基于三维坐姿的压力数据自动标注方法

针对当前传统坐姿标注方法，存在操作繁琐、人工标注效率低等问题，提出了一种基于三维坐姿的压力数据自动标注方法。该方法基于双指针时间戳匹配实现双目视觉数据和压力数据同步实时采集；采取归一化、自适应中值滤波处理压力数据，去除压力数据量纲影响和尖峰噪声；利用二维姿态估计与匹配优化、坐标变换和多点三角测量处理视觉数据，提取6个三维人体关键点信息；构造基于坐姿特征的骨架图和基于邻接节点间空间三维投影角度特征，建立一个基于三维坐姿的标注信息生成模型；利用标注信息标注压力数据，构建带标注的坐姿压力数据集。实际应用中，单个样本数据对视觉-压力端采集平均时间差仅为21 ms,生成标注准确率达98.98%,自动标注平均耗时0.199 s,标注速度较人工标注提升13.3倍。

Automatic annotation method for pressure data based on three-dimensional sitting posture

To address the problems of cumbersome operation and low efficiency of manual annotation in current traditional sitting postureannotation methods, this article proposes an automatic annotation method for pressure data based on three-dimensional sitting posture.Real time synchronous collection of binocular visual data and pressure data based on dual pointer timestamp matching. The normalizationand adaptive median filtering are utilized to process pressure data, and remove the dimensional influence and peak noise of pressuredata. By using 2D pose estimation and matching optimization, coordinate transformation, and multi-point triangulation to process visualdata, 6 key point information of the 3D human body are extracted. A skeleton image based on sitting posture features and a 3D projectionangle feature between adjacent nodes are constructed, and a 3D sitting posture based annotation information generation model isformulated. The annotation information is utilized to label the pressure data and create an annotated sitting pressure dataset. In practicalapplications, the average time difference between real-time synchronous collection of single sample data is only 21 ms, and the accuracyof label generation is 98. 98% . The average time for automatic labeling is 0. 199 s, and the labeling speed is 13. 3 times faster thanmanual labeling.