Rapid identification method for abnormal behaviors of autism spectrum disorder children

Abstract

The invention belongs to the technical field of behavior recognition, and provides a rapid recognition method for abnormal behaviors of children with autism spectrum disorder, and the method comprises the steps: obtaining the position information and face orientation of all children in each frame of image of a free activity video of the children; based on the position information, analyzing the position movement and the activity range of each child in different frames of images to obtain the activity tendency of each child; performing clustering analysis on the position information of all children in each frame of image, and adjusting clustering parameters in combination with activity tendencies to obtain an optimal comparison range of each child; analyzing active and passive interaction conditions between each child and other children in the corresponding optimal comparison range based on the face orientation to obtain an interaction tendency and an interaction weight of each child; and obtaining the abnormal behavior degree of each child in combination with the activity tendency, the interaction tendency and the interaction weight, and completing the recognition of the abnormal behavior of the child. The method provided by the invention can quickly and accurately identify children with abnormal behaviors based on the adjusted optimal comparison range.

Description

Technical Field

[0001] The present invention relates to the technical field of behavior recognition, and in particular to a method for quickly recognizing abnormal behaviors of children with autism spectrum disorders. Background Art

[0002] Autism spectrum disorder (ASD) is a neurodevelopmental disorder. Children with ASD may show social withdrawal, delayed language development, strong interest in specific things, and repetitive behaviors. Abnormal behavior identification of children with ASD can significantly improve the quality of life and social skills of children with ASD by monitoring children's social interactions, communication skills, and repetitive behaviors, observing children's eye contact, language expression, play style, and response to environmental changes, and combining standardized scales and intelligent recognition technology to screen for potential autistic characteristics.

[0003] When using posture estimation and behavior recognition technology to identify abnormal behaviors of children with autism spectrum disorders, it is usually necessary to collect videos of children's behaviors in their living environment for analysis, such as at home or in kindergarten. Videos of children's behaviors in kindergartens can well show how children interact and communicate with others and how they play, and are suitable as analysis samples. When analyzing the interactions between children, if the appropriate comparison object for analyzing the interactions is not selected, the analysis of the children's interactions will be inaccurate. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for quickly identifying abnormal behaviors of children with autism spectrum disorders.

[0005] According to the present invention, a method for quickly identifying abnormal behaviors of children with autism spectrum disorders is provided, the method comprising: Obtain the position information and facial orientation of all children in each frame of the video of children's free activities; Based on the position information, analyzing the position movement and activity range of each child in different frame images to obtain the activity tendency of each child; Performing cluster analysis on the position information of all children in each frame of the image, and adjusting the clustering parameters in combination with the activity tendency to obtain the best contrast range for each child; Based on the facial orientation, analyzing active and passive interactions between each child and other children within the corresponding optimal comparison range to obtain an interaction tendency and an interaction weight of each child; Combining the activity tendency, the interaction tendency and the interaction weight, obtaining a behavior abnormality degree of each child; Children with behavioral abnormalities are identified based on the degree of behavioral abnormality.

[0006] In some embodiments of the present invention, obtaining the position information and facial orientation of all children in each frame of a video of children's free activities includes: Get videos of children doing free activities; The YOLO model is used to identify the positions and facial orientations of all children in each frame of the free activity video, and the outer contour of each child is marked in the form of a rectangular bounding box to obtain the position information and facial orientation of all children in each frame.

[0007] In some embodiments of the present invention, based on the position information, the position movement and activity range of each child in different frame images are analyzed to obtain the activity tendency of each child, including: Based on the position information, analyzing the position change of each child in two adjacent frames of images to obtain a position movement parameter; Based on the position information, analyzing the maximum activity range of each child in all frame images to obtain activity range parameters; The activity tendency of each child is obtained according to the position movement parameter and the activity range parameter.

[0008] In some embodiments of the present invention, the maximum activity range is the minimum circumscribed circle radius of all position information of the child in all frame images.

