Method and apparatus for classifying postures using random forest

Abstract

The present invention relates to a method and apparatus for classifying postures using a random forest, and, according to an aspect of the present invention, the method for classifying postures using a random forest comprises the steps of: calculating joint coordinates corresponding to a central position of a face and joint coordinates corresponding to both shoulders from a front evaluation image in which the upper body of a seated person is captured from the front, to thereby detect objects corresponding to the face and shoulders; calculating a face area ratio that is a ratio of a face area of a face region detected in the front evaluation image to a preset face area for a correct posture; calculating a plurality of angles formed by the center of the face and both shoulders, by using the calculated joint coordinates; generating evaluation data including the joint coordinates corresponding to the central position of the face, the face area ratio, and the plurality of angles, all of which are calculated from the front evaluation image; and inputting the evaluation data into a pre-trained random forest model to classify a posture.

Description

Method and apparatus for posture classification using random forest

[0001] The present invention relates to a technology for classifying human posture through image analysis.

[0002] As computer use has become widespread and usage time has increased, the number of people suffering from spinal disorders, such as neck and lower back problems, due to poor sitting posture is on the rise. Although awareness of the importance of proper posture is growing, people unconsciously adopt incorrect postures when sitting for long periods. Furthermore, even if individuals are aware that they are adopting an incorrect posture, there is a problem in that it is difficult to objectively identify specifically what kind of incorrect posture they are using.

[0003] Accordingly, various technologies for measuring the posture of seated individuals are being introduced. First, existing technologies that classify posture based on measurements taken by equipment worn on the subject's body can cause discomfort during measurement and present difficulties in analyzing posture during daily activities.

[0004] Existing technology that measures posture by taking a side view of the spine alignment by shooting from the side of a seated person has the problem that a relatively large space is required for shooting because the camera must be spaced a certain distance away to capture the entire side view of the upper body.

[0005] Furthermore, existing technology that classifies posture based on video filmed from the front of a seated person has limitations in that it can only determine whether the posture is incorrect or if there is forward head posture, but cannot classify posture in detail.

[0006] The objective of the present invention is to solve the above problem by providing a posture classification method and apparatus using a random forest capable of detailed posture classification from a plurality of postures, including a straight posture, a forward head posture, a posture leaning backward, a posture leaning to the left, and a posture leaning to the right, from a frontal image captured by a camera installed on a monitor.

[0007] The objectives of the present invention are not limited to those mentioned above, and other unmentioned objectives will be clearly understood from the description below.

[0008] A posture classification method using a random forest according to one aspect of the present invention for achieving the aforementioned purpose relates to a method performed by an electronic device, comprising: a step of detecting objects corresponding to a face and shoulders by calculating joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders using a pre-set skeletalization algorithm on an evaluation frontal image of a seated person taken from the front; a step of calculating a face area ratio, which is the ratio of the face area according to the number of pixels belonging to the face region detected in the evaluation frontal image to the face area pre-set for a correct posture; a step of calculating a plurality of angles formed by the center of the face and both shoulders using the joint coordinates calculated in the step of detecting objects; a step of generating evaluation data including joint coordinates corresponding to the center position of the face calculated for the evaluation frontal image, the face area ratio, and the plurality of angles; and a step of classifying a posture into one of pre-set postures by inputting the evaluation data into a pre-trained random forest model.

[0009] A posture classification device using a random forest according to another aspect of the present invention includes a memory for storing preset commands and a processor for executing the commands stored in the memory. As the processor executes the commands stored in the memory, it detects objects corresponding to the face and shoulders by calculating joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders using a preset skeletalization algorithm on an evaluation frontal image of a seated person taken from the front. It also calculates a face area ratio, which is the ratio of the face area to the face area for a correct posture, based on the number of pixels belonging to the face area detected in the evaluation frontal image. Furthermore, it calculates a plurality of angles formed by the center of the face and both shoulders using the joint coordinates calculated for the center position of the face and both shoulders. It generates evaluation data including the joint coordinates corresponding to the center position of the face, the face area ratio, and the plurality of angles calculated for the evaluation frontal image. Finally, it inputs the evaluation data into a pre-trained random forest model to classify the posture into one of the preset postures.

