A method of emotion prediction based on group temperature monitoring

Abstract

The present invention proposes a method for predicting emotions based on group temperature monitoring, comprising: continuously acquiring infrared thermal images of the monitored group; monitoring the body temperature of each person in the monitored group based on each infrared thermal image, determining in real time whether each person's body temperature is abnormal, and acquiring each person's temperature monitoring information; determining whether each person's body temperature is abnormal within a preset period based on all temperature monitoring information acquired during the observation period, conducting a real-time emotional survey on those judged to be abnormal, and conducting emotional surveys on all people at regular intervals during the observation period; establishing an emotional prediction model based on the emotional survey results of all people and whether the corresponding body temperatures are abnormal; and inputting specific parameters of any person into the emotional prediction model to predict emotions. The present invention solves the problem of accuracy prediction in non-contact emotional prediction methods and can timely predict abnormal emotions of each person in the group.

Description

Technical Field

[0001] The present invention relates to a mood prediction method based on group temperature monitoring. Background Art

[0002] With the continuous development of economy and technology and the increasingly accelerated pace of life, people are facing increasing pressure in their lives, and more and more people are developing various psychological disorders or illnesses. Adolescents are a particularly high-risk group. Facing critical turning points in their studies, as well as the middle school entrance exams and college entrance exams, they are easily affected by hormones during development and are more susceptible to psychological disorders or illnesses. Furthermore, adolescents spend most of their time under high pressure to study, and parents are busy with work and household chores, making it difficult for them to promptly detect their children's emotional changes, which can further exacerbate their psychological disorders or illnesses. If we can predict the emotional changes of groups in specific situations or environments, many psychological disorders and illnesses can be addressed or avoided.

[0003] Currently, emotion prediction generally includes two types: contact-based and non-contact. Non-contact emotion prediction is more suitable for use within groups or in specific environments, and it also offers a degree of concealment. Non-contact methods primarily rely on recognizing facial micro-expressions for prediction, but these are difficult to detect, which can lead to increased errors in recognition and significantly impact the accuracy of detection results.

[0004] Research has shown that the hypothalamus is the central conductor of emotional and body temperature fluctuations. Emotional changes influence the hypothalamus, which in turn secretes hormones that influence temperature fluctuations. When people are excited, nervous, or angry, their body temperature rises; when they are depressed or overly sad, their temperature drops. Emotion and body temperature are closely linked. While body temperature monitoring technology is mature and accurate, there is currently no non-contact method for predicting emotions based on body temperature. Summary of the Invention

[0005] In order to solve the problem of accuracy prediction of non-contact emotion prediction methods in the prior art, the present invention proposes an emotion prediction method based on group temperature monitoring.

[0006] The technical solution of the present invention is achieved as follows:

[0007] A method for predicting emotions based on group temperature monitoring comprises: continuously acquiring infrared thermal images of a monitored group; monitoring the body temperature of each person in the monitored group based on each infrared thermal image; determining in real time whether the body temperature of each person is abnormal, and acquiring temperature monitoring information of each person; determining whether the body temperature of each person is abnormal within a preset period based on all temperature monitoring information acquired during an observation period; conducting a real-time emotion survey on persons determined to be abnormal, and conducting a regular emotion survey on all persons during the observation period; establishing an emotion prediction model based on the emotion survey results of all persons and whether the corresponding body temperatures are abnormal; and inputting specific parameters of a person into the emotion prediction model to predict emotions.

[0008] Preferably, the body temperature of each person in the monitored group is monitored based on each infrared thermal image, and the body temperature of each person is the temperature of several parts of each person, and the several parts can be arbitrarily selected from the following parts: forehead, cheek, neck, corner of the eye, arm, chest, foot and hand; the temperature monitoring information of each part is the average value of the pixel value of the position corresponding to the part on each infrared thermal image; the temperature of several parts of each person is the body temperature of that person.

