CN114358194A - Gesture tracking based detection method for abnormal limb behaviors of autism spectrum disorder - Google Patents
Gesture tracking based detection method for abnormal limb behaviors of autism spectrum disorder Download PDFInfo
- Publication number
- CN114358194A CN114358194A CN202210014731.6A CN202210014731A CN114358194A CN 114358194 A CN114358194 A CN 114358194A CN 202210014731 A CN202210014731 A CN 202210014731A CN 114358194 A CN114358194 A CN 114358194A
- Authority
- CN
- China
- Prior art keywords
- video
- child
- motion
- children
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention relates to an autism spectrum disorder abnormal limb behavior detection method based on posture tracking, which belongs to the technical field of video behavior analysis, and comprises the steps of preprocessing an acquired child diagnosis video, extracting detection frames of all people in the video by using a retraining target detection algorithm YOLOv3, cutting to obtain a sub-graph, detecting human body joint points of the sub-graph by using an HRNet human posture estimation method, extracting apparent characteristics of the sub-graph by using an OSNet pedestrian re-recognition network for multi-target tracking to obtain motion tracks of left and right shoulders, elbows and wrists of children, calculating a motion speed mean value and a motion speed standard deviation of the motion tracks, performing leave-one-out cross verification on the obtained characteristics by using a forest random algorithm, and finally performing comprehensive judgment on classification results; the invention detects the motion tracks of the left shoulder, the right shoulder, the elbow and the wrist of the child, is more comprehensive than single joint detection information, extracts two types of characteristics related to the motion speed and comprehensively judges, thereby improving the detection accuracy.
Description
Technical Field
The invention belongs to the technical field of video behavior analysis, and particularly relates to a gesture tracking-based method for detecting abnormal limb behaviors of autism spectrum disorder.
Background
Autism Spectrum Disorder (ASD), also known as Autism, is a relatively serious developmental Disorder. The autism spectrum disorder patients mainly have three types of core symptoms of social communication disorder, narrow interests and stereotypical and repeated behavior patterns. According to the statistics of the world health organization, one of every 160 children is the autism spectrum disorder child in the world, the number of the autism spectrum disorder children in China is increased by 20 ten thousand every year, and the number of the autism spectrum disorder children in China is far more than 1000 ten thousand so far. At present, clinically, the definitive diagnosis of autism spectrum disorder is generally 3-4 years old. Therefore, the detection of the abnormal limb behaviors of the children with early autism spectrum disorder is researched through the video behavior analysis and deep learning technology, the doctor is helped to carry out auxiliary judgment, and early discovery and early treatment are realized, so that different treatment methods are adopted for early intervention, the symptoms of the children with the autism spectrum disorder are relieved, and the method has important clinical and social meanings.
At present, three types of methods are applied to the diagnosis of children with autism spectrum disorder, the first method is a traditional medical method, the diagnosis of children with autism spectrum disorder is medically performed, and the diagnosis is mainly performed by depending on parents filling out questionnaires and evaluating different evaluation scales for children, wherein, ADOS-2 is the most authoritative evaluation method at present, the evaluation scales with different grades are adopted according to different ages and language levels, and the evaluation is performed on the children with autism spectrum disorder by observing the performances of the children in games and using materials, and the evaluation is divided into three grades, namely light, medium and heavy. The children grade needs to be judged by an assessment physician with professional training, but the number of professional assessment personnel is small, the requirement for early diagnosis of the autism spectrum disorder patient is difficult to meet, the system is complex, the time consumption is long, and the system has certain subjectivity. The second method is acceleration measurement, which requires multiple sensor devices to be worn by the child to extract parameters, but this method causes discomfort and rejection by the child, thereby affecting the measurement results of the child. The third method is to use computer vision technology, but most of the methods focus on eye tracking and face recognition, and neglect analysis of the child limb behaviors. And children with autism spectrum disorder sometimes show weak or obvious behavior confusion, so that the limb behavior analysis of the children is favorable for early discovery of the autism spectrum disorder.