[0010] According to the present invention, there is an effect that allows for posture classification even in a relatively narrow space through a frontal image captured by a camera installed on a monitor.

[0011] According to the present invention, it has the effect of enabling detailed classification of postures in a plurality of postures, including a correct posture, a forward head posture, a posture leaning backward, a posture leaning to the left, and a posture leaning to the right.

[0012] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description in the claims.

[0013] FIG. 1 is a flowchart of a posture classification method using a random forest according to an embodiment of the present invention.

[0014] FIG. 2 is a block diagram of a posture classification device using a random forest according to another embodiment of the present invention.

[0015] FIG. 3a shows an example of a frontal image corresponding to a correct posture among the postures classified according to an embodiment of the present invention.

[0016] FIG. 3b shows an example of a frontal image corresponding to a turtle neck posture among the postures classified according to an embodiment of the present invention.

[0017] FIG. 3c shows an example of a frontal image corresponding to a left-leaning posture among the postures classified according to an embodiment of the present invention.

[0018] FIG. 3d is an example of a frontal image corresponding to a posture leaning to the right among the postures classified according to an embodiment of the present invention.

[0019] FIG. 3e shows an example of a frontal image corresponding to a backward leaning posture among the postures classified according to an embodiment of the present invention.

[0020] A posture classification method using a random forest according to one aspect of the present invention for achieving the aforementioned purpose relates to a method performed by an electronic device, comprising: a step of detecting objects corresponding to a face and shoulders by calculating joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders using a pre-set skeletalization algorithm on an evaluation frontal image of a seated person taken from the front; a step of calculating a face area ratio, which is the ratio of the face area according to the number of pixels belonging to the face region detected in the evaluation frontal image to the face area pre-set for a correct posture; a step of calculating a plurality of angles formed by the center of the face and both shoulders using the joint coordinates calculated in the step of detecting objects; a step of generating evaluation data including joint coordinates corresponding to the center position of the face calculated for the evaluation frontal image, the face area ratio, and the plurality of angles; and a step of classifying a posture into one of pre-set postures by inputting the evaluation data into a pre-trained random forest model.

[0021] A posture classification device using a random forest according to another aspect of the present invention includes a memory for storing preset commands and a processor for executing the commands stored in the memory. As the processor executes the commands stored in the memory, it detects objects corresponding to the face and shoulders by calculating joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders using a preset skeletalization algorithm on an evaluation frontal image of a seated person taken from the front. It also calculates a face area ratio, which is the ratio of the face area to the face area for a correct posture, based on the number of pixels belonging to the face area detected in the evaluation frontal image. Furthermore, it calculates a plurality of angles formed by the center of the face and both shoulders using the joint coordinates calculated for the center position of the face and both shoulders. It generates evaluation data including the joint coordinates corresponding to the center position of the face, the face area ratio, and the plurality of angles calculated for the evaluation frontal image. Finally, it inputs the evaluation data into a pre-trained random forest model to classify the posture into one of the preset postures.

[0022] The advantages and features of the present invention, and the methods for achieving them, will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the claims. Meanwhile, the terms used in this specification are for describing the embodiments and are not intended to limit the present invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text.

[0023] The present invention relates to a technology for classifying postures through video analysis of a seated person.

[0024] The present invention is characterized by the technical feature that posture classification is possible from a frontal image obtained from a camera that captures the front of a seated person, such as a camera installed on a monitor, thereby enabling posture analysis even in a relatively narrow space, and detailed posture classification is possible in five postures including upright posture, forward head posture, leaning back posture, leaning to the left posture, and leaning to the right posture.

[0025] These technical features can be achieved by a configuration that calculates six explanatory variables including the coordinates of the center of the face position (x-coordinate value, y-coordinate value) in a frontal image, the ratio of the face area to the correct posture, and the three angles formed by the face and the shoulders, calculates weights based on the importance of each explanatory variable using a random forest model, and classifies the posture by applying the values ​​reflecting the weights of the explanatory variables to the random forest model.

[0026] Referring to FIG. 2, a posture classification device (100) using a random forest according to another embodiment of the present invention may be configured to include a communication interface (110), a memory (120), and a processor (130).

[0027] The communication interface (110) communicates with an external electronic device and may receive input from the external electronic device or output data processing results to the user through the external electronic device.