[0009] Preferably, the body temperature of each person in the monitored group is monitored based on each infrared thermal image, and the specific method for judging in real time whether the body temperature of each person is abnormal is: calculating the temperature of each part of each person based on each infrared thermal image, calculating the average temperature value of the temperature of the same part of all people, and calculating the temperature difference between the temperature of each part of each person and the average temperature value of the part; giving a temperature difference threshold, if the absolute value of the temperature difference of any part of any person is greater than the temperature difference threshold, it is judged that the body temperature of this part of this person is abnormal.

[0010] Preferably, based on all temperature monitoring information obtained during the observation period, it is determined whether the body temperature of each person is abnormal within a preset period, specifically including: starting from the starting point of the observation, counting all periods in which abnormalities continuously occur in each part of each person within each preset period, giving a period threshold, and recording periods greater than the period threshold as abnormal periods; counting the total length of all abnormal periods in each part of each person within the preset period, and calculating the ratio of the total length to the length of the preset period, giving a ratio threshold, and judging the specific part of the specific person corresponding to the ratio greater than the ratio threshold as abnormal within the preset period, and issuing a warning if the judgment result of any part of any person is abnormal within each preset period.

[0011] Preferably, an emotional survey is conducted in real time on people whose results are judged to be abnormal, specifically including: establishing a questionnaire that can reflect emotional categories, where the emotional categories can be set to happy, depressed, angry, anxious and normal; and requiring people whose results are judged to be abnormal within a preset time period to fill out a copy of the questionnaire.

[0012] Preferably, an emotional survey is conducted on all persons at regular intervals during the observation period, specifically including: having each person fill out a questionnaire at regular intervals, recording the group activities participated in by all persons during the observation period and the time of participation.

[0013] Preferably, an emotion prediction model is established based on the results of the emotion survey of all persons and whether the corresponding human body temperature is abnormal, specifically including: identifying the emotion category of the corresponding person based on the questionnaire filled out by the person conducting the real-time emotion survey, identifying the emotion category of each person at each time period based on the questionnaire filled out by all persons in the timed emotion survey, combining whether the human body temperature of each person in the corresponding time period is abnormal, and the data of group activities in the corresponding time, using machine learning methods to train based on all the data to establish an emotion prediction model.

[0014] Preferably, the specific parameters of any person input into the emotion prediction model for predicting emotions refer to whether the temperature of each part of the person is abnormal and the group activities that the person participated in at the same time.

[0015] Preferably, the emotion prediction method further includes: after predicting the emotion, conducting an emotion survey on the person predicted to have abnormal emotion, adding whether the temperature of each part of the person is abnormal, the group activities participated in at the same time and the emotion survey results to the training data for establishing the emotion prediction model, and correcting the emotion prediction model.

[0016] Preferably, the infrared thermal image of the monitored group is continuously acquired, specifically comprising: continuously acquiring the initial infrared thermal image of the monitored group using an infrared camera; performing median filtering and denoising on the initial infrared thermal image to obtain a preliminary denoised image; setting an effective temperature threshold, removing pixels that are not within the effective temperature threshold in the preliminary denoised image to obtain effective pixels; setting a local area threshold, calculating the pixel difference between each effective pixel and the effective pixels in the local area around it, and calculating the average value of all pixel differences of each effective pixel, statistically calculating the distribution histogram of the average values ​​of all effective pixels, and calculating the average difference of the preliminary denoised image calculated based on the difference as the noise level, setting a Gaussian filter according to the noise level, and performing Gaussian filtering on the preliminary denoised image to obtain a secondary denoised image; setting a frame number threshold, the secondary denoised image before the first frame number threshold is the infrared thermal image of the monitored group, and in the continuously processed secondary denoised image, the noise level is updated every frame number threshold, and Gaussian filtering is re-performed according to the updated noise level to obtain the infrared thermal image of the monitored group.