Disclosure of Invention
In order to solve the problems in the prior art, the invention aims to provide a posture tracking-based method for detecting the abnormal limb behaviors of the autism spectrum disorder, which is used for detecting the motion tracks of the left shoulder, the right shoulder, the elbow and the wrist of a child, obtaining more comprehensive information than single joint detection, extracting two types of characteristics related to the motion speed for comprehensive judgment and improving the accuracy of detecting the autism spectrum disorder.
The invention utilizes a target detection algorithm YOLOv3, a human body posture estimation method HRNet and a pedestrian re-recognition network OSNet to track the human body posture of a child, calculates the motion speed mean value and the motion speed standard deviation of the motion trail of the left shoulder, the right shoulder, the elbow and the wrist of the child, carries out leave-one cross validation on the obtained characteristics by using a random forest algorithm, and comprehensively judges the classification result to obtain whether the child is normal or not, comprising the following steps:
1.1 video acquisition of children: the method comprises the following steps of collecting monitoring videos of doctors when the doctors diagnose children, and acquiring child limb behavior video data in the interaction process of the children and the doctors:
1.1.1 screening out a child video with an evaluation scale of ADOS-2 and evaluation grades of T, 1 and 2 from the acquired monitoring video, and roughly intercepting the monitoring video after a doctor makes the same limb movement;
1.1.2 preprocessing the video of the children obtained in the step 1.1.1, converting the video into frame images, selecting continuous 750 frame images without noise interference, and naming uniformly;
1.1.3, converting 750 frames of images obtained in the step 1.1.2 into a 30s child video, wherein the sampling frame rate is 25 frames/second;
1.2 tracking the motion trail of the children: for the children videos obtained in the step 1.1.3, motion tracks of the left shoulder, the right shoulder, the elbow and the wrist of the children are obtained by utilizing a human body posture tracking algorithm, and the method comprises the following steps:
1.2.1 human body frame detection: carrying out reasonable frame extraction on the preprocessed frame images obtained in the step 1.1.2, making a human body frame detection data set, wherein the number of the frame images which are acquired in total is 1223, then carrying out manual marking, training a YOLOv3 network, inputting the video of the child obtained in the step 1.1.3 into a retrained YOLOv3 network, obtaining detection frames of the child, the doctor and the parent in the video, and cutting the detection frames to obtain sub-images;
1.2.2 detection of human joint points: for the subgraph obtained in the step 1.2.1, detecting the joint points of the human body by using an HRNet human body posture estimation method to obtain the position coordinates and confidence coefficients of all the joint points;
1.2.3 multi-target tracking: for the sub-image obtained in the step 1.2.1, extracting the apparent characteristics of the sub-image by using an OSNet pedestrian re-identification network, calculating the probability that different people belong to the same target, performing data association, and distributing a digital ID (identity) to each person in the video frame image;
1.2.4 through the steps 1.2.1, 1.2.2 and 1.2.3, the ID information of the child in the video can be obtained, and the motion tracks of the left shoulder, the right shoulder, the elbow and the wrist joint points of the child can be obtained by tracking;
1.3 analyzing the motion track: calculating the motion speed of the motion tracks of the joints of the left shoulder, the right shoulder, the elbow and the wrist of the child obtained by tracking in the step 1.2.4, and extracting the mean value and the standard deviation of the motion speed as characteristic vectors, wherein the method comprises the following steps:
1.3.1, screening the motion tracks of the joints of the left shoulder, the right shoulder, the elbow and the wrist of the child, and removing samples with the frame number less than 700 frames and obviously wrong tracking of the posture of the human body;
1.3.2 extracting the position information of the x-axis coordinate change and the y-axis coordinate change of six joint points of the left shoulder, the right shoulder, the elbow and the wrist from the sample obtained in the step 1.3.1, and calculating the movement speed;
1.3.3, calculating the motion speed mean values of six joint points of the left shoulder, the right shoulder, the elbow and the wrist according to the motion speeds obtained in the step 1.3.2 to obtain a fused speed mean value feature vector;
1.3.4 calculating the standard deviation of the movement speeds of the six joint points of the left shoulder, the right shoulder, the elbow and the wrist according to the movement speeds obtained in the step 1.3.2 to obtain a fused speed standard deviation feature vector;