[0028] The memory (120) includes a computer-readable recording medium and can store an operating system and preset instructions.

[0029] Each step of the posture classification method using a random forest according to one embodiment of the present invention may be implemented as computer instructions that perform designated functions, which are loaded into the processor or memory of an electronic device capable of data processing (e.g., a general-purpose computer, a special-purpose computer, a portable notebook computer, a network computer).

[0030] For the convenience of the following explanation, drawing symbols for functionally consistent content will be consistent, and explanations will be avoided.

[0031] Referring to FIG. 1, the processor (130) can receive learning frontal images of a person taking a posture corresponding to each of the preset postures from a user through a communication interface (110) as it executes commands stored in memory (120) (S101).

[0032] Here, the pre-set postures may include a straight posture, a forward head posture, a posture leaning backward, a posture leaning to the left, and a posture leaning to the right.

[0033] The processor (130) can detect objects corresponding to the face and shoulders by using a skeleton algorithm pre-set for each of the frontal images for training to calculate joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders (S103).

[0034] Here, the skeletalization algorithm can utilize the media pipe developed by Google.

[0035] The processor (130) applies a frontal training image to a preset skeletalization algorithm to calculate joint coordinates corresponding to the center position of the face, joint coordinates corresponding to the left shoulder, and joint coordinates corresponding to the right shoulder in each frontal training image, and can detect objects corresponding to the face and shoulders for each frontal training image (S103).

[0036] The processor (130) receives a training image in which a rectangular box area corresponding to a person's face is labeled in a frontal image of a person, and applies each training frontal image to a pre-trained object recognition algorithm to detect a rectangular face area corresponding to a face in an image input from the outside, thereby detecting a face area in each training frontal image and calculating a face area based on the number of pixels belonging to the detected face area, and can calculate a face area ratio which is the ratio of the calculated face area to a pre-set standard face area for a correct posture (S105).

[0037] Here, the object recognition algorithm can be the SSD (Single Shot MultiBox Detector) algorithm.

[0038] The preset reference face area for a correct posture can be pre-set to correspond to the number of pixels belonging to the rectangular area corresponding to the face in the training frontal image corresponding to the correct posture.

[0039] In other words, the pre-set standard face area for a correct posture refers to the face area detected in the frontal image when in a correct posture.

[0040] The processor (130) can calculate the face area ratio for each training frontal image by dividing the detected face area in each training frontal image by a reference face area set for a correct posture (S105).

[0041] Afterwards, the processor (130) can calculate multiple angles formed by the center of the face and both shoulders using the joint coordinates calculated in step S103, which detects objects in the frontal image for training (S107).

[0042] The processor (130) can calculate three angles formed by the face center and both shoulders by calculating a first angle based on the angle between two vectors calculated by subtracting the joint coordinates corresponding to the face center and the joint coordinates corresponding to the left shoulder, respectively, from the joint coordinates corresponding to the right shoulder and the joint coordinates corresponding to the left shoulder, calculating a second angle based on the inverse tangent of the ratio of the difference in the horizontal and vertical axes of the joint coordinates corresponding to the face center and the joint coordinates corresponding to the right shoulder, and calculating a third angle based on the inverse tangent of the ratio of the difference in the horizontal and vertical axes of the joint coordinates corresponding to the face center and the joint coordinates corresponding to the left shoulder (S107).

[0043] The processor (130) has joint coordinates corresponding to the center position of the face according to the following mathematical formula ( , ), joint coordinates corresponding to the left shoulder( , ) and joint coordinates corresponding to the right shoulder( , From ) the first angle ( ), second angle( ), third angle( ) can be produced.

[0044]

[0045]

[0046]

[0047] A processor (130) can generate a dataset by creating a plurality of rows corresponding to training frontal images corresponding to preset postures and columns corresponding to joint coordinates, face area ratios, three angles, and postures corresponding to the center position of the face among the preset postures, and by substituting the joint coordinates, face area ratios, and three angles corresponding to the center position of the face calculated from each training frontal image and the preset classification values ​​for the posture corresponding to each training frontal image among the preset postures as matrix values, setting the joint coordinates, face area ratios, and three angles corresponding to the center position of the face as explanatory variables and setting the preset classification values ​​for the posture corresponding to each training frontal image as response variables (S109).

