摘要目的:正确的胸外按压姿势（chest compression posture, CCP）是完成高质量心肺复苏（cardiopulmonary resuscitation, CPR）的重要基础，但目前关注CCP的研究十分有限。本研究设计新的CPR按压姿势自动分析程序，拟实现对CCP监测达到客观化、标准化和自动化的目的。方法:本研究共招募15人参与现场试验，其中专业组11人，非专业组4人。分别于正前方和45度侧面用ZED双摄像头同时记录按压视频数据，所有参与人员均在Smartman模拟人上进行连续的120次持续胸外按压操作。3位专家对CPR视频进行独立标注，智能算法提取人体骨骼点用于后续分析和模型开发。专业组和业余组两组率的比较采用卡方检验进行统计分析。结果:研究分析发现，腕部用力、手指未翘起、重心偏移、肘部弯曲是其中发生率最高的错误。通过专业组规范数据集共28 800组人体骨骼点坐标数据计算手臂角度合理范围为左臂169.24°~180.00°，右臂角度为168.49°~180°。相同的方法，得到重心角度合理范围为0.00°~18.46°。在此基础上，构建的基于双ZED的CPR按压姿势检测模型可以较准确的识别出CPR的按压姿势错误（准确率91.31%，敏感度80.16%，特异度93.53%）。结论:本研究创新性的提出对CPR按压姿势进行客观评价的方法，并且在此基础上构建了基于双ZED摄像头的CPR按压姿势检测模型，可以较准确的识别出CPR的按压姿势错误，以实现CPR培训质量控制可以更加的自动化和标准化。

abstractsObjective:Correct chest compression posture (CCP) is an important basis for high-quality cardiopulmonary resuscitation, but the research on CCP was still very limited. In this study, a new automatic analysis model was developed to achieve the purpose of objectification, standardization and automation of CCP monitoring.Methods:A total of 15 participants, including 11 professionals and 4 nonprofessionals, were recruited to participate in the field experiment. The video data were recorded simultaneously with zed cameras in the front and 45-degree sides. All participants performed 120 consecutive external chest compression operations on the Smartman CPR simulator. Three experts annotated the videos independently. An intelligent algorithm was used to extract human bone points for subsequent analysis and model development. The chi-square test was used to compare the rates of the professional and nonprofessional groups.Results:The results showed that problems with wrists, fingers, center of body weight and elbow bending had the highest incidence. Through 28 800 sets of standard human skeleton point coordinate data, we obtained a reasonable range of arm angles of 169.24°- 180.00° for the left arm and 168.49°-180.00° for the right arm. By the same method, the reasonable range of the center of gravity angle is 0.00°-18.46°. Based on these results, a new chest compression posture detection model based on a dual ZED camera was developed, which can accurately identify CCP errors (accuracy 91.31%; sensitivity 80.16%; specificity 93.53%).Conclusions:This study innovatively proposed an objective evaluation method for CCP. Moreover, a new chest compression posture detection model based on a dual ZED camera was developed, which can accurately identify CCP errors to achieve automation and standardization of quality control in CPR training.