1.4 training the fused speed mean feature vector obtained in the step 1.3.3 and the fused speed standard deviation feature vector obtained in the step 1.3.4 by using a random forest, comprising the following steps:
1.4.1 taking a sample with the diagnosis result of the doctor evaluation scale as a light grade as a normal sample, taking a sample with the diagnosis result of the doctor evaluation scale as a medium grade and a sample with the diagnosis result of the doctor evaluation scale as a heavy grade as an abnormal sample, setting a normal sample label as 1, and setting an abnormal sample label as-1;
1.4.2, in the training process, one-leave cross validation is adopted, namely, under the condition that the total number of samples is N, each sample is independently used as a validation set, and the rest N-1 samples are used as training sets;
1.4.3 the random forest is a classifier composed of a plurality of decision trees, the construction of the random forest needs to be composed of two aspects of random selection of data and random selection of characteristics, and the classification result finally output by the random forest is determined by the mode of the classification result output by each decision tree;
1.4.4 respectively constructing sample sets for the fused speed mean characteristic vector obtained in the step 1.3.3 and the fused speed standard deviation characteristic vector obtained in the step 1.3.4, setting 2/3 samples with the total number of the extracted original sample sets replaced each time as a sub-sample set, and obtaining two optimal random forest models by adjusting the number of random forests including decision trees, the randomly selected characteristic number and the maximum depth of each decision tree;
1.5 comprehensive judgment of abnormal limb behaviors of autism spectrum disorder: and (3) comprehensively judging the classification result according to the two optimal random forest models obtained in the step (1.4.4), and defining a judgment rule: if the classification results obtained by the two optimal random forest models are both-1, judging that the children are autism spectrum disorder; and if at least one of the classification results obtained by the two optimal random forest models is 1, judging the child to be normal.
The calculated movement velocity of step 1.3.2 is calculated by the following formula:
v＝s/t
wherein: s is the distance and t is the time;
according to the position information of the x-axis coordinate change and the y-axis coordinate change of the six joint points of the left shoulder, the elbow and the wrist obtained in the step 1.3.2, the motion of the child is non-uniform curvilinear motion, the video frame rate of the child is 25 frames/s, the speed between two adjacent frames is obtained, the time interval is set to be 1 for calculation convenience, the motion track between two adjacent frames is quantized into an approximate straight line, and the speed calculation formula obtained by the pythagorean theorem is as follows:
wherein: the subscript t +1 denotes the t +1 th frame information, and the subscript t denotes the t-th frame information.
The invention has the beneficial effects that:
1. the invention adopts an autism spectrum disorder abnormal limb behavior detection method based on posture tracking, firstly, the collected monitoring video when the children are diagnosed is preprocessed, the target detection algorithm YOLOv3, the human posture estimation method HRNet and the pedestrian re-recognition network OSNet are utilized to track the human posture of the children, wherein, for the target detection algorithm YOLOv3, the preprocessed video frame image is reasonably framed, a human detection frame data set with 1223 data quantity is constructed, the YOLOv3 is retrained, the human posture tracking effect can be improved, the motion trail obtained by human posture tracking is screened, the number of frames is less than 700 frames and the human posture tracking obvious samples are removed, the samples meeting the conditions are subjected to the calculation of the motion speed mean value and the motion speed standard deviation, the obtained characteristics are trained by a random forest, and obtaining two optimal random forest models, and finally, comprehensively judging the classification result to obtain an optimal recognition result.
2. The invention adopts the body behavior detection to carry out auxiliary judgment on the autism spectrum disorder besides the existing eye movement tracking and face recognition methods, combines the information of six body joints of the left shoulder, the right shoulder, the elbow and the wrist, is more comprehensive than the information obtained by single joint information, simultaneously carries out cross validation on the obtained characteristics by using a random forest, solves the problem of small sample number, comprehensively judges the classification result, improves the accuracy, carries out auxiliary judgment on whether children suffer from the autism spectrum disorder, can realize early discovery and early treatment and has important significance for relieving the autism spectrum disorder disease.