[0009] In some embodiments of the present invention, cluster analysis is performed on the position information of all children in each frame of the image, and the clustering parameters are adjusted in combination with the activity tendency to obtain the optimal comparison range for each child, including: Analyze the Euclidean distance between the location information of any two children to obtain the initial similarity between any two children; Identify active children based on the described activity tendencies; In combination with the activity tendency, the initial similarity between the active child and other children is adjusted to obtain a modified similarity of the active child, and then a modified similarity matrix is ​​obtained; Based on the modified similarity matrix, cluster analysis is performed on the position information of all children in each frame of image to obtain a clustering result for each frame of image; The clustering results of all frame images are analyzed to obtain the best contrast range for each child.

[0010] In some embodiments of the present invention, analyzing the clustering results of all frame images to obtain the best contrast range for each child includes: taking the clustering results of all frame images that are consistent with the first The children belong to the same cluster the most times and are closest to each other children, as the first The optimal contrast range for each child.

[0011] In some embodiments of the present invention, based on the facial orientation, the active and passive interactions between each child and other children within the corresponding optimal contrast range are analyzed to obtain the interaction tendency and interaction weight of each child, including: establishing a facing vector according to the facial orientation; Recording the line connecting each child and other children in the corresponding optimal comparison range as a comparison baseline; Based on the orientation vector and the comparison baseline, the active and passive interactions between each child and other children within the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child.

[0012] In some embodiments of the present invention, based on the orientation vector and the comparison baseline, the active and passive interactions between each child and other children in the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child, including: Analyzing the angle relationship between the orientation vector of each child and the comparison baseline to obtain a first position relationship; Analyze the angle relationship between the orientation vectors of other children within the optimal comparison range corresponding to each child and the comparison baseline to obtain a second position relationship; Combining the first position relationship and the second position relationship to obtain the interaction tendency of each child; According to the second position relationship, the interaction weight of each child is obtained.

[0013] In some embodiments of the present invention, the activity tendency, the interaction tendency and the interaction weight are combined to obtain the abnormal behavior degree of each child, including: The interaction weight is used as the weight of the interaction tendency, and 1 minus the interaction weight is used as the weight of the activity tendency to obtain the behavior abnormality degree of each child.

[0014] In some embodiments of the present invention, identifying children with abnormal behavior according to the degree of abnormal behavior includes: Preset abnormal thresholds; When the degree of abnormal behavior of the child is greater than or equal to the abnormal threshold, identifying that the child has abnormal behavior; The autism risk of the children was assessed using the Child Autism Screening Scale and the Child Autism Assessment Scale.

[0015] It can be seen from the above embodiments that the method for quickly identifying abnormal behaviors of children with autism spectrum disorders provided by the embodiments of the present invention has the following beneficial effects: The present invention acquires the position information and facial orientation of all children in each frame image of a video of children's free activities, so as to analyze the behavior performance of children in an active state of autonomous action; based on the position information, the position movement and activity range of each child in different frame images are analyzed to obtain the activity tendency of each child. The activity tendency of the child, as an important indicator for judging abnormal behavior of the child, can reflect the possibility that the child has autism spectrum disorder to a certain extent; cluster analysis is performed on the position information of all children in each frame image, and the clustering parameters are adjusted in combination with the activity tendency to obtain the best contrast range of each child. The acquisition of the best contrast range provides a true and reliable contrast range for the subsequent analysis of the children's interaction situation, thereby improving the accuracy of the final recognition result; based on the facial orientation, the active and passive interaction between each child and other children in the corresponding best contrast range is analyzed to obtain the interaction tendency and interaction weight of each child. The interaction tendency of the child, as another important indicator for judging abnormal behavior of the child, can also reflect the possibility that the child has autism spectrum disorder to a certain extent; therefore, the behavior abnormality degree of each child is obtained in combination with the activity tendency, the interaction tendency and the interaction weight; and children with abnormal behavior are identified according to the behavior abnormality degree. The present invention provides a method that can quickly and accurately identify children with abnormal behaviors based on the adjusted optimal comparison range and combined with the children's activity tendency and interaction tendency.

