DTW-based escalator pedestrian real-time abnormal posture recognition and emergency processing method
Through the combination of DTW algorithm and BlazePose tool, the joint trajectory flow of pedestrian postures is extracted and matched and recognized, which solves the real-time and accuracy of pedestrian posture monitoring on escalators, and realizes timely identification and emergency treatment of abnormal postures, which significantly improves the Escalator safety.
Patent Information
- Application Number
- CN202410969256.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-12
AI Technical Summary
The existing technology is difficult to achieve real-time posture monitoring and accurate identification of pedestrians on escalators, resulting in lagging accident response or false alarms, and the inability to effectively improve escalator safety.
Using DTW-based method, the joint trajectory flow of pedestrian poses is extracted through MediaPipe's BlazePose tool, timing feature vectors are generated, and standard library of normal and abnormal pose features is established. The DTW algorithm is used to match and identify abnormal poses, real-time pedestrians on escalators are realized. Abnormal posture detection and automatically trigger emergency response measures.
Real-time effective detection of pedestrians on escalators, timely identification of abnormal posture behaviors, and automatic emergency response measures have been taken, which significantly improves pedestrians' safety in escalators and reduces potential safety hazards.
Smart Images

Figure CN118942031A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of machine vision recognition, and in particular to a real-time abnormal posture recognition and emergency processing method for escalator pedestrians based on DTW. Background Art
[0002] In modern urban transportation systems, escalators have become an indispensable facility in public places such as subways, shopping malls and office buildings. With the increase in passenger flow, the safety issue of escalators has become increasingly prominent. Escalator failures and improper use often lead to accidents such as pedestrian falls and crush injuries, which can even endanger life in serious cases. Therefore, how to monitor the posture of pedestrians on escalators in real time and take emergency measures immediately when abnormal postures occur has become an important issue in improving public safety.
[0003] At present, escalator safety monitoring mainly relies on manual inspections and cameras installed around escalators. These methods have some limitations: manual inspections cannot achieve real-time monitoring, there is a reaction lag, and it is impossible to avoid injuries; although camera monitoring can provide a certain degree of real-time performance, due to the complex escalator environment and the diverse postures of pedestrians, methods based on simple image processing have difficulties in accurately identifying abnormal pedestrian postures. In addition, camera monitoring is greatly affected by factors such as ambient light and camera installation angle, and is prone to false alarms or missed alarms.
[0004] In order to overcome the above shortcomings, there is an urgent need for a method that can recognize the posture of escalator pedestrians in real time, immediately trigger an early warning for abnormal postures, and feed back emergency treatment measures to the escalator control end. Summary of the invention
[0005] In order to solve the problems of the prior art, an embodiment of the present invention provides a method for real-time abnormal posture recognition and emergency handling of escalator pedestrians based on DTW. The technical solution is as follows:
[0006] A DTW-based real-time abnormal posture recognition and emergency handling method for escalator pedestrians is provided, comprising the following steps:
[0007] Collect normal and abnormal posture movement samples of passengers when the escalator is in operation;
[0008] Performing smoothing preprocessing on the collected samples;
[0009] Using a human pose estimation tool to extract a human joint node trajectory stream of the processed pose samples;
[0010] Extract the angle cosine value group of the joint vector frame by frame as a feature;
[0011] Combining the features in time sequence to generate a posture time sequence feature vector;
[0012] Establish a standard library of normal posture features and an abnormal posture feature library, and define adaptive thresholds;
[0013] Collect real-time pedestrian posture samples and generate real-time posture feature vectors;
[0014] Perform DTW matching on the real-time posture feature vector and the standard samples in the normal posture library;
[0015] Determine whether the real-time posture sample belongs to a normal posture;
[0016] If it does not belong to the normal posture, DTW matching is performed with the standard samples in the abnormal posture library to determine which type of abnormal posture the real-time posture sample belongs to;
[0017] Output gesture recognition results;
[0018] Adopt corresponding response strategies according to the recognition results, and control the escalator motor to implement emergency measures when abnormal posture is detected.
[0019] Furthermore, the human body posture estimation tool is a real-time human body posture estimation model based on machine learning.
[0020] Furthermore, the abnormal posture includes but is not limited to abnormal elevator postures such as leaning forward, leaning backward, leaning sideways, and squatting.