[0017] The beneficial effects of the present invention are as follows: the present invention proposes an emotion prediction method based on group temperature monitoring, which monitors the body temperature of each person in the group based on each continuously collected infrared thermal image, and determines in real time whether each person's body temperature is abnormal; then, based on the temperature monitoring information of all people during the entire observation period, it determines whether the body temperature of any one person during a preset period is abnormal, rather than directly determining the temperature of the monitored person, thereby improving the accuracy of the judgment, combining the emotion survey during the observation period, increasing the accuracy of establishing the emotion model, and improving the accuracy of emotion prediction. The observation period is longer than the preset period, and setting a preset period within the observation period can more timely and accurately feedback the body temperature of the monitored person, thereby increasing the accuracy of judging the body temperature of each person during the observation period.

[0018] The emotion prediction method of the present invention can promptly predict emotional abnormalities of any individual in a group, allowing for timely intervention and prevention of, or even resolution of, psychological disorders or illnesses. The non-contact emotion prediction method is also more conducive to protecting the privacy of the individuals being monitored.

[0019] The emotion prediction method of the present invention predicts the emotions of any person in a group, conducts an emotion survey on the person predicted to have abnormal emotions, adds whether the temperature of each part of the person is abnormal, the group activities participated in at the same time, and the emotion survey results to the training data for establishing the emotion prediction model, corrects the emotion prediction model, and continuously improves the accuracy of the model prediction. DETAILED DESCRIPTION

[0020] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0021] Example 1: A method for predicting emotions based on group temperature monitoring, comprising the following steps: 1. using an infrared camera to continuously acquire initial infrared thermal images of the monitored group without contact; 2. performing denoising on the initial infrared thermal images to obtain infrared thermal images of the monitored group, wherein the infrared thermal images are represented by a pixel matrix, and the values ​​of the pixel matrix represent the thermal radiation intensity values ​​directly related to the temperature of the measured human body; the infrared thermal images are processed based on the infrared camera calibration data so that each pixel value of the infrared thermal image represents a temperature value; 3. monitoring the body temperature of eight parts of each person in the monitored group based on each infrared thermal image, i.e., each person in the monitored group is measured at a specific temperature. The temperature of an individual and his / her eight body parts are one-to-one corresponding. The specific method is to perform face recognition on the first frame of infrared thermal image to obtain each person's personal information, use the MobileNetSSD algorithm to detect the human body in each infrared thermal image in turn, use the SORT algorithm to track the detected human body, and use the openpose method to detect the bone points of each person, obtain the position of each person and the position of each person's forehead, cheek, neck, corner of the eye, arm, chest, foot and hand in the corresponding time. The eight parts selected by each person are the same; 4. Real-time judgment of whether the body temperature of each person's eight body parts is abnormal. Each person's eight body parts The human body temperature of any part of the body is the average value of the pixel values ​​corresponding to the position of the part on the infrared thermal image. On the same infrared thermal image, the average temperature value of the temperature of the same part of all people is calculated, and the temperature difference between the temperature of each part of each person and the average temperature value of the part is calculated. A temperature difference threshold is given. If the absolute value of the temperature difference of any part of any person is greater than the temperature difference threshold, the human body temperature of this part of the person is judged to be abnormal, and the temperature monitoring information of each person is obtained. The temperature monitoring information of each person includes the normal or abnormal information of the human body temperature of eight parts of each person; 5. Set the observation period and the preset period. The preset period is within the observation period. For example, the observation period is 5 days, and the preset period is 45 minutes. That is, starting from the starting point of the observation, every 45 minutes, it is judged whether the body temperature of each person is abnormal within these 45 minutes. Starting from the starting point of the observation, all the periods in which abnormalities appear continuously in each part of each person within 45 minutes are counted. A period threshold is given, and the period greater than the period threshold is recorded as an abnormal period. The total length of all abnormal periods in each part of each person is counted, and the ratio of the total length to 45 minutes is calculated. A ratio threshold is given, and the specific part of the specific person corresponding to the ratio greater than the ratio threshold is judged to be abnormal. In each preset period, if the judgment result of any part of any person is abnormal, a warning will be issued;6. Establish a questionnaire that can reflect emotion categories. Emotion categories can be set as happy, depressed, angry, anxious, and normal. For all people judged as abnormal in each preset time period in step 5, have them fill out a questionnaire in real time. Based on the completed questionnaire, identify which emotion category each person's emotion belongs to. For all people in the observation period, each person is regularly asked to fill out the questionnaire and identify which emotion category the corresponding person's emotion belongs to based on the completed questionnaire. Record the group activities and time of participation of all people in the observation period. 7. Based on the emotion category identified each time, whether the body temperature of each person in the corresponding time period is abnormal, and the data of group activities in the corresponding time, use the RNN machine learning method to train all data to establish an emotion prediction model. 8. The emotion prediction model then predicts the emotion of any person in the group based on whether the body temperature of the person is abnormal and the group activities the person participates in. 9. Based on the prediction results, conduct an emotion survey on the person predicted to have abnormal emotions. The abnormal temperature of each part of the person, the group activities participated in at the same time, and the emotion survey results are added to the training data for establishing the emotion prediction model to revise the emotion prediction model and improve the accuracy of the emotion prediction model. Predicting each person's emotions can help them take timely, targeted measures to prevent or resolve psychological disorders or illnesses, and even help treat them if they are discovered in a timely manner.