[0016] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A schematic diagram of the basic flow of a method for rapidly identifying abnormal behaviors in children with autism spectrum disorders provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the method for quickly identifying abnormal behaviors of children with autism spectrum disorders proposed by the present invention, its specific implementation method, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. Terms such as "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a circuit structure, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such article or device. In the absence of further restrictions, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the article or device including the element.

[0021] A method for quickly identifying abnormal behaviors of children with autism spectrum disorders provided by this embodiment will be described in detail below in conjunction with the accompanying drawings.

[0022] See also Figure 1 , which shows the basic process of a method for quickly identifying abnormal behaviors of children with autism spectrum disorders provided by an embodiment of the present invention.

[0023] like Figure 1 As shown, an embodiment of the present invention provides a method for quickly identifying abnormal behaviors of children with autism spectrum disorders, which specifically includes the following steps: S100: Acquire the position information and facial orientation of all children in each frame of the video of children's free activities.

[0024] First, we obtain videos of children’s free activities in kindergarten. Specifically, we use the surveillance cameras in the kindergarten classroom to obtain 30 minutes of free activity videos (time when children are independently active, playing and communicating without teacher guidance) at a fixed position. The video specification is 30 frames per second, which is used to analyze children’s behavioral performance in the state of autonomous action.

[0025] Then, the existing pre-trained YOLO model for person recognition is used to identify the position and facial orientation of all children in each frame of the free activity video, and the outer contour of each child is marked in the form of a rectangular bounding box. The center of the rectangular bounding box represents the position of each child, and the position information and facial orientation of all children in each frame are obtained.

[0026] S200: Based on the position information, the position movement and activity range of each child in different frame images are analyzed to obtain the activity tendency of each child.

[0027] Children with autism spectrum disorders tend to play alone during free activities and are unable to actively participate in interactive group activities. By analyzing the positions and movement trajectories of all children in the children's free activity videos, we can obtain the activity tendencies of each child. The following situations may occur in children during free activities: (1) Multiple children play together. Such children have more interactions with their group and will cluster with other children in position, but their overall movement trajectory may not change much; (2) Children play in multiple groups or alone. These children are usually lively, with large changes in movements and movement trajectories, and may interact more with others; (3) Children play alone, such as introverted children or children with autism spectrum disorders, who are not good at actively integrating into the group. They may play alone in the same place for a long time and hardly interact with others.

[0028] First, the overall movement of the child is analyzed. If the child's position movement is small and the range of activity is small throughout the free activity video, it reflects that the child has a low activity tendency and is more likely to have an autism spectrum disorder.

[0029] Based on the above analysis, this embodiment analyzes the position movement and activity range of each child in different frame images based on the position information to obtain the activity tendency of each child. Further, based on the position information, the position change of each child in two adjacent frame images is analyzed, that is, the position change of each child in each frame image and its adjacent previous frame image is analyzed, and the position change is recorded as the displacement of each child in each frame image, and the displacement of the first frame is recorded as 0; the cumulative sum of the displacement of the children in all frame images is calculated to obtain the position movement parameter. Based on the position information, the maximum activity range of each child in all frame images is analyzed at the same time to obtain the activity range parameter, wherein the maximum activity range is the minimum circumscribed circle radius of all position information of the child in all frame images. Then, according to the position movement parameter and the activity range parameter, the activity tendency of each child is obtained.

[0030] Build Activity trends of children The calculation formula is: In the formula, Indicates The activity tendency of each child; Indicates The minimum circumscribed circle radius of all positions of a child in all frame images; Indicates the total number of frames of the free-motion video; Indicates Children in Displacement of frame images; represents the linear normalization function.