[0048] The processor (130) can train a random forest model with the generated dataset (S111).

[0049] At this time, the processor (130) can classify the data corresponding to some rows in the generated dataset as training data and the data corresponding to the remaining rows as evaluation data, train a random forest model with the training data, and input the evaluation data into the trained random forest model to classify the pose and check the performance of the random forest model.

[0050] Referring to Figures 3a to 3e, it can be seen that the center of the face is similar in the upright posture, forward head posture, and backward leaning posture, but there are differences in the face area and the angle formed by the center of the face and both shoulders.

[0051] Also, it can be observed that while the facial surface area is similar in the left-leaning posture and the right-leaning posture, the center of the face is different.

[0052] Based on this, the present invention generates a dataset including data comprising coordinates corresponding to the center position of the face, the ratio of the face area to the correct posture, and three angles formed by the face and the shoulder, and a classification value corresponding to the posture as a supervised learning value for each data, thereby enabling the construction of a random forest model capable of classifying five detailed postures including correct posture, forward head posture, leaning backward, leaning to the left, and leaning to the right through the dataset.

[0053] And, the processor (130) can calculate weights for each explanatory variable based on the variable importance for each explanatory variable derived through the learned random forest model (S113).

[0054] The processor (130) can calculate weights for each explanatory variable based on the MeanDecreaseGini (MDI) index value, which indicates the degree to which each explanatory variable reduces the impurity of the random forest among the variable importance values ​​calculated from the learned random forest model, such that explanatory variables with relatively large MDI index values ​​have larger weight values ​​than explanatory variables with relatively small MDI index values ​​(S113).

[0055] The processor (130) can classify the pose using data classified as evaluation data in the dataset, but in actual implementation, a trained Random Forest model is deployed and used to classify the pose by receiving a frontal image as input. Therefore, the following description assumes a situation in which the pose is classified by receiving a frontal image as input for evaluation.

[0056] The steps described below for detecting the face and shoulders (S117), calculating the face area ratio (S119), and calculating multiple angles formed by the face and shoulders (S121) are identical in execution method to the steps described above for detecting the face and shoulders (S103), calculating the face area ratio (S105), and calculating multiple angles formed by the face and shoulders (S107), differing only in the applied data being distinguished as frontal images for training and frontal images for evaluation; therefore, specific details will not be explained redundantly.

[0057] The processor (130) receives an evaluation frontal image of a person's upper body through a communication interface (110) (S115), calculates joint coordinates corresponding to the center position of the face, joint coordinates corresponding to the left shoulder, and joint coordinates corresponding to the right shoulder in the evaluation frontal image using a skeletalization algorithm pre-set in the evaluation frontal image, and can detect objects corresponding to the face and shoulders (S117).

[0058] The processor (130) receives a training image in which a rectangular box area corresponding to a person's face is labeled in an evaluation frontal image, applies a pre-trained object recognition algorithm to detect a rectangular face area corresponding to a face in an image input from the outside, detects a face area, calculates a face area based on the number of pixels belonging to the detected face area, and divides the face area calculated in the evaluation frontal image by a reference face area pre-set for a correct posture to calculate a face area ratio which is the ratio of the reference face area pre-set for a correct posture to the face area calculated in the evaluation frontal image (S119).

[0059] Subsequently, the processor (130) can calculate three angles formed by the face center and both shoulders by using the joint coordinates calculated in step S117, which detects an object in a frontal image for evaluation, to calculate a first angle based on the angle between two vectors calculated by subtracting the joint coordinates corresponding to the face center and the joint coordinates corresponding to the left shoulder, respectively, and calculating a second angle based on the inverse tangent of the ratio of the difference in the horizontal and vertical axes of the joint coordinates corresponding to the face center and the joint coordinates corresponding to the right shoulder, respectively, and calculating a third angle based on the inverse tangent of the ratio of the difference in the horizontal and vertical axes of the joint coordinates corresponding to the face center and the joint coordinates corresponding to the left shoulder, respectively (S121).

[0060] The processor (130) can generate evaluation data including joint coordinates corresponding to the center position of the face calculated for the frontal image for evaluation, a face area ratio, and a plurality of angles (S123).

