CN112381072B - Human body abnormal behavior detection method based on time-space information and human-object interaction - Google Patents
Human body abnormal behavior detection method based on time-space information and human-object interaction Download PDFInfo
- Publication number
- CN112381072B CN112381072B CN202110030865.2A CN202110030865A CN112381072B CN 112381072 B CN112381072 B CN 112381072B CN 202110030865 A CN202110030865 A CN 202110030865A CN 112381072 B CN112381072 B CN 112381072B
- Authority
- CN
- China
- Prior art keywords
- human
- information
- result
- frame
- data
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 42
- 230000003993 interaction Effects 0.000 title claims abstract description 27
- 206010000117 Abnormal behaviour Diseases 0.000 title claims abstract description 21
- 230000009471 action Effects 0.000 claims abstract description 20
- 230000002159 abnormal effect Effects 0.000 claims abstract description 12
- 230000004927 fusion Effects 0.000 claims abstract description 10
- 238000002372 labelling Methods 0.000 claims abstract description 9
- 230000008859 change Effects 0.000 claims abstract description 7
- 230000005484 gravity Effects 0.000 claims abstract description 6
- 238000000034 method Methods 0.000 claims description 14
- 238000012544 monitoring process Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 5
- 238000013480 data collection Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 230000002452 interceptive effect Effects 0.000 claims description 4
- 230000009194 climbing Effects 0.000 abstract description 3
- 230000002085 persistent effect Effects 0.000 abstract 2
- 230000006399 behavior Effects 0.000 description 22
- 206010016173 Fall Diseases 0.000 description 6
- 238000005457 optimization Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 208000012661 Dyskinesia Diseases 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Alarm Systems (AREA)
Abstract
The invention discloses a human body abnormal behavior detection method based on time-space information and human-object interaction, which comprises the following steps: s1, data acquisition and labeling; s2, extracting the position information of the people and the objects; s3, extracting motion information of people and objects; s4, modeling the characteristic interaction relationship of the people and the objects; s5, behavior classification and fusion; and S6, optimizing the detection result. Aiming at the problems of abnormal actions of falling, climbing and limb conflict and detection of persistent abnormal states, the abnormal actions are assisted and judged in a human interaction mode, the persistent states of the abnormal actions are detected by combining the change condition of the gravity center, and meanwhile, the normal actions of walking, standing and sitting can be detected besides the abnormal actions.
Description
Technical Field
The invention relates to the technical field of computer vision and deep learning, in particular to a human body abnormal behavior detection method based on space-time information and human-object interaction.
Background
The human body abnormal behavior detection has important application in the fields of security and intelligent monitoring, so that the pressure of manual monitoring is relieved to a great extent, and the detection efficiency is improved. The existing solutions adopt manual feature extraction motion features for judgment, and the accuracy rate is low in actual real scene application; however, some current methods based on deep learning can only detect one abnormal behavior and cannot adapt to automatic determination of multiple abnormal behaviors under real conditions. However, abnormal movements such as climbing and falling down have a certain specificity, and it is necessary to detect not only the abnormal movement being performed by the actor in real time but also to be able to continuously determine the state of the abnormal movement. For example, after falling down, the user may lie still on the table and climb up, and then continuously walk on the table or other auxiliary objects, which all bring challenges to the existing detection technology, and the existing method cannot detect the continuous state of the abnormal motion, so a new technical method is needed to solve the problem.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a human body abnormal behavior detection method based on space-time information and human-object interaction, and solves the problems in the background art.
In order to achieve the purpose, the invention provides the following technical scheme: a human body abnormal behavior detection method based on space-time information and human-object interaction comprises the following steps: s1, data acquisition and labeling; s2, extracting the position information of the people and the objects; s3, extracting motion information of people and objects; s4, modeling the characteristic interaction relationship of the people and the objects; s5, behavior classification and fusion; s6, optimizing the detection result; the abnormal behavior refers to a behavior beyond a normal range, has certain scene correlation, and represents an unacceptable behavior in the scene.