[0021] Furthermore, the angle cosine value group of the joint vector includes:
[0022] The cosine of the angle between the head and torso and the shoulders;
[0023] The cosine of the angle between the shoulders and the center of the torso;
[0024] The cosine of the angle between the shoulder, upper arm and forearm;
[0025] The cosine of the angle between the elbow, upper arm and forearm;
[0026] The cosine value of the angle formed by the waist, torso and thigh;
[0027] The cosine of the angle between the hip, trunk and thigh;
[0028] The cosine value of the angle between the knee, thigh and calf;
[0029] The cosine value of the angle formed by the ankle, calf and foot;
[0030] And other characteristic angle cosine values that characterize the connections of the main joints of the human body.
[0031] Furthermore, the method for defining the adaptive threshold includes:
[0032] Perform DTW similarity measurement on intra-class feature vectors of the same posture in the standard library;
[0033] Calculate the mean and standard deviation of all DTW indicator values within the class;
[0034] Set a threshold coefficient k, whose value range is determined based on the actual application scenario;
[0035] Calculates the adaptive upper and lower thresholds.
[0036] Furthermore, the DTW matching process includes:
[0037] Construct a distance matrix;
[0038] Calculate the cumulative distance of the optimal regularized path;
[0039] Calculate DTW similarity.
[0040] Furthermore, the step of determining whether the real-time posture sample belongs to a normal posture includes:
[0041] Matching the current real-time posture feature vector with a standard normal posture sample group in a normal posture feature standard library;
[0042] Based on the matching result and the preset threshold, it is determined whether the current real-time posture is a normal posture.
[0043] Furthermore, the step of determining which type of abnormal posture the real-time posture sample belongs to includes:
[0044] Matching the current real-time posture feature vector with a standard abnormal posture sample group in an abnormal posture feature standard library;
[0045] Traversing the abnormal posture feature standard library to obtain the abnormal posture set to which the current real-time posture feature vector may belong;
[0046] Matching the current real-time posture feature vector to a posture in a set of abnormal postures to which it may belong;
[0047] The abnormal posture type to which the current real-time posture feature vector belongs is determined based on the matching result.
[0048] Furthermore, the emergency treatment measures include but are not limited to adjusting the escalator operation state or stopping the escalator operation.
[0049] Furthermore, the method further comprises the following steps:
[0050] Select representative samples from the sample set of a given posture as the standard posture sample of the posture;
[0051] These standard posture samples are matched first during real-time recognition.
[0052] The technical solution provided by the embodiment of the present invention has the following beneficial effects:
[0053] The present invention provides a real-time abnormal posture recognition and emergency treatment method for escalator pedestrians based on DTW. The BlazePose tool integrated in the MediaPipe software package is used to extract the joint trajectory of the real-time posture of pedestrians. After obtaining the effective time series features, the DTW algorithm is used to match and classify the standard postures, so as to achieve real-time and effective detection of the postures of pedestrians on the escalator and timely identify abnormal posture behaviors. In addition, the abnormal posture detection method described in the present invention can be automatically associated with the escalator motor after outputting the results, and corresponding emergency treatment can be performed, so as to provide preventive measures before potential safety hazards occur, thereby providing effective protection for the safety of pedestrians on the escalator. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.
[0055] Figure 1 This is a flow chart of a method for real-time abnormal posture recognition and emergency handling of escalator pedestrians based on DTW according to an embodiment of the present invention;
[0056] Figure 2 is a schematic diagram of human joint points extracted by BlazePose in an embodiment of the present invention;
[0057] Figure 3 is a schematic diagram of camera settings in an embodiment of the present invention;
[0058] Figure 4 is a flow chart of a method for establishing a standard model library of escalator pedestrian postures and a threshold value delineation method in an embodiment of the present invention;
[0059] Figure 5 It is a flow chart of the abnormal posture recognition and emergency handling method of escalator pedestrians in an embodiment of the present invention. DETAILED DESCRIPTION
[0060] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0061] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0062] The present invention provides a real-time abnormal posture recognition and emergency handling method for escalator pedestrians based on DTW, see Figure 1 , including the following steps:
[0063] Collect normal and abnormal posture movement samples of passengers when the escalator is in operation;
[0064] Performing smoothing preprocessing on the collected samples;
[0065] Using a human pose estimation tool to extract a human joint node trajectory stream of the processed pose samples;
[0066] Extract the angle cosine value group of the joint vector frame by frame as a feature;
[0067] Combining the features in time sequence to generate a posture time sequence feature vector;
[0068] Establish a standard library of normal posture features and an abnormal posture feature library, and define adaptive thresholds;
[0069] Collect real-time pedestrian posture samples and generate real-time posture feature vectors;
[0070] Perform DTW matching on the real-time posture feature vector and the standard samples in the normal posture library;
[0071] Determine whether the real-time posture sample belongs to a normal posture;
[0072] If it does not belong to the normal posture, DTW matching is performed with the standard samples in the abnormal posture library to determine which type of abnormal posture the real-time posture sample belongs to;
[0073] Output gesture recognition results;
[0074] Adopt corresponding response strategies according to the recognition results, and control the escalator motor to implement emergency measures when abnormal posture is detected.