[0061] The processor (130) can generate evaluation data consisting of values ​​with weights reflected by applying the weights for each explanatory variable calculated in step S113, which calculates weights for each explanatory variable, to the joint coordinates corresponding to the center position of the face, the face area ratio, and the three angles calculated for the frontal image for evaluation, respectively (S123).

[0062] The processor (130) can generate evaluation data consisting of weighted values ​​by multiplying the weights of the joint coordinates corresponding to the center position of the face calculated for the frontal image for evaluation, the face area ratio, and the explanatory variables corresponding to each of the three angles (S123).

[0063] The processor (130) can classify the posture into one of five postures, including a straight posture, a turtle neck posture, a posture leaning backward, a posture leaning to the left, and a posture leaning to the right, by inputting evaluation data into a pre-trained random forest model (S125).

[0064] According to the present invention, posture classification is possible using only a frontal image capturing the front of the user's upper body, such as the top of a monitor, without capturing the entire sitting posture.

[0065] Accordingly, it has the advantage of being able to classify the posture of a seated person even in relatively narrow spaces without being restricted by the size of the space.

[0066] In addition, according to the present invention, by constructing a random forest model with a dataset including the coordinates of the face center position calculated from a frontal image, the ratio of the face area to the correct posture, and three angles formed by the face and the shoulder, it is possible to classify five detailed postures including correct posture, forward head posture, leaning backward, leaning to the left, and leaning to the right.

[0067] In addition, by reflecting weights based on variable importance derived from a random forest model trained on the coordinates of the center of the face, the ratio of the face area to the correct posture, and the three angles formed by the face and shoulders, and then applying them to the random forest model, high accuracy can be guaranteed for five detailed classifications.

[0068] Experiments were conducted to evaluate the performance of a posture classification method using a random forest according to an embodiment of the present invention.

[0069] Twenty adult men and women, consisting of 12 men and 8 women, were selected as test subjects. A camera was installed on the computer monitor screens of the subjects, and each subject was instructed to assume a straight posture, a forward head posture, a posture leaning backward, a posture leaning to the left, and a posture leaning to the right. The subjects' frontal views were filmed, and 100 experimental frontal images were acquired as data.

[0070] The experiment was performed in a Windows 10 Pro, Intel® Core™ i7 10700k, CPU@3.80GHz, and Python 3.6 environment.

[0071] The x-coordinates and y-coordinates of the center of the face derived from the experimental frontal image, the ratio of the face area to the correct posture, and the three angles formed by the face and shoulders were calculated, and weights were assigned to each explanatory variable according to the variable importance of the six explanatory variables, including the x-coordinates and y-coordinates of the center of the face derived from the random forest model, the ratio of the face area to the correct posture, and the three angles formed by the face and shoulders, as shown in the table below.

[0072] Variable Name Weight Value Face Area Ratio 0.3 X-coordinate of face center 0.2 Face-Left Shoulder Angle (2nd Angle) 0.15 Face-Right Shoulder Angle (3rd Angle) 0.15 Face-Both Shoulder Angle (1st Angle) 0.15 Y-coordinate of face center 0.05

[0073] In addition, a Random Forest classifier, a Support Vector Machine classifier, a Logistic Regression, and a K-Nearest Neighbors model were constructed according to the present invention, respectively, and experiments were conducted using data composed of values ​​reflecting weights according to an embodiment of the present invention. The number of decision trees in the Random Forest was set to 150, and since the number of data acquired for the experiment was small, cross-validation was performed with 10 to verify the constructed models, and the experimental results are as shown in the table below.

[0074] Classifier Accuracy 1 2 3 4 5 6 7 8 9 10 Average Random Forest (The present invention) 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 9 1.0 1.0 9 9 9 Support Vector Machine 0.9 0.9 9 9 1.0 0.8 0.9 9 9 1.0 0.9 9 9 9 Logistic Regression 0.8 0.9 9 9 8 0.9 9 9 9 7 1.0 0.9 9 9 9 8 K-Nearest Neighbors 1.0 0.7 0.8 9 9 6 0.9 9 7 0.8 9 9 8 8 8 9 8 1 0

[0075] Referring to Table 2, it was confirmed that when classified by the random forest model according to the present invention, it achieved an average accuracy of 0.98, which is higher than the average accuracy of other existing classifiers, which is in the 0.8 range. In other words, according to the present invention, it was confirmed that it is possible to accurately classify five detailed postures even in situations where data for model construction is relatively insufficient.