Drawings
FIG. 1 is a flow chart of a method for detecting abnormal limb behaviors of autism spectrum disorders based on posture tracking;
FIG. 2 is an effect diagram of human body posture tracking of a single frame image;
FIG. 3 is a track image of a certain detected child's left shoulder movement speed;
fig. 4 is a flow chart for determining whether a child is autism spectrum disorder.
Detailed Description
The following describes the implementation process of the present invention with reference to the attached drawings.
As shown in fig. 1, the invention provides a posture tracking-based abnormal limb behavior detection method for autism spectrum disorder, which comprises the following steps:
1.1 video acquisition of children: the method comprises the following steps of collecting monitoring videos of doctors when the doctors diagnose children, and acquiring child limb behavior video data in the interaction process of the children and the doctors:
1.1.1 screening out the videos of children with an evaluation scale of ADOS-2 and evaluation grades of T, 1 and 2 from the collected monitoring videos, wherein most of the videos are children with the evaluation scale grade of 1, the children are aged 31 months or more and are not often expressed by phrases, and after a doctor makes the same limb activity, the monitoring videos are roughly intercepted, and the time is controlled between 31s and 35 s;
1.1.2 preprocessing the video of the children obtained in the step 1.1.1, converting the video into frame images, selecting continuous 750 frame images without noise interference, and naming uniformly;
1.1.3, converting 750 frames of images obtained in the step 1.1.2 into a 30s child video, wherein the sampling frame rate is 25 frames/second;
1.2 tracking the motion trail of the children: for the children videos obtained in the step 1.1.3, motion tracks of the left shoulder, the right shoulder, the elbow and the wrist of the children are obtained by utilizing a human body posture tracking algorithm, and the method comprises the following steps:
1.2.1 human body frame detection: for the child video obtained in the step 1.1.3, a body frame is detected by using YOLOv3 to obtain a detection frame of all people in a video frame image, the detection frame is cut to obtain a sub-image, the scene contains children, doctors and parents, so that the body posture of the children needs to be obtained by adopting a body posture estimation and multi-target tracking algorithm, the scene is small, the situation that the multi-target tracking is inaccurate due to too short distance between the children and the doctors or the parents is easy to occur, in order to improve the accuracy of the tracking of the body posture, reasonable frame extraction operation is carried out on the preprocessed frame image obtained in the step 1.1.2, a body frame detection data set is manufactured, the number of the frame images which are collected in total is 1223, then manual labeling is carried out, the body detection frame is framed as a true value, a YOLOv3 network is trained, the detection effect of the body detection frame is improved, and the tracking effect of the body posture is further improved;
1.2.2 detection of human joint points: for the subgraph obtained in the step 1.2.1, a human body joint point detection is carried out by using an HRNet human body posture estimation method, the HRNet is parallelly connected with a high-resolution sub-network to a low-resolution sub-network instead of serial connection, high-resolution characteristics are kept, and more accurate position information is obtained by repeating multi-scale information fusion;
1.2.3 multi-target tracking: the workflow of the multi-target tracking algorithm comprises five steps, namely, an original frame of a given video; operating the human body detector to obtain a detection frame of the human body; for each detected body frame, calculating different features, typically motion features and appearance features; calculating the similarity; performing association, and distributing a digital ID to each person, wherein the apparent characteristics of the subgraph obtained in the step 1.2.1 are extracted by using an OSNet pedestrian re-identification network, and the digital ID is distributed to each person in the video frame image;
1.2.4 through the steps 1.2.1, 1.2.2 and 1.2.3, the ID information of the child in the video can be obtained, and the motion tracks of the left shoulder, the right shoulder, the elbow and the wrist joint points of the child can be obtained by tracking;
1.3 analyzing the motion track: calculating the motion speed of the motion tracks of the joints of the left shoulder, the right shoulder, the elbow and the wrist of the child obtained by tracking in the step 1.2.4, and extracting the mean value and the standard deviation of the motion speed as characteristic vectors, wherein the method comprises the following steps:
1.3.1, screening the motion tracks of the joints of the left shoulder, the right shoulder, the elbow and the wrist of the child, and removing samples with the frame number less than 700 frames and obviously wrong tracking of the posture of the human body;
1.3.2 extracting the position information of the x-axis coordinate change and the y-axis coordinate change of six joint points of the left shoulder, the right shoulder, the elbow and the wrist from the sample obtained in the step 1.3.1, and calculating the movement speed by the following formula:
v＝s/t
wherein: s is the distance and t is the time;
according to the obtained position information of the x-axis coordinate change and the y-axis coordinate change of the six joint points of the left shoulder, the right shoulder, the elbow and the wrist, the motion of the child is not uniform-speed curvilinear motion, the video frame rate of the child is 25 frames/s, the speed between two adjacent frames is obtained, the time interval is set to be 1 for convenience of calculation, the motion trail between the two adjacent frames is quantized into an approximate straight line, and the speed calculation formula is obtained by the pythagorean theorem as follows:
wherein: subscript t +1 represents the t +1 th frame information, and subscript t represents the t th frame information;
1.3.3, calculating the motion speed mean values of six joint points of the left shoulder, the right shoulder, the elbow and the wrist according to the motion speeds obtained in the step 1.3.2 to obtain a fused speed mean value feature vector;
1.3.4 calculating the standard deviation of the movement speeds of the six joint points of the left shoulder, the right shoulder, the elbow and the wrist according to the movement speeds obtained in the step 1.3.2 to obtain a fused speed standard deviation feature vector;
1.4 training the fused speed mean feature vector obtained in the step 1.3.3 and the fused speed standard deviation feature vector obtained in the step 1.3.4 by using a random forest, comprising the following steps:
1.4.1 taking a sample with the diagnosis result of the doctor evaluation scale as a light grade as a normal sample, taking a sample with the diagnosis result of the doctor evaluation scale as a medium grade and a sample with the diagnosis result of the doctor evaluation scale as a heavy grade as an abnormal sample, setting a normal sample label as 1, and setting an abnormal sample label as-1;
1.4.2 in the random forest training process, a leave-one cross validation is adopted, namely under the condition that the total number of samples is N, each sample is independently used as a validation set, and the rest N-1 samples are used as training sets, so that the problem of small number of samples is effectively solved;
1.4.3 the random forest is a classifier composed of a plurality of decision trees, each decision tree is not related, the classification result finally output by the random forest is determined by the mode of the classification result output by each decision tree, the construction of the random forest needs to be composed of two aspects of random selection of data and random selection of characteristics, wherein the random selection of the data refers to that n samples are randomly extracted from an original sample set and put back to serve as a sub-sample set; the random selection of the features refers to that d features are randomly extracted from all the features to serve as splitting features of each decision tree, wherein elements in each sub sample set can be repeated, elements in the same sub sample set can also be repeated, and the number of the finally constructed sub sample sets is the same as that of the decision trees contained in the random forest;
1.4.4 respectively constructing sample sets for the fused speed mean characteristic vector obtained in the step 1.3.3 and the fused speed standard deviation characteristic vector obtained in the step 1.3.4, setting 2/3 samples with the total number of the extracted original sample sets replaced each time as a sub-sample set, and obtaining two optimal random forest models by adjusting the number of random forests including decision trees, the randomly selected characteristic number and the maximum depth of each decision tree;
1.5 comprehensive judgment of abnormal limb behaviors of autism spectrum disorder: and (3) comprehensively judging the classification result according to the two optimal random forest models obtained in the step (1.4.4), and defining a judgment rule: if the classification results obtained by the two optimal random forest models are both-1, judging that the children are autism spectrum disorder; and if at least one of the classification results obtained by the two optimal random forest models is 1, judging the child to be normal.