[0031] Indicates the first The cumulative sum of the displacement distances of the children represents the position movement parameter; Reflect the The size of the activity range of a child in all frame images represents the maximum activity range; Indicates The greater the value, the greater the change in the action trajectory of the child. The more activities a child engages in and the larger the scope of his activities, the greater his tendency to be active.

[0032] Similarly, obtain the activity tendencies of all children.

[0033] S300: Performing cluster analysis on the position information of all children in each frame of the image, and adjusting the clustering parameters in combination with the activity tendency to obtain the best comparison range for each child.

[0034] Activity tendency reflects the overall positional activities of each child. There are also other situations, such as multiple children playing together or introverted children having lower activity tendencies. The overall activity tendency alone cannot fully and accurately show whether a child has a tendency to be lonely. Therefore, further analysis of each child's local movements and interactions is needed.

[0035] Before analyzing the children's interactions, we first cluster the distances between all children, so as to analyze whether there are any abnormalities in the children's actions and their interactions with others by comparing the associations between children in the same cluster. Children who are closer to each other are more likely to play together, so we cluster the children's location information. If the comparison range is not selected properly, it may lead to incorrect judgment of the interaction object when judging the children's interactions in the future, which may lead to errors in the judgment of the children's interactions.

[0036] Based on the above analysis, this embodiment performs cluster analysis on the position information of all children in each frame image, and adjusts the clustering parameters in combination with the activity tendency to obtain the optimal comparison range for each child. Further, it includes analyzing the Euclidean distance between the position information of any two children to obtain the initial similarity between any two children; identifying active children according to the activity tendency; adjusting the initial similarity between active children and other children in combination with the activity tendency to obtain the corrected similarity of active children, and then obtaining the corrected similarity matrix; based on the corrected similarity matrix, cluster analysis is performed on the position information of all children in each frame image to obtain the clustering result of each frame image; analyzing the clustering results of all frame images to obtain the optimal comparison range for each child.

[0037] Taking any frame image as an example, the specific explanation is as follows: use the AP clustering algorithm to cluster all children. Due to the existence of children with strong activity tendencies, these children may be far away from the location where other children gather in some frame images during their activities, causing the location of their corresponding points to show certain outlier characteristics. However, in fact, it is because of their strong activity tendencies that they are outliers rather than being unsociable. The existence of such children may lead to inaccurate clustering results. Therefore, in order to obtain more accurate clustering results, the similarity values ​​of different individuals in clustering are adjusted using the children's activity tendencies to make the clustering results more accurate, thereby obtaining a more reference comparison range for each child. The specific steps are as follows: Get similarity and reference. One child corresponds to one point, and the center of the bounding box still represents the position of each child. The opposite of the Euclidean distance between any two points is used as the distance between any two points. and Point The initial similarity between , and The similarity between every two points and the reference degree of each point to itself constitutes a similarity matrix. For any point , take the minimum value in the similarity matrix as the point Reference , similarity matrix and reference degree are well-known technologies and will not be described here. In order to make children with high activity tendency be better classified into other clusters during clustering rather than showing outlier characteristics (too far from the cluster center), the initial similarity of children with high activity tendency is adjusted. Therefore, an active threshold is preset, which can be 0.7, and all children with activity tendency greater than or equal to the activity threshold are screened out and recorded as active children. For example, take the first The initial similarity between the active child and other children is adjusted: In the formula, Indicates The corrected similarity between the active child and one other child; Indicates The initial similarity between the active child and one other child; Indicates that except Any child other than the active child; Indicates activity tendencies of active children; Indicates the active threshold.

[0038] Indicates The activity tendency of active children exceeds the activity threshold. The value range is [0.7,1], and the activity tendency The larger the The smaller the value of is a negative number ( The location of the active child is The opposite number of the Euclidean distance between the position points of the children), the greater the activity tendency, The larger the adjusted similarity value is.