[0075] The specific contents are as follows:
[0076] The present invention provides a method for monitoring abnormal posture of escalator pedestrians based on machine vision and dynamic time warping (DTW) algorithm, which can realize dynamic real-time recognition of abnormal posture of escalator pedestrians. The method obtains the real-time video stream of pedestrians riding the escalator through a camera, and then extracts the joint trajectory stream of the pedestrians' riding posture. The method for extracting the joint trajectory stream of the pedestrians' riding posture is completed by BlazePose of MediaPipe.
[0077] MediaPipe is an open source framework provided by Google, designed for building cross-platform, multimodal machine learning pipelines, suitable for real-time computing tasks such as computer vision and augmented reality. The BlazePose component is an efficient instant body posture tracking and estimation model, which is used to identify and locate the main joints of the human body in real-time video, including shoulders, elbows, wrists, hips, knees and ankles, to achieve rapid extraction of human joint motion trajectory streams. BlazePose is efficient and low in resource usage, and is particularly suitable for mobile devices and other resource-constrained environments.
[0078] Traditional posture recognition methods based on spatial coordinate systems and vector calculations have extremely high requirements on the number, accuracy, installation position and angle of cameras and multi-viewpoint cross-calibration algorithms. In order to overcome the above shortcomings, the present invention extracts the cosine of the angle between the vectors of each joint node of the human body as a single-frame static feature.
[0079] Although everyone's height and body shape are different, the human body's movements when performing postures are similar, that is, for the same posture, the joint angles of different people's limbs are similar. This feature extraction method can offset the differences caused by the near-big-far-small effect and the different body shapes and postures of the human body. At the same time, the number of cameras can be reduced, and the installation position and angle settings can be more flexible without the need for fine calibration.
[0080] Since different samples of the same posture in the standard posture sample library often have different durations, and the posture samples collected in real time are of irregular duration. In order to measure the similarity between the real-time samples and the standard samples, the present invention uses the DTW algorithm to match the two. DTW is an algorithm used to calculate the similarity of two unequal time series, which can effectively handle dynamic changes and deformations in time series data. Applying DTW to escalator pedestrian posture recognition can capture the movement characteristics of pedestrians on the escalator in a more real-time and effective manner, and can identify abnormal postures while they are occurring, without relying on the completeness of the posture.
[0081] Once an abnormal posture is detected, the system will immediately trigger an early warning and initiate an emergency response mechanism, such as slowing down or stopping the escalator to avoid accidents, greatly reducing the possibility of physical harm to pedestrians.
[0082] Reference Figure 4 The present invention proposes a method for establishing a standard model library for abnormal postures of escalator pedestrians and defining an adaptive threshold. The method is based on the BlazePose tool integrated in the MediaPipe framework. After importing the basic model library in the package, the human joint trajectory stream is extracted, the cosine features between joint vectors are calculated frame by frame, a standard library of abnormal posture features is constructed, and the DTW threshold limit within the same posture class is calculated.
[0083] The specific implementation steps are as follows:
[0084] Step 1: Use the escalator monitoring camera to collect normal and abnormal posture samples of passengers during escalator operation and perform smoothing. The camera position and angle are as follows: Figure 3 shown.
[0085] Furthermore, abnormal postures include leaning forward, leaning backward, leaning sideways, squatting and other abnormal elevator postures.
[0086] Step 2: Use BlazePose to process the pose action samples and extract the human joint node trajectories to obtain the pose trajectory stream of the samples.