[0076] A person skilled in the art to which the present invention pertains will understand that the present invention may be implemented in other specific forms without altering its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the claims and their equivalents should be interpreted as being included within the scope of the present invention.

[0077] The present invention relates to a technology for classifying human posture through image analysis.

Claims

1. A method performed by an electronic device, A step of detecting objects corresponding to the face and shoulders by using a pre-set skeletalization algorithm on an evaluation frontal image of a seated person's upper body taken from the front to calculate joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders; A step of calculating a face area ratio, which is the ratio of the face area based on the number of pixels belonging to the face region detected in the above-mentioned frontal image for evaluation to the face area for a correct posture; A step of calculating a plurality of angles formed by the face center and both shoulders using joint coordinates calculated in the step of detecting the object above; A step of generating evaluation data including joint coordinates, face area ratios, and multiple angles corresponding to the center position of the face calculated for the above-mentioned frontal image for evaluation; A method for classifying a pose using a random forest, comprising the step of inputting the evaluation data into a pre-trained random forest model to classify the pose into one of the preset poses.

2. In Paragraph 1, Prior to the step of receiving the above-mentioned frontal image for evaluation, A step of training a random forest model with a dataset generated by setting joint coordinates corresponding to the center position of the face calculated from each frontal image corresponding to the preset postures, face area ratio, and multiple angles as explanatory variables, and setting a preset classification value for the posture corresponding to each frontal image among the preset postures as a response variable; A method for classifying details using a random forest, further comprising the step of calculating weights for each explanatory variable based on variable importance derived through a trained random forest model.

3. In Paragraph 2, The step of generating the above evaluation data is In the step of calculating the weights for each explanatory variable above, the calculated weights for each explanatory variable are applied to the joint coordinates, face area ratio, and multiple angles corresponding to the center position of the face calculated for the frontal image for evaluation, respectively, to generate evaluation data consisting of values ​​with applied weights. Detail classification method using random forest.

4. In Paragraph 1, Pre-set postures Includes good posture, forward head posture, leaning back, leaning to the left, and leaning to the right. Detail classification method using random forest.

5. Memory for storing preset commands; and A processor that executes instructions stored in the above memory; comprising, The above processor By executing the commands stored in the above memory, Objects corresponding to the face and shoulders are detected by calculating joint coordinates corresponding to the center position of the face and joint coordinates corresponding to both shoulders using a pre-configured skeletalization algorithm on an evaluation frontal image of the upper body of a seated person taken from the front, and Calculate the face area ratio, which is the ratio of the face area based on the number of pixels belonging to the face region detected in the above-mentioned frontal image for evaluation to the face area for a correct posture, and Using the joint coordinates calculated for the center of the face and both shoulders, multiple angles formed by the center of the face and both shoulders are calculated, and Evaluation data is generated that includes joint coordinates, face area ratios, and multiple angles corresponding to the center position of the face calculated for the above-mentioned frontal image for evaluation, and Inputting the above evaluation data into a pre-trained Random Forest model to classify the posture into one of the preset postures. A posture classification device using random forest.

6. In Paragraph 5, The above processor A Random Forest model is trained using a dataset generated by setting joint coordinates corresponding to the center position of the face, face area ratios, and multiple angles calculated from each frontal image corresponding to a preset posture as explanatory variables, and setting a preset classification value for the posture corresponding to each frontal image among the preset postures as the response variable, and Calculating weights for each explanatory variable based on variable importance derived from the trained random forest model. A posture classification device using random forest.

7. In Paragraph 6, The above processor The method of generating evaluation data composed of weighted values ​​by applying the calculated weights for each explanatory variable to the joint coordinates, face area ratio, and multiple angles corresponding to the center position of the face calculated for the aforementioned frontal image for evaluation. A posture classification device using random forest.

8. In Paragraph 5, Pre-set postures Includes good posture, forward head posture, leaning back, leaning to the left, and leaning to the right. A posture classification device using random forest.

Images