Preferably, the data collection and labeling in step S1 includes: collecting normal actions and abnormal actions in video monitoring, cutting video data, generating initial spatial positions of people and objects through an SSD target detection network, and finally, manually correcting the generated position information by using a simple marking tool, correcting and detecting inaccurate object positions to obtain accurate position information; the normal action refers to an action which can be accepted in a monitoring scene, and the normal action comprises walking, sitting or standing; and abnormal actions represent actions that are not accepted in the scene, and the abnormal actions comprise falling, climbing or limb conflict.
Preferably, the simple marking tool is used for correcting the position information of the frame, reading and displaying the picture and the corresponding person and object frame thereof, judging whether the position of the display frame is accurate or not, and redrawing a new frame through a mouse, wherein the new data can cover the old data.
Preferably, the extracting of the position information of the person and the object in step S2 includes fine-tuning the collected data set by the SSD object detection network pre-trained on the MS COCO data set to accurately detect the position of the person and the object.
Preferably, the fine tuning means that on the basis of a model pre-trained by an MS COCO data set, only the last two layers of the network are retrained for training data, and parameters of the remaining layers are kept unchanged.
Preferably, the extracting of the motion information of the person and the object in step S3 includes using a 3D-shuffle network as a backbone network of the spatio-temporal motion information, taking an input segment composed of the current frame and the previous 15 frames of data as input data, performing feature extraction on the input 16 frames of data, and finally obtaining a spatio-temporal information feature map of a single frame.
Preferably, the modeling of the human-object feature interaction relationship in step S4 includes applying the position information of the human and the object obtained in step S2 to the feature map extracted in step S3 to obtain spatiotemporal feature information, and individually cropping the features of the human and the object to perform the interaction modeling, wherein the formula is as follows:
wherein
representing the correlation of the spatio-temporal features of the ith individual with the ensemble of individual object features,
representing spatiotemporal motion characteristics of an ith person;
representing a feature of a jth object;
representing the current frame object feature set.
Representing a model of the relationship of a person to an object,
representing the result of integrating multiple character relationship models.
Preferably, the behavior classification and fusion in step S5 includes: the method comprises the following steps of respectively carrying out behavior classification on human motion information and a human-object interaction relation model, fusing two classification results to obtain a primary detection result, wherein a fusion formula is as follows:
whereinCshow that
And
the obtained action classification results are fused with the classification scores of (1),
represents the classification result score obtained by the human motion information,
representing the classification result scores obtained by modeling the interaction relationship between the human and the object,
is a learnable hyper-parameter, which indicates the importance of the result if
If the correlation between the behavior and the object is less than 0.5, the correlation between the behavior and the object is small, the model focuses more on the classification result of the human motion information, and otherwise, the model focuses more on the classification result of the interactive relation modeling of the human and the object.
Preferably, the optimizing of the detection result in step S6 includes: judging whether the action of falling over the ground is detected or not according to the preliminary detection result of the previous frame, and if the action of falling over the ground is not detected, taking the preliminary detection result of the previous frame as a final result and outputting a behavior category; if detecting a falling motionCalculating the gravity center point of the human body through the position frame, and calculating the speed change information of the adjacent frames to obtain
，
Indicating the velocity change information of the adjacent frames
And a threshold value
Making a comparison, if less than the threshold
If so, indicating that the mobile terminal is still in a falling state, and covering the detected result with the result; if it is greater than or equal to the threshold value
Then, the state that the model is not in a fallen state is indicated, the result detected by the model is taken as a final result, and the behavior category is output.
The invention has the beneficial effects that: by the method, the target detection module can accurately position the specific spatial positions of the behavior people and the objects, and the model can finally give the behavior types of the behavior people. And finally, drawing the human body frame and the behavior category on an original picture (without an object frame), and recording the abnormal behavior category. The method mainly utilizes human-object interaction modeling analysis, behavior classification fusion and optimization of results based on a gravity center speed model, adopts a human interaction mode to assist in judging abnormal behaviors, and detects the persistence state of the abnormal behaviors by combining the change condition of the gravity center. Meanwhile, the invention can detect normal actions of walking, standing and sitting besides abnormal actions.