[0087] Furthermore, in order to simplify the calculation, in the gesture recognition process, the head expression nodes with lower requirement levels are ignored, only node 0 is retained, and some palm nodes are ignored.
[0088] Step 3: Extract features from the trajectory flow samples frame by frame, extract the cosine values between the node vectors as features, and convert the node position information into cosine features.
[0089] Furthermore, for the three-dimensional vector and The cosine of the angle The formula is:
[0090]
[0091] For the sake of convenience, the cosine of the angle between O as the vertex and A and B as the end points of the vectors is abbreviated as cos(O,A,B) below.
[0092] Further, Figure 2 Based on the joint node numbers shown, the extracted angle vectors are:
[0093] cos(11,0,12),cos(12,0,11),cos(12,14,24),cos(12,11,24),cos(14,12,16),cos(16,14,22),cos(16,14,18),cos(16,18 ,22),cos(24,12,26),cos(24,23,26),cos(24,12,26),cos(26,24,28),cos(28,26,32),cos(28,26,30),cos(28,30,32),co s(11,13,23),cos(11,12,23),cos(13,11,15),cos(15,13,21),cos(15,13,17),cos(15,17,21),cos(23,11,25),cos(23,24,25),cos(23,11,25),cos(25,23,27),cos(27,25,31),cos(27,25,29),cos(27,29,31), etc., that is, one frame of whole body joint signal can get a cosine feature S with a length of 28 t
[0094] S t =(cosθ t1 ,cosθ t2 ,…,cosθ t28 )
[0095] Where t is the current frame number.
[0096] Arrange the cosine features in time sequence to obtain the posture time sequence feature vector
[0097] V={S 1 ,S 2 ,S 3 ,…,S n}
[0098] Where n is the total number of frames.
[0099] Step 4: The posture time series feature vector set is used to form a normal posture feature standard library and an abnormal posture feature standard library according to the subordinate posture. The DTW similarity of the intra-class feature vectors of the same posture in the standard library is measured to obtain an adaptive threshold. The specific steps are as follows:
[0100] 4.1 For a given posture P = {V 1 ,V 2 ,V 3 ,…,V r}, r is the total number of samples in P, matching any two posture temporal feature vectors V belonging to P i and Vj , find V i and V j The DTW similarity between them.
[0101] Further, the DTW similarity is defined as:
[0102] Assume two time series x(i) and y(j) of length n and m respectively, construct the distance matrix D
[0103]
[0104] Where d(i,j) is the Euclidean distance between two points.
[0105] The DTW similarity is the sum of the cumulative distances r(i,j) of the optimal regular path, which should satisfy the requirement that the regular path must be from d(1,1) to d(m,n) and cannot cross or intersect.
[0106] r(i,j)=d(i,j)+min(|r(i-1,j-1),r(i,j-1),r(i-1,j)|)
[0107] DTW=min(|r(m,n)|)
[0108] 4.2 For a given posture P = {V 1 ,V 2 ,V 3 ,…,V r}, r is the total number of samples in P, and the adaptive threshold is formed by all valid and non-repetitive DTW indicator values in the class.
[0109] Table 1 DTW index values
[0110]
[0111] Furthermore, the so-called valid and non-repetitive DTW index values are the non-zero parts in the upper triangle of the above table.
[0112] The adaptive threshold limit is defined as:
[0113]
[0114] in is the average value of all valid and non-repetitive DTW index values in the class. std is the standard deviation of all valid and non-repetitive DTW indicator values in the class. k is the threshold coefficient, the recommended range is [3,4], and 3.5 is taken in this case.
[0115] Below the DTW limit Limit - or above the upper DTW limit Limit+ is considered not to belong to posture P.
[0116] 4.3 For a given posture P = {V 1 ,V 2 ,V 3 ,…,V r}, r is the total number of samples in P. The top 10 samples with the lowest sum of DTW index values obtained by matching other vectors pairwise (i.e., summing and sorting Table 1 by row and taking the top 10 samples with the smallest sum) are called standard posture samples of posture P and will be matched first in real-time recognition.
[0117] Reference Figure 5 The present invention proposes a real-time recognition method for abnormal postures of escalator pedestrians based on DTW, which is used to determine whether the real-time actions of pedestrians belong to a certain abnormal posture. If it belongs to a certain abnormal posture, the escalator motor is controlled to implement emergency measures according to different posture corresponding strategies. The specific implementation steps are as follows:
[0118] Step 1: When the escalator is in operation, the camera collects video samples of pedestrians’ postures in real time.