[0022] In step 1 of Example 1, initial infrared thermal images are continuously acquired. Each initial infrared thermal image is acquired at a corresponding time point. Therefore, the real-time determination of whether the human body temperature at the eight locations in each infrared thermal image is normal or abnormal has a corresponding time point. A group in the present invention refers to two or more people.

[0023] In the second step of the first embodiment, each of the continuously acquired initial infrared thermal images is subjected to denoising in sequence. Specifically, the initial infrared thermal image is subjected to median filtering denoising to obtain a preliminary denoised image; an effective temperature threshold is set, and pixels that are not within the effective temperature threshold are removed from the preliminary denoised image to obtain effective pixels; a local area threshold is set, and the pixel difference between each effective pixel and the effective pixels in the surrounding local area is calculated, and the average of all pixel differences of each effective pixel is calculated, and a distribution histogram of the average values ​​of all effective pixels is calculated. The average difference of the preliminary denoised image calculated based on the difference is recorded as the noise level, and the preliminary denoised image is Gaussian filtered according to the noise level to obtain a secondary denoised image; a frame number threshold is set, and the secondary denoised image before the initial frame number threshold is the infrared thermal image of the monitored group. In the continuously processed secondary denoised images, the noise level is updated every frame number threshold, and Gaussian filtering is re-performed based on the updated noise level to obtain an infrared thermal image of the monitored group. This denoising effect can better preserve the boundaries of the infrared thermal image while removing noise.

[0024] In step 3 of Example 1, facial recognition is performed on the first frame of infrared thermal image to obtain personal information of each person in the group, so as to ensure that each person in the group matches the monitored temperature, the corresponding human body temperature abnormality and emotional category, and the group activities participated in, so as to avoid mismatching between the mobile personnel appearing during the observation period and the monitored temperature, whether the temperature is abnormal, the emotional category, and the group activities participated in. If no one leaves the monitored group during the observation period, human body tracking and skeleton point detection are used on each infrared thermal image to obtain the human body position and skeleton point position of each person at the corresponding time point, and temperature monitoring is performed on each person to obtain whether the human body temperature is abnormal, the emotional category, and the group activities participated in by each person during the observation period. There is no need to perform face recognition on every infrared thermal image during the observation period; if someone leaves and comes back in the middle, face recognition is performed again, and the human body temperature of this person is monitored again. The human body temperature, whether the human body temperature is abnormal, the emotional category, and the group activities participated in before the person leaves are connected with the human body temperature, whether the human body temperature is abnormal, the emotional category, and the group activities participated in after the person comes back at the corresponding time points to obtain whether the human body temperature is abnormal, the emotional category at the corresponding time, and the group activities participated in by the person during the entire observation period; if a new person joins, face recognition is first performed on the person to obtain personal information, and the temperature, whether the human body temperature is abnormal, the emotional category, and the group activities participated in are monitored from that period to the remaining observation period. Face recognition can be performed using any method that can achieve this function. Face recognition can also be replaced by any other way or method that can accurately identify each person in the group to ensure that each person's body temperature abnormality data and the emotional category at the corresponding time and the group activities participated in can be correctly matched during the observation period.

