基于集成学习的人眼定位和头部状态检测研究

Research on Eye Location and Head State Detection Based on Integrated Learning

针对人眼定位中的实时性检测问题,设计了基于Viola-Jones算法的人眼实时检测系统,通过MATLAB控制外接或网络摄像头对检测到的人眼图片进行实时读取和实时定位,通过正面图、侧视图、俯仰图的检测,表明人眼实时定位系统检测效果较好并具有很好的鲁棒性。针对小数据集对头部状态的分类效果较差的情况,使用随机森林算法将俯仰角和偏航角的分别用HOG-LBP融合特征和Haar-like特征分类,再将得到的俯仰角和偏转角进行融合,在Pointing’04数据集上对比直接分类准确率提升了4.5%。

For real-time detection of eye location problem, a real-time detection system of human eye based on Viola Jones algorithm is designed. Through the MATLAB, the real-time reading and positioning of the detected human eye pictures are controlled by the external or webcam. Through side view, top view, and upward view, it is shown that the human eye real-time positioning system has good detection effect and robustness. In view of the poor classification effect of small data sets on the head state, the random forest algorithm is used to classify the pitching Angle and yaw Angle by HOG-LBP fusion feature and haar-like feature respectively, and then the obtained pitching Angle and deflection Angle are fused. The accuracy of direct classification is improved by 4.5% compared with the point’04 data set.