[0039] Similarly, the corrected similarity between all active children and all other children is obtained. The similarity of all children to all other children is formed into a corrected similarity matrix , where the data corresponding to active children in the similarity matrix are the modified similarities, and the data corresponding to other inactive children are the initial similarities.

[0040] Initialize the attraction matrix And the attribution matrix The attraction matrix is ​​usually And the attribution matrix Initialized to zero matrix. Attraction matrix Any element of Indicate point Point The attractiveness of When a child is the center of a cluster The tendency of a child in the cluster with which he is the center; the belonging degree matrix Any element of Indicate point As the The availability of cluster centers for each point.

[0041] Iteratively update the attraction matrix And the attribution matrix By modifying the similarity matrix And the attribution matrix Attraction Matrix Iterative update, and then through the attraction matrix Attribution Matrix The iterative update method is a prior art and will not be described in detail here.

[0042] Setting up the decision matrix ,in , ,when When it is greater than 0, it is considered that the point is the cluster center until the decision matrix The result is considered to have converged if the number of times the cluster center obtained in the continuous constant exceeds the preset control parameter. The preset control parameter here is usually 15, that is, 15 consecutive decision matrices The cluster centers in do not change. The decision matrix After convergence, all points are assigned to the nearest cluster center to achieve clustering.

[0043] Similarly, the clustering results of all frame images are obtained.

[0044] After obtaining the clustering results of each frame image, analyze the clustering results of all frame images to obtain the best contrast range for each child. Specifically, among the clustering results of all frame images, take the one that matches the first The children belong to the same cluster the most times and are closest to each other children, The value of can be 5, as the The best comparison range for the child, and the other children within the best comparison range are The children were matched with the control children.

[0045] This step uses the AP clustering algorithm to cluster all children, and adjusts the similarity value in the cluster based on each child's activity tendency to obtain the best comparison range for each child for subsequent analysis of children's interactions. Taking any child as an example, the other children included in the best comparison range of the child are the objects that are most likely to interact with the child. By analyzing the interaction between these objects and the child, it can be further determined whether the child has abnormal behavioral characteristics of autism spectrum disorder.

[0046] S400: Based on the facial orientation, the active and passive interactions between each child and other children within the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child.

[0047] For introverted children and children with autism spectrum disorders, although they are less active in interaction, introverted children behave normally in interaction with others, while children with autism spectrum disorders can hardly complete normal interaction with others. For example, after being interacted with by others, they cannot make normal eye contact or verbal responses, and they almost never make eye contact with others. By analyzing the interaction between each child and other children in their best comparison range, the interaction tendency of each child can be obtained.

[0048] Based on the above analysis, this embodiment analyzes the active and passive interactions between each child and other children in the corresponding optimal comparison range based on facial orientation, and obtains the interaction tendency and interaction weight of each child. It further includes: establishing a facing vector according to facial orientation, the starting point of the facing vector is the position of the child, the size of the facing vector is a unit vector, and the facing vector is only used to indicate the facing direction; the line connecting each child and any other child in the corresponding optimal comparison range is recorded as a comparison baseline, and there is a comparison baseline between any two children; based on the facing vector and the comparison baseline, the active and passive interactions between each child and other children in the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child. Furthermore, based on the orientation vector and the comparison baseline, the active and passive interactions between each child and other children within the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child, including: analyzing the angular relationship between the orientation vector of each child and the comparison baseline to obtain a first position relationship; analyzing the angular relationship between the orientation vectors of other children within the corresponding optimal comparison range of each child and the comparison baseline to obtain a second position relationship; combining the first position relationship and the second position relationship to obtain the interaction tendency of each child; and obtaining the interaction weight of each child based on the second position relationship.

[0049] Build The calculation formula for the interaction tendency of a child is: In the formula, Indicates the interaction tendencies of the children; Indicates the total number of frames of the free-motion video; Indicates the optimal comparison range size, here ; Indicates Frame The orientation vector of the first child is The angle between the comparison baselines, The comparison baseline is The child and his corresponding 1. The connection between the children in the comparison group; Indicates Frame The corresponding child The orientation vectors of the first and second children were compared The angle between the comparison baselines; represents the cosine function; represents the maximum value function; represents the linear normalization function.