[0025] In the first embodiment, during the continuous observation period, people are allowed to leave or join, and people are allowed to leave and then come back or not. During the continuous observation period, it is necessary to accurately record whether the body temperature of each person is abnormal, the emotional category at the corresponding time, and the collective group activities participated in.

[0026] The MobileNetSSD algorithm, SORT algorithm, and OpenPose method in step 3 of Example 1 are all existing algorithms. The selected body parts for each person can be other body parts in addition to the eight selected parts, and can be selected based on actual needs. Human body detection in infrared thermal images can utilize existing HOG (Histogram of Oriented Grids) and SSD (Histogram of Oriented Grids) algorithms, YOLO (You Only Look Once: Unified, Real-Time Object Detection), R-CNN (Region-Convolutional Neural Networks), or other target detection algorithms. Detected human bodies can be tracked using the MIL (Multiple Instance Learning) algorithm, KCF (Kernelized Correlation Filter) algorithm, TLD (Tracking-Learning-Detection) algorithm, MedianFlow algorithm, GoTrun algorithm, MOSSE (Minimum Output Sum of Squared Error Filter) algorithm, or other target tracking algorithms. Human skeletal point detection can also utilize any other algorithm that can achieve the corresponding function.

[0027] In step 4 of Example 1, the average temperature of each part of all people in the group and the temperature difference of each part of each person are calculated, and the difference between the temperature differences is used to determine whether each part of each person is abnormal. This is mainly because the human body temperature is affected by various factors. The accuracy of directly determining abnormalities based on each person's absolute body temperature is low. The absolute temperature of the human body is calculated using infrared thermal images captured by an infrared camera, with an absolute error accuracy of 1 to 2 degrees. However, the relative temperature of different people in the group is calculated with a relative error accuracy of up to 0.03 degrees. Therefore, the present invention determines whether the temperature of an individual is abnormal by statistically analyzing the difference in body temperature between individuals and groups, thereby improving the accuracy of detection. A group refers to two or more people. Generally speaking, the larger the number of people in the group, the more accurate the measurement of individual body temperature abnormalities.

[0028] In step 5 of Example 1, if the judgment result of any part of any person is abnormal within each preset time period, a warning will be issued. The warning method is diverse, and can be a voice prompt or an indicator light prompt, which is to remind the monitoring personnel to conduct an emotional survey in real time on all people whose judgment results are abnormal within each preset time period.

[0029] The questionnaire in step 6 of Example 1 can be a questionnaire in various forms, and the questions and answers in the questionnaire are designed in a variety of ways. The questionnaire can adopt international standards, such as the PHQ-9 emotional self-assessment form, or a customized questionnaire can be used to reflect the emotional categories of the respondents in the group. Of course, any method that can help obtain the emotional categories of the observed persons can also be used here to conduct emotional surveys; emotional categories can also include other forms such as sadness, sadness, fear, etc., and different emotional categories can be selected according to the different groups being monitored; group activities are recorded according to actual conditions, such as indoor cultural classes, indoor self-study, outdoor activities, indoor meals, indoor work, indoor meetings, etc. By recording group activity events, the macro or overall information of the group can be obtained, individual differences can be highlighted, and abnormal people in the group can be obtained to improve the accuracy of the emotional model establishment.