The performance analysis of the method for detecting the abnormal limb behaviors of the autism spectrum disorder based on the posture tracking provided by the invention is as follows: three metrics are defined that judge the performance of the methods herein: sensitivity (SE), Specificity (SP), Accuracy (AC), the specific formula is as follows:
wherein: define True Positives (TP): the children with autism spectrum disorder are judged as children with autism spectrum disorder. False Positive (FP): the normal children are judged to be autistic spectrum disorder children. True Negative (TN): the normal child is judged as a normal child. False Negative (FN): the children with autism spectrum disorder are judged as normal children. The results are shown in the following table:
the invention adopts the characteristics of speed mean value and speed standard deviation to carry out comprehensive judgment, and the Sensitivity (SE) is 61.54%, the Specificity (SP) is 65.22% and the Accuracy (AC) is 62.90% which can be obtained from the table 1.
Claims (2)
1. The method for detecting the abnormal limb behaviors of the autism spectrum disorder based on posture tracking is characterized by comprising the following steps of:
1.1 video acquisition of children: the method comprises the following steps of collecting monitoring videos of doctors when the doctors diagnose children, and acquiring child limb behavior video data in the interaction process of the children and the doctors:
1.1.1 screening out a child video with an evaluation scale of ADOS-2 and evaluation grades of T, 1 and 2 from the acquired monitoring video, and roughly intercepting the monitoring video after a doctor makes the same limb movement;
1.1.2 preprocessing the video of the children obtained in the step 1.1.1, converting the video into frame images, selecting continuous 750 frame images without noise interference, and naming uniformly;
1.1.3, converting 750 frames of images obtained in the step 1.1.2 into a 30s child video, wherein the sampling frame rate is 25 frames/second;
1.2 tracking the motion trail of the children: for the children videos obtained in the step 1.1.3, motion tracks of the left shoulder, the right shoulder, the elbow and the wrist of the children are obtained by utilizing a human body posture tracking algorithm, and the method comprises the following steps:
1.2.1 human body frame detection: carrying out reasonable frame extraction on the preprocessed frame images obtained in the step 1.1.2, making a human body frame detection data set, wherein the number of the frame images which are acquired in total is 1223, then carrying out manual marking, training a YOLOv3 network, inputting the video of the child obtained in the step 1.1.3 into a retrained YOLOv3 network, obtaining detection frames of the child, the doctor and the parent in the video, and cutting the detection frames to obtain sub-images;
1.2.2 detection of human joint points: for the subgraph obtained in the step 1.2.1, detecting the joint points of the human body by using an HRNet human body posture estimation method to obtain the position coordinates and confidence coefficients of all the joint points;
1.2.3 multi-target tracking: for the sub-image obtained in the step 1.2.1, extracting the apparent characteristics of the sub-image by using an OSNet pedestrian re-identification network, calculating the probability that different people belong to the same target, performing data association, and distributing a digital ID (identity) to each person in the video frame image;
1.2.4 through the steps 1.2.1, 1.2.2 and 1.2.3, the ID information of the child in the video can be obtained, and the motion tracks of the left shoulder, the right shoulder, the elbow and the wrist joint points of the child can be obtained by tracking;
1.3 analyzing the motion track: calculating the motion speed of the motion tracks of the joints of the left shoulder, the right shoulder, the elbow and the wrist of the child obtained by tracking in the step 1.2.4, and extracting the mean value and the standard deviation of the motion speed as characteristic vectors, wherein the method comprises the following steps:
1.3.1, screening the motion tracks of the joints of the left shoulder, the right shoulder, the elbow and the wrist of the child, and removing samples with the frame number less than 700 frames and obviously wrong tracking of the posture of the human body;
1.3.2 extracting the position information of the x-axis coordinate change and the y-axis coordinate change of six joint points of the left shoulder, the right shoulder, the elbow and the wrist from the sample obtained in the step 1.3.1, and calculating the movement speed;