[0050] Indicates The cosine value of the angle between the orientation vector of the first child and the comparison baseline represents the first position relationship, reflecting the Children The active interaction of the first comparison child is in the range of [-1,1]. The larger the value, the more The more consistent the orientation vector of the child is with the direction of the baseline, the The more likely a child is to The comparison children actively interacted; Indicates The corresponding child The orientation vectors of the first and second children were compared The cosine value of the angle between the two comparison baselines indicates the second position relationship, reflecting the The corresponding child The comparison children The interaction of the first child, The passive interaction of the first child. When the values ​​of these two are larger, Children and their The more interaction there is between the two contrasting children, the greater the interaction tendency. Winning The maximum value of , that is, taking the result of the contrast children with the most likely interaction in the best contrast range; It indicates the cumulative sum of interactions in all frame images. The larger the value, the The greater the interaction tendency of the child.

[0051] Similarly, the interaction tendency of all children is obtained. The greater the interaction tendency of a child, the more likely the child is to have effective interaction with other children.

[0052] According to the second position relationship, the interaction weight of each child is obtained, that is, the first The passive interaction of each child in all frames is analyzed, and the interaction weight is constructed. The interaction weight of each child The calculation formula is: In the formula, Indicates The interaction weights of the children; Indicates the total number of frames of the free-motion video; Indicates the optimal comparison range size, here ; Indicates Frame The corresponding child The orientation vectors of the first and second children were compared The angle between the comparison baselines; represents the cosine function; represents the maximum value function. The average of the passive interactions of the children in all frames.

[0053] S500: Combining the activity tendency, interaction tendency and interaction weight, the behavior abnormality degree of each child is obtained.

[0054] The activity tendency, interaction tendency and interaction weight are combined to obtain the behavior abnormality of each child, including: using the interaction weight as the weight of the interaction tendency, and using 1 minus the interaction weight as the weight of the activity tendency to obtain the behavior abnormality of each child.

[0055] Build Children with abnormal behavior The calculation method is: In the formula, Indicates The degree of behavioral abnormality of the child; Indicates The interaction weights of the children; Indicates The activity tendency of each child; Indicates the interaction tendencies of the children; Represents a natural constant An exponential function with base ; Represents the inverse normalization function, x is the independent variable.

[0056] Indicates that when the interaction weight is small, Individual children rarely interact with other children, and activity tendency is used as the main reference for calculating behavioral abnormality; When the interaction weight is significant, the interaction tendency is used as the main reference for calculating the abnormality of the behavior.

[0057] Similarly, obtain the degree of behavioral abnormality of all children.

[0058] S600: Identify children with behavioral abnormalities according to the degree of behavioral abnormality.

[0059] According to the degree of behavioral abnormality, children with abnormal behavior are identified. Specifically, a preset abnormal threshold value can be 0.7; when the degree of abnormal behavior of a child is greater than or equal to the abnormal threshold value, the child is identified as having abnormal behavior; through guardians and teachers, standardized assessment scales such as M-CHAT (Child Autism Screening Scale) and CARS (Child Autism Assessment Scale) are used to assess the risk of autism in children. According to the assessment results, timely intervention is carried out on the abnormal behavior of children with autism spectrum disorders and the help of professional doctors is sought.

[0060] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for quickly identifying abnormal behaviors of children with autism spectrum disorders, characterized in that: The method comprises: Obtain the position information and facial orientation of all children in each frame of the video of children's free activities; Based on the position information, analyzing the position movement and activity range of each child in different frame images to obtain the activity tendency of each child; Performing cluster analysis on the position information of all children in each frame of the image, and adjusting the clustering parameters in combination with the activity tendency to obtain the best contrast range for each child; Based on the facial orientation, analyzing active and passive interactions between each child and other children within the corresponding optimal comparison range to obtain an interaction tendency and an interaction weight of each child; Combining the activity tendency, the interaction tendency and the interaction weight, obtaining a behavior abnormality degree of each child; Children with behavioral abnormalities are identified based on the degree of behavioral abnormality.

2. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 1, characterized in that: Get the location information and facial orientation of all children in each frame of the video of children's free activities, including: Get videos of children doing free activities; The YOLO model is used to identify the positions and facial orientations of all children in each frame of the free activity video, and the outer contour of each child is marked in the form of a rectangular bounding box to obtain the position information and facial orientation of all children in each frame.

3. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 1 or 2, characterized in that: Based on the position information, the position movement and activity range of each child in different frame images are analyzed to obtain the activity tendency of each child, including: Based on the position information, analyzing the position change of each child in two adjacent frames of images to obtain a position movement parameter; Based on the position information, analyzing the maximum activity range of each child in all frame images to obtain activity range parameters; The activity tendency of each child is obtained according to the position movement parameter and the activity range parameter.

4. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 3, characterized in that: The maximum activity range is the minimum circumscribed circle radius of all position information of the child in all frame images.

5. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 1, characterized in that: Perform cluster analysis on the position information of all children in each frame of the image, and adjust the clustering parameters based on the activity tendency to obtain the optimal comparison range for each child, including: Analyze the Euclidean distance between the location information of any two children to obtain the initial similarity between any two children; Identify active children based on the described activity tendencies; In combination with the activity tendency, the initial similarity between the active child and other children is adjusted to obtain a modified similarity of the active child, and then a modified similarity matrix is ​​obtained; Based on the modified similarity matrix, cluster analysis is performed on the position information of all children in each frame of image to obtain a clustering result for each frame of image; The clustering results of all frame images are analyzed to obtain the best contrast range for each child.

6. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 5, characterized in that: Analyze the clustering results of all frame images to obtain the best contrast range for each child, including: taking the clustering results of all frame images that are consistent with the first The children belong to the same cluster the most times and are closest to each other children, as the first The optimal contrast range for each child.

7. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 1, characterized in that: Based on the facial orientation, active and passive interactions between each child and other children within the corresponding optimal contrast range are analyzed to obtain the interaction tendency and interaction weight of each child, including: establishing a facing vector according to the facial orientation; Recording the line connecting each child and other children in the corresponding optimal comparison range as a comparison baseline; Based on the orientation vector and the comparison baseline, the active and passive interactions between each child and other children within the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child.

8. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 7, characterized in that: Based on the orientation vector and the comparison baseline, the active and passive interactions between each child and other children in the corresponding optimal comparison range are analyzed to obtain the interaction tendency and interaction weight of each child, including: Analyzing the angle relationship between the orientation vector of each child and the comparison baseline to obtain a first position relationship; Analyze the angle relationship between the orientation vectors of other children within the optimal comparison range corresponding to each child and the comparison baseline to obtain a second position relationship; Combining the first position relationship and the second position relationship to obtain the interaction tendency of each child; According to the second position relationship, the interaction weight of each child is obtained.

9. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 1, characterized in that: Combining the activity tendency, the interaction tendency and the interaction weight, the behavior abnormality degree of each child is obtained, including: The interaction weight is used as the weight of the interaction tendency, and 1 minus the interaction weight is used as the weight of the activity tendency to obtain the behavior abnormality degree of each child.

10. The method for rapidly identifying abnormal behaviors of children with autism spectrum disorders according to claim 1 or 9, characterized in that: Based on the degree of behavioral abnormality, children with behavioral abnormalities are identified, including: Preset abnormal thresholds; When the degree of abnormal behavior of the child is greater than or equal to the abnormal threshold, identifying that the child has abnormal behavior; The autism risk of the children was assessed using the Child Autism Screening Scale and the Child Autism Assessment Scale.

Images