[0030] In step 6 of Example 1, a real-time emotional survey is conducted on individuals identified as abnormal. This real-time survey refers to the immediate conduct of an emotional survey on individuals identified as abnormal within a preset time period. A scheduled emotional survey is conducted on all individuals within the observation period, regardless of whether or not there are any abnormal individuals. The scheduled survey can be set to occur at regular intervals or at a specific time, depending on the actual monitored population and the environment in which the population resides. Individuals identified as abnormal are those identified as abnormal in any part of the body.

[0031] In step 4 of the first embodiment, j is used to represent the number of any of the eight parts of a person, j = 1, 2, 3, ..., 8. There are P people in the current infrared thermal image, i is used to represent the number of each person, i = 1, 2, 3, ..., P. The temperature of each part of each person is expressed as T(i, j). The average temperature value of the temperature of the same part of all people is The difference between the temperature of any part of each person and the average temperature of that part is Given a temperature difference threshold T, if the absolute value of the temperature difference of any part j of any person i is greater than the temperature difference threshold T, it is judged that this part of the person is abnormal at the corresponding time point when the infrared thermal image is obtained.

[0032] The RNN machine learning algorithm in step 7 of Example 1 is an existing algorithm, and any algorithm that can achieve the corresponding function can also be used for processing, such as other ML related analysis algorithms.

[0033] The emotion prediction in step 8 of Example 1 is generally performed for all people in the group with abnormal body temperatures, but it can also be performed for each individual. Step 8 then predicts the emotion of any person in the group based on whether the person's body temperature is abnormal and the group activities the person has participated in. The abnormality of any person's body temperature and the group activities the person has participated in can refer to the person's abnormal body temperature and the group activities they have participated in on a particular occasion, or they can refer to the person's abnormal body temperature and the group activities they have participated in over a period of time. If only the person's abnormal body temperature and the group activities they have participated in on a particular occasion are considered, the prediction accuracy will be lower because a single temperature situation is subject to randomness. If the person's abnormal body temperature and the group activities they have participated in are considered over a period of time, the prediction accuracy will be higher.

[0034] Step 9 in the first implementation can further modify the emotion prediction model, increasing its accuracy. This accuracy increases with the addition of more data. Even after removing step 9, the emotion prediction model can still predict the emotions of the monitored population. Abnormal emotions in step 9 can be defined as depression, anger, or anxiety. Abnormal emotions can also be defined based on the specific situation. Generally, negative emotions are considered abnormal.

[0035] In the first embodiment, the temperatures of several body parts of each person in a group are monitored simultaneously. The temperature difference of any body part of each person in the group is compared with the average temperature value of the same body part of the rest of the people, and then compared with the temperature threshold. Finally, it is determined whether the body temperature of this part of the person is abnormal, thereby obtaining the temperature monitoring information of each person. Instead of directly detecting the body temperature of each person, which is affected by various factors, the present invention improves the stability of obtaining the temperature monitoring information of each person.

[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A mood prediction method based on group temperature monitoring, characterized in that: include: Continuously acquire infrared thermal images of the monitored group; monitor the body temperature of each person in the monitored group based on each infrared thermal image; Determine in real time whether each person's body temperature is abnormal and obtain each person's temperature monitoring information; Based on all the temperature monitoring information obtained during the observation period, determine whether each person's body temperature is abnormal within the preset period; Conduct emotional surveys on people whose results are judged to be abnormal in real time, and conduct emotional surveys on all people at regular intervals during the observation period; establish an emotional prediction model based on the emotional survey results of all people and whether the corresponding body temperature is abnormal; input the specific parameters of any person into the emotional prediction model to predict emotions, and monitor the body temperature of each person in the monitored group based on each infrared thermal image. The body temperature of each person is the temperature of several parts of each person, and the several parts can be arbitrarily selected from the following parts: forehead, cheek, neck, corner of the eye, arm, chest, foot and hand; the temperature monitoring information of each part is the average value of the pixel value of the corresponding position of the part on each infrared thermal image; the temperature of several parts of each person is the body temperature of that person, and the body temperature of each person in the monitored group is monitored based on each infrared thermal image, and the body temperature of each person is judged in real time The specific method of determining whether it is abnormal is as follows: based on each infrared thermal image, the temperature of each part of each person is calculated, the average temperature value of the temperature of the same part of all people is calculated, and the temperature difference between the temperature of each part of each person and the average temperature value of the part is calculated; a temperature difference threshold is given, and if the absolute value of the temperature difference of any part of any person is greater than the temperature difference threshold, the body temperature of this part of the person is judged to be abnormal; the specific parameters of any person are input into the emotion prediction model. The specific parameters in the predicted emotion refer to whether the temperature of each part of the person is abnormal and the collective activities that the person participated in at the same time; after predicting the emotion, an emotion survey is conducted on the person whose emotion is predicted to be abnormal, and whether the temperature of each part of the person is abnormal, the collective activities participated in at the same time, and the emotion survey results are added to the training data for establishing the emotion prediction model, and the emotion prediction model is corrected.