[0119] Step 2: After smoothing the pose video samples, BlazePose is used to perform real-time analysis to extract the trajectories of human joint nodes.
[0120] Step 3: Extract the cosine features of the joint vector angle frame by frame, and combine them in time sequence to obtain the current real-time posture feature vector.
[0121] Furthermore, the method for extracting cosine features is consistent with the method for extracting cosine features in step 3 of the method for establishing a standard model library of escalator pedestrian postures.
[0122] Step 4: Convert the current real-time posture feature vector V t Perform DTW matching with the standard normal posture sample group in the normal posture feature standard library to determine V t Is this a normal posture?
[0123] The specific approach is:
[0124] 4.1 The current real-time posture feature vector V t Match with the standard normal posture sample group in the normal posture feature standard library. If the obtained DTW mean is not within the threshold range of the normal posture feature standard library, it is considered not to be a normal posture, and jump to step 5;
[0125] The current real-time posture feature vector V tMatch with other sample groups in the normal posture feature standard library. If the obtained DTW mean is not within the threshold range of the normal posture feature standard library, it is considered that it does not belong to the normal posture, and jump to step 5; if the obtained DTW mean is within the threshold range of the normal posture feature standard library, it is considered that the current real-time posture feature vector belongs to the normal posture, and jump to step 6.
[0126] Step 5: Convert the current real-time posture feature vector V t Perform DTW matching with the standard abnormal posture sample group in the abnormal posture feature standard library to determine V t Whether it is an abnormal posture.
[0127] The specific approach is:
[0128] 5.1 Let the abnormal posture feature standard library φ = {P 1 ,P 2 ,P 3 ,…,P k}, k is the abnormal posture type, which is 5 in this example.
[0129] V t With abnormal posture P i ∈φ standard abnormal posture sample group. If the obtained DTW mean is not in P i If it is within the threshold range, it is considered not to belong to posture P i . Traverse φ and get V t The possible abnormal posture set φ'. If φ' is an empty set, then V t If the pose is unknown, skip to step 6.
[0130] 5.2 V t For abnormal posture P i ∈φ' to match other sample groups. Traverse φ' to obtain the DTW mean set obtained by Vt matching
[0131] 5.3 If V t With abnormal posture P i The DTW mean obtained by matching Then determine V t Abnormal posture P i .
[0132] Step 6: Output V t Gesture recognition type. If V t If it is a normal posture or an unknown posture, no processing will be performed; if it is an abnormal posture, the escalator motor will be controlled to implement emergency processing measures such as immediate shutdown, frequency reduction to low speed, or frequency reduction to standstill according to different posture corresponding strategies.
[0133] The present invention uses the DTW algorithm to perform real-time recognition of the posture of escalator pedestrians, overcoming the shortcomings of the traditional method in that it has extremely high requirements on the number, accuracy, installation position and angle of cameras and multi-viewpoint cross-calibration algorithm. At the same time, by extracting the cosine of the angle between the vectors of each joint node of the human body as a single-frame static feature, it effectively offsets the difference caused by the near-big-far-small effect and the different body shapes and postures of the human body. In addition, the present invention can recognize abnormal postures while they are occurring, without relying on the completeness of the posture, which greatly improves the safety of escalators.
[0134] In summary, the real-time abnormal posture recognition and emergency handling method of escalator passengers based on the DTW algorithm can quickly recognize abnormal postures of escalator passengers through efficient human posture estimation and accurate similarity calculation, and automatically operate the motor, take effective emergency measures in time, ensure the safety of pedestrians, and greatly improve the safety of escalators. The application of this technology will provide strong technical support for the safety management of public transportation and commercial places.
[0135] Compared with the prior art, the present invention has the following beneficial effects:
[0136] The present invention provides a real-time abnormal posture recognition and emergency treatment method for escalator pedestrians based on DTW. The BlazePose tool integrated in the MediaPipe software package is used to extract the joint trajectory of the real-time posture of pedestrians. After obtaining the effective time series features, the DTW algorithm is used to match and classify the standard postures, so as to achieve real-time and effective detection of the postures of pedestrians on the escalator and timely identify abnormal posture behaviors. In addition, the abnormal posture detection method described in the present invention can be automatically associated with the escalator motor after outputting the results, and corresponding emergency treatment can be performed, so as to provide preventive measures before potential safety hazards occur, thereby providing effective protection for the safety of pedestrians on the escalator.
