基于头部姿态估计的多人低头跟踪检测系统

Multi-person heads down tracking detection based on head pose estimation

文中设计了在课堂场景下的多人头部动作检测系统,具有以下特点:①使用普通摄像头进行检测,不需要依赖深度信息和人脸关键点信息,可以做到较为精确的头部估计;②使用基于脸部特征的目标追踪算法,在稳定检测到人脸的情况下能够进行有效的目标追踪以便后续进行分析;③使用卡尔曼滤波对结果进行优化,得出的头部角度更为稳定。文中在三个不同分辨率的场景下进行实验,获得了较好的效果。

In this paper,we design a multi-person heads down tracking detection system under environment of classroom,which has the following characters:1.head pose could estimate accurately without depth information or key-points on faces by utilizing the normal camera.2.The tracking algorithm based on the facial features can track the target effectively while the face is detected,the information of which could be used for future analysis.3.The Kalman Filter is used to optimize the results which indicates that results produced by the head pose estimation algorithm are more stable.The experiments are conducted in three situations with different resolution and gains great performance out of it.