2. The emotion prediction method according to claim 1, characterized in that Based on all the temperature monitoring information obtained during the observation period, it is determined whether the body temperature of each person is abnormal within the preset period, specifically including: starting from the starting point of the observation, counting all the periods in which abnormalities continuously occur in each part of each person within each preset period, giving a period threshold, and recording the period greater than the period threshold as an abnormal period; counting the total length of all abnormal periods in each part of each person within the preset period, and calculating the ratio of the total length to the length of the preset period, giving a ratio threshold, and judging the specific part of the specific person corresponding to the ratio greater than the ratio threshold as abnormal within the preset period; and issuing a warning if the judgment result of any part of any person is abnormal within each preset period.

3. The emotion prediction method according to claim 2, characterized in that Conducting real-time emotional surveys on people whose results are judged to be abnormal, specifically including: establishing a questionnaire that can reflect emotional categories, where the emotional categories can be set to happy, depressed, angry, anxious and normal; and requiring people whose results are judged to be abnormal within a preset time period to fill out a copy of the questionnaire.

4. The emotion prediction method according to claim 3, characterized in that Conduct emotional surveys on all persons at regular intervals during the observation period, specifically including: having each person fill out a questionnaire at regular intervals, and recording the group activities that all persons participated in during the observation period and the time of participation.

5. The emotion prediction method according to claim 4, characterized in that An emotion prediction model is established based on the results of the emotion survey of all people and whether the corresponding body temperature is abnormal. Specifically, it includes: identifying the emotion category of the corresponding person based on the questionnaire filled out by the people taking the real-time emotion survey, identifying the emotion category of each person at each time based on the questionnaire filled out by all people in the scheduled emotion survey, combining the data on whether the body temperature of each person is abnormal in the corresponding time period and the group activities in the corresponding time, and using machine learning methods to train based on all the data to establish an emotion prediction model.

6. The emotion prediction method according to claim 1, characterized in that Continuously acquiring infrared thermal images of the monitored group, specifically comprising: continuously acquiring initial infrared thermal images of the monitored group using an infrared camera; performing median filtering and denoising on the initial infrared thermal image to obtain a preliminary denoised image; setting an effective temperature threshold, removing pixels that are not within the effective temperature threshold in the preliminary denoised image to obtain effective pixels; setting a local area threshold, calculating the pixel difference between each effective pixel and the effective pixels in its surrounding local area, and calculating the average value of all pixel differences of each effective pixel, statistically calculating a distribution histogram of the average values ​​of all effective pixels, and calculating the average difference of the preliminary denoised image calculated based on the difference as the noise level, setting a Gaussian filter according to the noise level, and performing Gaussian filtering on the preliminary denoised image to obtain a secondary denoised image; setting a frame number threshold, and the secondary denoised image before the first frame number threshold is the infrared thermal image of the monitored group, and in the continuously processed secondary denoised image, updating the noise level every frame number threshold, and re-performing Gaussian filtering based on the updated noise level to obtain the infrared thermal image of the monitored group.