[0137] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.
Claims
1. A DTW-based real-time abnormal posture recognition and emergency handling method for escalator pedestrians, characterized in that: The following steps are involved: Collect normal and abnormal posture movement samples of passengers when the escalator is in operation; Performing smoothing preprocessing on the collected samples; Using a human pose estimation tool to extract a human joint node trajectory stream of the processed pose samples; Extract the angle cosine value group of the joint vector frame by frame as a feature; Combining the features in time sequence to generate a posture time sequence feature vector; Establish a standard library of normal posture features and an abnormal posture feature library, and define adaptive thresholds; Collect real-time pedestrian posture samples and generate real-time posture feature vectors; Perform DTW matching on the real-time posture feature vector and the standard samples in the normal posture library; Determine whether the real-time posture sample belongs to a normal posture; If it does not belong to the normal posture, DTW matching is performed with the standard samples in the abnormal posture library to determine which type of abnormal posture the real-time posture sample belongs to; Output gesture recognition results; Adopt corresponding response strategies according to the recognition results, and control the escalator motor to implement emergency measures when abnormal posture is detected.
2. The method according to claim 1, characterized in that The human body posture estimation tool is a real-time human body posture estimation model based on machine learning.
3. The method according to claim 1, characterized in that The abnormal posture includes but is not limited to non-normal elevator postures such as leaning forward, leaning backward, leaning sideways, and squatting.
4. The method according to claim 1, characterized in that: The angle cosine value group of the joint vector includes: The cosine of the angle between the head and torso and the shoulders; The cosine of the angle between the shoulders and the center of the torso; The cosine of the angle between the shoulder, upper arm and forearm; The cosine of the angle between the elbow, upper arm and forearm; The cosine value of the angle formed by the waist, torso and thigh; The cosine of the angle between the hip, trunk and thigh; The cosine value of the angle between the knee, thigh and calf; The cosine value of the angle formed by the ankle, calf and foot; And other characteristic angle cosine values that characterize the connections of the main joints of the human body.
5. The method according to claim 1, characterized in that The method for defining the adaptive threshold includes: Perform DTW similarity measurement on intra-class feature vectors of the same posture in the standard library; Calculate the mean and standard deviation of all DTW indicator values within the class; Set a threshold coefficient k, whose value range is determined based on the actual application scenario; Calculates the adaptive upper and lower thresholds.
6. The method according to claim 1, characterized in that The DTW matching process includes: Construct a distance matrix; Calculate the cumulative distance of the optimal regularized path; Calculate DTW similarity.
7. The method according to claim 1, characterized in that The step of determining whether the real-time posture sample belongs to a normal posture comprises: Matching the current real-time posture feature vector with a standard normal posture sample group in a normal posture feature standard library; Based on the matching result and the preset threshold, it is determined whether the current real-time posture is a normal posture.
8. The method according to claim 1, characterized in that The step of determining which type of abnormal posture the real-time posture sample belongs to comprises: Matching the current real-time posture feature vector with a standard abnormal posture sample group in an abnormal posture feature standard library; Traversing the abnormal posture feature standard library to obtain the abnormal posture set to which the current real-time posture feature vector may belong; Matching the current real-time posture feature vector to a posture in a set of abnormal postures to which it may belong; The abnormal posture type to which the current real-time posture feature vector belongs is determined based on the matching result.
9. The method according to claim 1, characterized in that: The emergency treatment measures include but are not limited to adjusting the escalator operation state or stopping the escalator operation.
10. The method according to claim 1, characterized in that The following steps are also included: Select representative samples from the sample set of a given posture as the standard posture sample of the posture; These standard posture samples are matched first during real-time recognition.
Citation Information
Patent Citations
Passenger abnormal behavior recognition method based on human skeleton sequence
CN109460702A
Escalator risk state monitoring method based on digital twinning
CN117237870A
Device for estimation of occupant posture
JP2011123733A
Realtime Pose recognition system using artificial intelligence and recognition method
KR102369152B1
Smart method and system for body building posture identification, assessment, warning and intensity estimation
WO2017156835A1
