验表明,LBP与XGBoost级联分类器组合分类效果良好,构建的MLP网络能够有效拟合滑窗采集样本的偏移值。

Detecting the location of mobile phone based on machine vision is an important method and basis for detecting drivers'telephone behavior.In order to solve the influence of illumination,posture and other factors on the detection effect in practical application,a call behavior detection algorithm based on LBP feature and cascaded XGBoost is proposed,which is used to screen handheld phone samples collected by sliding windows.At the same time,MLP(Multi-Layer Perceptron)neural network is used for regression calibration to get more accurate location.The feature of LBP improves the identification of call behavior.Cascaded XGBoost and MLP network regression improve the detection efficiency and location accuracy.Experiments show that the combination of LBP and XGBoost cascade classifier has good classification effect,and the constructed MLP network can effectively fit the migration value of sliding window samples.