基于红外测温的皮肤温度识别与热舒适预测OA

To address the difficulty of automatically acquiring whole-body segment temperatures and the infrared measurement bias caused by clothing coverage in individualized thermal comfort control,a segment-level temperature recognition method based on infrared temperature matrices and the human pose estimation model(YOLOv8-Pose)is proposed in this study.The proposed method achieves keypoint recognition through hierarchical transfer learning,and realizes the segmentation of 19 human body segments,quantification of clothing coverage,and temperature extraction based on geometric construction and clustering-based discrimination.Experimental results demonstrate that the multi-model cascaded keypoint recognition approach improves keypoint detection performance by 86.0%over the single-pass full-body keypoint recognition model under medium-to-high localization accuracy metrics.The extracted segment-level infrared temperatures show good agreement with contact-based measurements,with a root mean square error of 0.45℃.Furthermore,machine learning-based thermal sensation prediction using segment-level temperatures achieves an accuracy of 91.7%,which is significantly higher than that of the predicted mean vote model.