Drawings
FIG. 1 is a diagram of a network model of the present invention;
FIG. 2 is a flow chart of data collection and labeling according to the present invention;
FIG. 3 is a flow chart of the detection result optimization according to the present invention;
FIG. 4 is a flow chart of the model training and operation of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1-4, the model training and operation flow is shown in fig. 3, and the present invention provides a technical solution: a human body abnormal behavior detection method based on space-time information and human-object interaction comprises the following steps: (1) collecting and marking data; (2) extracting position information of people and objects; (3) extracting motion information of people and objects; (4) modeling the characteristic interaction relation of the human and the object; (5) behavior classification and fusion; (6) and optimizing the detection result.
(1) Data collection and annotation
The method collects normal actions and abnormal actions in a real video monitoring scene, in order to facilitate data labeling, video data of the real scene is cut, then an SSD target detection network is used for generating initial space positions of people and objects, a network model diagram is shown in figure 1, finally a simple labeling tool of the invention is used for manually correcting generated position information and correcting the position of an inaccurate object, and a specific flow is shown in figure 2.
Description of the simple labeling tool: the tool is mainly used for correcting the position information of the frame, the picture and the corresponding person and object frames thereof can be read and displayed, a user can judge whether the position of the display frame is accurate or not, a new frame is redrawn through a mouse, and the new data can cover the old data.
(2) Extracting position information of people and objects
The invention finely adjusts the collected data set by the SSD (Single Shot Multi Box Detector) target detection network pre-trained on the MS COCO data set so as to adapt to the target characteristics in the monitoring scene and accurately detect the positions of people and objects.
The fine adjustment mode comprises the following steps: on the basis of a model pre-trained by an MS COCO data set, only the last two layers of the network are retrained according to training data, and parameters of the other layers are kept unchanged.
(3) Extracting motion information of people and objects
In order to give consideration to both the running speed and the detection accuracy, the invention provides a main network using 3D-ShuffleNet as space-time motion information, and the specific process is as follows:
1) data sampling, the invention uses 16 frames of data as input, and the concrete sampling process is as follows: taking the current frame and the previous 15 frames of data to jointly form an input segment as input data;
2) the method comprises the steps of performing feature extraction on input 16 frame data by using a space-time downsampling mode, and finally obtaining a single-frame space-time information feature map by performing feature downsampling.
(4) Modeling human and object feature interaction relationships
The main process of the module comprises the following steps:
1) applying the position information obtained in the step (2) to the feature map obtained in the step (3) to obtain space-time feature information;
2) the characteristics of people and objects are cut out separately for interactive modeling analysis, and the formula is as follows:
wherein
Representing spatiotemporal motion characteristics of an ith person;
representing a feature of a jth object;
representing the current frame object feature set.
Representing a model of the relationship of a person to an object,
the result of integrating multiple character relationship models, both of which are implemented by convolutional neural networks, is represented.
(5) Behavior classification and fusion
The module mainly comprises three steps:
1) performing behavior classification on the human motion information obtained in the step (3);
2) performing behavior classification on the relation model established in the step (4);
3) the two classification results are fused, and the formula is as follows:
wherein,
represents 1) the obtained classification result score,
represents the classification result scores obtained by the relational modeling in 2),
is a learnable hyper-parameter, which represents the importance of the result, if the relationship between the behavior and the object is small, then
Smaller, models focus more on classification in 1)As a result, otherwise, the classification result in 2) is more important.
(4) Optimization of test results
This step is mainly used to optimize the detection result of the falling-down abnormal behavior, because there is a possibility that the human motion information is less after falling down and cannot be distinguished from the normal behavior by using the deep learning method alone, after the action of falling down is detected, the change of the center-of-gravity speed of the human body is calculated to assist in determining whether the human body is still in the falling-down state, and the optimization flow is as shown in fig. 4.
The optimization process of the current detection result is as follows:
1) judging whether the action of falling over the ground is detected or not according to the preliminary detection result of the previous frame, and if the action of falling over the ground is not detected, taking the preliminary detection result of the previous frame as a final result and outputting a behavior category; if the action of falling to the ground is detected, the second step is carried out;
2) calculating the gravity center point of the human body through the position frame, and calculating the speed change information of the adjacent frames to obtain
(ii) a 3) Will be provided with
And a threshold value
Making a comparison, if less than the threshold
If so, indicating that the mobile terminal is still in a falling state, and covering the detected result with the result; if it is greater than or equal to the threshold value
This indicates that the user is no longer in a fallen state (e.g., standing up from the fallen position), and the results detected by the model are used as the final behavior category.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that various changes in the embodiments and/or modifications of the invention can be made, and equivalents and modifications of some features of the invention can be made without departing from the spirit and scope of the invention.