1.3.3, calculating the motion speed mean values of six joint points of the left shoulder, the right shoulder, the elbow and the wrist according to the motion speeds obtained in the step 1.3.2 to obtain a fused speed mean value feature vector;
1.3.4 calculating the standard deviation of the movement speeds of the six joint points of the left shoulder, the right shoulder, the elbow and the wrist according to the movement speeds obtained in the step 1.3.2 to obtain a fused speed standard deviation feature vector;
1.4 training the fused speed mean feature vector obtained in the step 1.3.3 and the fused speed standard deviation feature vector obtained in the step 1.3.4 by using a random forest, comprising the following steps:
1.4.1 taking a sample with the diagnosis result of the doctor evaluation scale as a light grade as a normal sample, taking a sample with the diagnosis result of the doctor evaluation scale as a medium grade and a sample with the diagnosis result of the doctor evaluation scale as a heavy grade as an abnormal sample, setting a normal sample label as 1, and setting an abnormal sample label as-1;
1.4.2, in the training process, one-leave cross validation is adopted, namely, under the condition that the total number of samples is N, each sample is independently used as a validation set, and the rest N-1 samples are used as training sets;
1.4.3 the random forest is a classifier composed of a plurality of decision trees, the construction of the random forest needs to be composed of two aspects of random selection of data and random selection of characteristics, and the classification result finally output by the random forest is determined by the mode of the classification result output by each decision tree;
1.4.4 respectively constructing sample sets for the fused speed mean characteristic vector obtained in the step 1.3.3 and the fused speed standard deviation characteristic vector obtained in the step 1.3.4, setting 2/3 samples with the total number of the extracted original sample sets replaced each time as a sub-sample set, and obtaining two optimal random forest models by adjusting the number of random forests including decision trees, the randomly selected characteristic number and the maximum depth of each decision tree;
1.5 comprehensive judgment of abnormal limb behaviors of autism spectrum disorder: and (3) comprehensively judging the classification result according to the two optimal random forest models obtained in the step (1.4.4), and defining a judgment rule: if the classification results obtained by the two optimal random forest models are both-1, judging that the children are autism spectrum disorder; and if at least one of the classification results obtained by the two optimal random forest models is 1, judging the child to be normal.
2. The method for detecting abnormal limb behaviors of autism spectrum disorder based on posture tracking as claimed in claim 1, wherein the calculated movement velocity in step 1.3.2 is calculated by the following formula:
v＝s/t
wherein: s is the distance and t is the time;
according to the position information of the x-axis coordinate change and the y-axis coordinate change of the six joint points of the left shoulder, the elbow and the wrist obtained in the step 1.3.2, the motion of the child is non-uniform curvilinear motion, the video frame rate of the child is 25 frames/s, the speed between two adjacent frames is obtained, the time interval is set to be 1 for calculation convenience, the motion track between two adjacent frames is quantized into an approximate straight line, and the speed calculation formula obtained by the pythagorean theorem is as follows:
wherein: the subscript t +1 denotes the t +1 th frame information, and the subscript t denotes the t-th frame information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210014731.6A CN114358194A (en) | 2022-01-07 | 2022-01-07 | Gesture tracking based detection method for abnormal limb behaviors of autism spectrum disorder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210014731.6A CN114358194A (en) | 2022-01-07 | 2022-01-07 | Gesture tracking based detection method for abnormal limb behaviors of autism spectrum disorder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114358194A true CN114358194A (en) | 2022-04-15 |
Family
ID=81106975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210014731.6A Pending CN114358194A (en) | 2022-01-07 | 2022-01-07 | Gesture tracking based detection method for abnormal limb behaviors of autism spectrum disorder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114358194A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116453702A (en) * | 2023-03-24 | 2023-07-18 | 北京大学第六医院 | Data processing method, device, system and medium for autism behavior feature set |
