基于Haar-NMF特征和级联AdaBoost的脱岗检测算法

针对值班室人员的岗位执勤情况进行实时检测的需求,设计了一种脱岗检测系统,它使用低维的Haar-NMF特征代替Haar特征的方法以减少计算量,同时采用级联分类器代替单一的分类器结构以提高检测的准确性。该系统分为训练和测试两个部分,训练部分包括利用非负矩阵分解对Haar特征进行降维生成Haar-NMF特征和级联AdaBoost分类器的训练,经过数次训练后得到的各弱分类器根据训练过程中的权值加权组成一个强分类器,该分类器具有较高的学习效率,检测速度有明显提升。测试部分根据实际检测效果中存在的误检情况对检测算法进行修正和优化。实验验证了该系统具有相当高的检测率和较低的误检率,有助于避免在岗位无人值守时发生意外及损失财产。

Detection Algorithm for Off Position Personnel Based on Haar-NMF Features and Cascading AdaBoost

A novel detection system for off position personnel was designed for real-time detecting requirements.It replaces the Haar features with low-dimensional Haar-NMF,which can greatly reduce the calculation amount.At the same time,a cascaded classifier is used for instead of a single classifier structure to improve the accuracy of the detection.The system is divided into two parts,including training and testing.The training part includes training the cascading AdaBoost classifier and utilizing non-negative matrix factorization to lower the dimension of Haar features to generate Haar-NMF features.The weak classifiers obtained by several trainings can constitute a weighted strong classifier based on the weights in the training process.This classifier has a high learning efficiency and the detection speed is significantly improved.The testing part can optimize the detection algorithm based on error detections.This system can avoid accidents and property losses when the duty officers leave their posts and has a low false detection rate which is verified by experiments.