Claims (6)
1. A human body abnormal behavior detection method based on space-time information and human-object interaction is characterized by comprising the following steps: s1, data acquisition and labeling; s2, extracting the position information of the people and the objects; s3, extracting motion information of people and objects; s4, modeling the characteristic interaction relationship of the people and the objects; s5, behavior classification and fusion; s6, optimizing the detection result;
the data collection and labeling in step S1 includes: collecting normal actions and abnormal actions in video monitoring, cutting video data, generating initial spatial positions of people and objects through an SSD target detection network, and finally, manually correcting the generated position information by using a simple marking tool, correcting and detecting inaccurate object positions to obtain accurate position information;
the modeling of the human-object feature interaction relationship in the step S4 includes applying the position information of the human and the object obtained in the step S2 to the feature map extracted in the step S3 to obtain spatiotemporal feature information; the characteristics of the human and the object are cut out separately for interactive modeling, and the formula is as follows, R (P)i)＝Fα{Gβ(Pi,Oj),OjE.g. O, wherein R (P)i) Representing the correlation of the spatio-temporal features of the ith person with the totality of the individual object features, PiRepresenting spatiotemporal motion characteristics of an ith person; o isjRepresenting a feature of a jth object; o represents the object feature set of the current frame; gβRepresenting a model of the relationship of a person to an object, FαRepresenting a result of integrating the plurality of character relationship models;
the optimizing of the detection result in the step S6 includes: judging whether the action of falling the ground is detected or not according to the preliminary detection result of the previous frame, and if the action of falling the ground is not detected, judging that the previous frame is the next frameTaking the preliminary detection result of the frame as a final result and outputting a behavior category; if the action of falling to the ground is detected, the gravity center point of the human body is calculated through the position frame, and the speed change information of the adjacent frames is calculated to obtain ViWill ViComparing with the threshold value mu, if the threshold value mu is smaller than the threshold value mu, indicating that the ground is still in a fallen state, and overlaying the detected result with the result; if the value is larger than or equal to the threshold value mu, the state that the model is not in a fallen state is indicated, the result detected by the model is taken as a final result, and the behavior category is output.
2. The human body abnormal behavior detection method based on spatiotemporal information and human-object interaction according to claim 1, characterized in that: the simple marking tool is used for correcting the position information of the frame, reading and displaying the picture and the corresponding person and object frame, judging whether the position of the display frame is accurate or not, and redrawing a new frame through a mouse, wherein the new data can cover the old data.
3. The human body abnormal behavior detection method based on spatiotemporal information and human-object interaction according to claim 1, characterized in that: the extracting of the position information of the person and the object in the step S2 includes fine-tuning the acquired data set by the SSD object detection network pre-trained on the MS COCO data set to accurately detect the position of the person and the object.
4. The human body abnormal behavior detection method based on spatiotemporal information and human-object interaction according to claim 3, characterized in that: the fine tuning means that only the last two layers of the network are retrained aiming at training data on the basis of a model pre-trained by an MS COCO data set, and parameters of the other layers are kept unchanged.
5. The human body abnormal behavior detection method based on spatiotemporal information and human-object interaction according to claim 1, characterized in that: the extracting of the motion information of the person and the object in the step S3 includes using a 3D-shuffle network as a backbone network of the spatio-temporal motion information, taking an input segment composed of the current frame and the previous 15 frames of data as input data, performing feature extraction on the input 16 frames of data, and finally obtaining a spatio-temporal information feature map of a single frame.