| CN116602663A (en) * | 2023-06-02 | 2023-08-18 | 深圳市震有智联科技有限公司 | Intelligent monitoring method and system based on millimeter wave radar |
| CN116665310A (en) * | 2023-07-28 | 2023-08-29 | 中日友好医院(中日友好临床医学研究所) | Method and system for identifying and classifying tic disorder based on weak supervision learning |
| CN117351405A (en) * | 2023-12-06 | 2024-01-05 | 江西珉轩智能科技有限公司 | Crowd behavior analysis system and method |
-
2022
- 2022-01-07 CN CN202210014731.6A patent/CN114358194A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116453702A (en) * | 2023-03-24 | 2023-07-18 | 北京大学第六医院 | Data processing method, device, system and medium for autism behavior feature set |
| CN116453702B (en) * | 2023-03-24 | 2023-11-17 | 北京大学第六医院 | Data processing method, device, system and medium for autism behavior feature set |
| CN116602663A (en) * | 2023-06-02 | 2023-08-18 | 深圳市震有智联科技有限公司 | Intelligent monitoring method and system based on millimeter wave radar |
| CN116602663B (en) * | 2023-06-02 | 2023-12-15 | 深圳市震有智联科技有限公司 | Intelligent monitoring method and system based on millimeter wave radar |
| CN116665310A (en) * | 2023-07-28 | 2023-08-29 | 中日友好医院(中日友好临床医学研究所) | Method and system for identifying and classifying tic disorder based on weak supervision learning |
| CN116665310B (en) * | 2023-07-28 | 2023-11-03 | 中日友好医院(中日友好临床医学研究所) | Method and system for identifying and classifying tic disorder based on weak supervision learning |
| CN117351405A (en) * | 2023-12-06 | 2024-01-05 | 江西珉轩智能科技有限公司 | Crowd behavior analysis system and method |
| CN117351405B (en) * | 2023-12-06 | 2024-02-13 | 江西珉轩智能科技有限公司 | Crowd behavior analysis system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114358194A (en) | Gesture tracking based detection method for abnormal limb behaviors of autism spectrum disorder | |
| CN109543526B (en) | True and false facial paralysis recognition system based on depth difference characteristics | |
| Chen et al. | Computerized wrist pulse signal diagnosis using modified auto-regressive models | |
| CN104783782B (en) | Electrocardiosignal automatic testing method and device | |
| CN110110633B (en) | Method for automatically identifying and analyzing hemiplegic gait based on machine learning | |
| CN109815858B (en) | Target user gait recognition system and method in daily environment | |
| CN108511055B (en) | Ventricular premature beat recognition system and method based on classifier fusion and diagnosis rules | |
| CN108256453A (en) | A kind of method based on one-dimensional ECG signal extraction two dimension CNN features | |
| CN110755085B (en) | Motion function evaluation method and equipment based on joint mobility and motion coordination | |
| CN112617835B (en) | Multi-feature fusion fatigue detection method based on transfer learning | |
| CN111653356A (en) | New coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning | |
| CN111403026A (en) | Facial paralysis grade assessment method | |
| Tobias et al. | CNN-based deep learning model for chest X-ray health classification using tensorflow | |
| CN106778047A (en) | A kind of traditional Chinese medical science facial diagnosis integrated system based on various dimensions medical image | |
| CN108305680A (en) | Intelligent parkinsonism aided diagnosis method based on multi-element biologic feature and device | |
| CN110379509A (en) | A kind of Breast Nodules aided diagnosis method and system based on DSSD | |
| CN107358646A (en) | A kind of fatigue detecting system and method based on machine vision | |
| CN111401435B (en) | Human body motion mode identification method based on motion bracelet | |
| CN113768519A (en) | Method for analyzing consciousness level of patient based on deep learning and resting state electroencephalogram data | |
| Yang et al. | Automatic detection pipeline for accessing the motor severity of Parkinson’s disease in finger tapping and postural stability | |
| CN115349828A (en) | Neonate pain assessment system based on computer deep learning | |
| CN114550299A (en) | System and method for evaluating daily life activity ability of old people based on video | |
| CN103735247B (en) | For the palm colour analysis method of health status identification | |
| Guan | Application of logistic regression algorithm in the diagnosis of expression disorder in Parkinson's disease | |
| CN110123280A (en) | A kind of finger dexterity detection method based on the identification of intelligent mobile terminal operation behavior |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |