面向消防演练应用的姿势识别和目标定位

Posture Recognition and Target Localization for Fire Drill Application

针对传统消防演练中指导员的主观判断存在局限性的问题,本文提出了一种应用于消防演练的姿势识别和目标定位的方法。首先,利用双目相机获取图像并提取人体骨架信息,选取合适的关节点特征构建人体姿势识别算法;其次,针对具有低纹理区域目标定位问题,提出一种融合点-线特征的立体匹配算法实现对目标的精确定位;进一步地,设计了基于距离自适应的线特征描述改进方法降低了立体匹配的计算复杂度。实验结果表明,该系统消防动作平均识别率为94%,对灭火器定位平均误差在1.3cm以内,可以为消防演练提供有效的辅助信息。

Aiming at the limitation of subjective judgment of instructors in traditional fire drill,this paper presents a method of posture recognition and target location for auxiliary monitoring system used in fire drill.Firstly,the binocular camera is used to obtain the scene image and the extract the human skeleton information.and the appropriate node features were selected to construct the human posture recognition algorithm.Secondly,aiming at the target localization problem with low texture area,a stereo matching algorithm integrating point-line features is proposed to achieve accurate target localization.Furthermore,the computational complexity of stereo matching is reduced by improving the line feature description based on distance adaptive.The experimental results show that the average action recognition rate of the system is 94%,and the average positioning error of the fire extinguisher is less than 1.3cm,which provides effective auxiliary information for fire drill.