6. The human body abnormal behavior detection method based on spatiotemporal information and human-object interaction according to claim 1, characterized in that: the behavior classification and fusion in step S5 includes: the method comprises the following steps of respectively carrying out behavior classification on human motion information and a human-object interaction relation model, fusing two classification results to obtain a primary detection result, wherein a fusion formula is as follows: c ═ 1-theta ═ S1+θ*S2Wherein C represents S1And S2The classification scores of (a) are fused to obtain an action classification result, S1Score of classification result obtained by representing human motion information, S2And if theta is less than 0.5, the correlation between the behavior and the object is small, the model pays more attention to the classification result of the human motion information, and otherwise, the model pays more attention to the classification result of the interactive relationship modeling between the person and the object.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110030865.2A CN112381072B (en) | 2021-01-11 | 2021-01-11 | Human body abnormal behavior detection method based on time-space information and human-object interaction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110030865.2A CN112381072B (en) | 2021-01-11 | 2021-01-11 | Human body abnormal behavior detection method based on time-space information and human-object interaction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112381072A CN112381072A (en) | 2021-02-19 |
| CN112381072B true CN112381072B (en) | 2021-05-25 |
Family
ID=74590054
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110030865.2A Active CN112381072B (en) | 2021-01-11 | 2021-01-11 | Human body abnormal behavior detection method based on time-space information and human-object interaction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112381072B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114926837B (en) * | 2022-05-26 | 2023-08-04 | 东南大学 | Emotion recognition method based on human-object space-time interaction behavior |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104510475A (en) * | 2014-12-15 | 2015-04-15 | 中国科学院计算技术研究所 | Human body falling-down detection method and system |
| CN110321780A (en) * | 2019-04-30 | 2019-10-11 | 苏州大学 | Exception based on spatiotemporal motion characteristic falls down behavioral value method |
| CN110569773A (en) * | 2019-08-30 | 2019-12-13 | 江南大学 | Double-flow network behavior identification method based on space-time significance behavior attention |
| CN111310689A (en) * | 2020-02-25 | 2020-06-19 | 陕西科技大学 | Method for recognizing human body behaviors in potential information fusion home security system |
| CN111325073A (en) * | 2018-12-17 | 2020-06-23 | 上海交通大学 | Monitoring video abnormal behavior detection method based on motion information clustering |
| CN111738218A (en) * | 2020-07-27 | 2020-10-02 | 成都睿沿科技有限公司 | Human body abnormal behavior recognition system and method |
| CN111898514A (en) * | 2020-07-24 | 2020-11-06 | 燕山大学 | Multi-target visual supervision method based on target detection and action recognition |
| CN112149616A (en) * | 2020-10-13 | 2020-12-29 | 西安电子科技大学 | Figure interaction behavior recognition method based on dynamic information |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102236783A (en) * | 2010-04-29 | 2011-11-09 | 索尼公司 | Method and equipment for detecting abnormal actions and method and equipment for generating detector |
| WO2018126323A1 (en) * | 2017-01-06 | 2018-07-12 | Sportlogiq Inc. | Systems and methods for behaviour understanding from trajectories |
| US11450145B2 (en) * | 2017-04-12 | 2022-09-20 | Disney Enterprise, Inc. | System and method for monitoring procedure compliance |
| US10572723B2 (en) * | 2017-12-07 | 2020-02-25 | Futurewei Technologies, Inc. | Activity detection by joint human and object detection and tracking |
| CN110555404A (en) * | 2019-08-29 | 2019-12-10 | 西北工业大学 | Flying wing unmanned aerial vehicle ground station interaction device and method based on human body posture recognition |
| CN111339668B (en) * | 2020-02-28 | 2022-05-10 | 西南交通大学 | Crowd evacuation visualization method based on emotion cognition |
| CN111582122B (en) * | 2020-04-29 | 2021-03-16 | 成都信息工程大学 | System and method for intelligently analyzing behaviors of multi-dimensional pedestrians in surveillance video |
| CN111709306B (en) * | 2020-05-22 | 2023-06-09 | 江南大学 | Double-flow network behavior identification method based on multilevel space-time feature fusion enhancement |
