Research On Pose Recognition Algorithm Of Human Thermal Discomfort For Green Intelligent Buildings

20% of global energy consumption comes from building energy consumption,and HVAC systems account for 50% of building energy consumption.Due to the inability to accurately sense the thermal sensation of indoor personnel in real time,supercooling or superheating supply brings a lot of energy waste.Non-contact human thermal discomfort detection based on visual perception is one of the methods to alleviate this problem,and has become a research hotspot at home and abroad in recent years.Due to individual differences,the accuracy of existing visual detection methods needs to be improved urgently.In this thesis,from the perspective of human pose detection,the thermal discomfort of human body is studied.The main work is as follows:(1)Build a human thermal discomfort pose dataset(PORT: Pose Dataset of Human Thermal Discomfort).Based on the Fanger theory,through the subjective data of a large number of questionnaires,it is confirmed that there is a correlation between human body pose and human thermal discomfort.On this basis,17 poses related to thermal discomfort were defined,and 18 subjects were invited to collect relevant video data and subjective thermal sensations in a temperature and humidity controlled environment.This thesis preprocesses the data,and then constructs a dataset of thermal discomfort pose,including 320,000 thermal discomfort pose pictures,and the corresponding thermal sensation data.(2)Thermal discomfort gesture recognition algorithm based on skeleton key points.This thesis firstly studies the thermal discomfort pose of the human body from the perspective of skeleton key points,and proposes a pose recognition framework based on skeleton key points.The algorithm framework includes two parts: 1)a skeleton key point acquisition module and 2)a gesture recognition module.The difference in the position of the person in the image will affect the result of skeletal key point coordinate estimation.Aiming at this problem,this thesis firstly eliminates the influence of individual location differences through normalization.The human thermal discomfort pose is a continuous video frame.In this thesis,the correlation between frames is used to construct a deep learning network.In this thesis,the method of horizontal comparison(between individuals)is used to train and test between different individuals to verify the impact of individual differences on the algorithm.The experimental results show that the proposed algorithm can effectively identify the individual pose information,so as to realize the detection of human thermal discomfort.(3)End-to-end thermal discomfort pose detection algorithm.The human thermal discomfort pose detection algorithm in the skeleton key point mode is beneficial to edge deployment,but its detection accuracy needs to be improved due to the loss of a large amount of human information.In response to this problem,this thesis explores from an end-to-end perspective,and proposes an endto-end thermal discomfort gesture recognition algorithm.On the basis of massive data preprocessing,this thesis sorts out the correspondence between thermal discomfort sequence frames and thermal sensations.In order to build a balance between detection accuracy and detection speed,this thesis selects two modes of 8 frames and 16 frames in continuous poses for algorithm training and verification.The algorithm can effectively extract spatial features between image sequences,and fuse low-level and high-level features to capture the associated information of the entire action sequence.Aiming at individual differences,the algorithm has been tested horizontally,and the results show the effectiveness and robustness of the algorithm.