| CN111797705A (en) * | 2020-06-11 | 2020-10-20 | 同济大学 | Action recognition method based on character relation modeling |
| CN111767888A (en) * | 2020-07-08 | 2020-10-13 | 北京澎思科技有限公司 | Object state detection method, computer device, storage medium, and electronic device |
| CN112052795B (en) * | 2020-09-07 | 2022-10-18 | 北京理工大学 | Video behavior identification method based on multi-scale space-time feature aggregation |
-
2021
- 2021-01-11 CN CN202110030865.2A patent/CN112381072B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104510475A (en) * | 2014-12-15 | 2015-04-15 | 中国科学院计算技术研究所 | Human body falling-down detection method and system |
| CN111325073A (en) * | 2018-12-17 | 2020-06-23 | 上海交通大学 | Monitoring video abnormal behavior detection method based on motion information clustering |
| CN110321780A (en) * | 2019-04-30 | 2019-10-11 | 苏州大学 | Exception based on spatiotemporal motion characteristic falls down behavioral value method |
| CN110569773A (en) * | 2019-08-30 | 2019-12-13 | 江南大学 | Double-flow network behavior identification method based on space-time significance behavior attention |
| CN111310689A (en) * | 2020-02-25 | 2020-06-19 | 陕西科技大学 | Method for recognizing human body behaviors in potential information fusion home security system |
| CN111898514A (en) * | 2020-07-24 | 2020-11-06 | 燕山大学 | Multi-target visual supervision method based on target detection and action recognition |
| CN111738218A (en) * | 2020-07-27 | 2020-10-02 | 成都睿沿科技有限公司 | Human body abnormal behavior recognition system and method |
| CN112149616A (en) * | 2020-10-13 | 2020-12-29 | 西安电子科技大学 | Figure interaction behavior recognition method based on dynamic information |
Non-Patent Citations (3)
| Title |
|---|
| iCAN: Instance-Centric Attention Network for Human-Object Interaction Detection;Chen Gao等;《arXiv:1808.10437v1》;20180830;第1-13页引言、第3节，图3 * |
| 基于时空交互注意力模型的人体行为识别算法;潘娜等;《激光与光电子学进展》;20200930;第57卷(第18期);第181506-1 - 181506-9页 * |
| 基于深度学习的行为检测方法综述;高陈强等;《重庆邮电大学学报(自然科学版)》;20201231;第32卷(第6期);第991-1002页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112381072A (en) | 2021-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830252B (en) | Convolutional neural network human body action recognition method fusing global space-time characteristics | |
| US11983919B2 (en) | Video anomaly detection method based on human-machine cooperation | |
| CN111797771B (en) | Weak supervision video behavior detection method and system based on iterative learning | |
| CN110532970B (en) | Age and gender attribute analysis method, system, equipment and medium for 2D images of human faces | |
| CN103324937B (en) | The method and apparatus of label target | |
| CN104123545B (en) | A kind of real-time human facial feature extraction and expression recognition method | |
| CN111832400B (en) | Mask wearing condition monitoring system and method based on probabilistic neural network | |
| CN110674785A (en) | Multi-person posture analysis method based on human body key point tracking | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN113807276A (en) | Smoking behavior identification method based on optimized YOLOv4 model | |
| CN112906631B (en) | Dangerous driving behavior detection method and detection system based on video | |
| CN108288015A (en) | Human motion recognition method and system in video based on THE INVARIANCE OF THE SCALE OF TIME | |
| US20200306585A1 (en) | Jump counting method for jump rope | |
| CN109389185A (en) | Use the video smoke recognition methods of Three dimensional convolution neural network | |
| CN112801000B (en) | Household old man falling detection method and system based on multi-feature fusion | |
| US20230237694A1 (en) | Method and system for detecting children's sitting posture based on face recognition of children | |
| CN112381072B (en) | Human body abnormal behavior detection method based on time-space information and human-object interaction | |
| CN115346149A (en) | Rope skipping counting method and system based on space-time diagram convolution network | |
| CN110674887A (en) | End-to-end road congestion detection algorithm based on video classification | |
| CN104463909A (en) | Visual target tracking method based on credibility combination map model | |
| CN111091069A (en) | Power grid target detection method and system guided by blind image quality evaluation | |
| CN117137435B (en) | Rehabilitation action recognition method and system based on multi-mode information fusion | |
| US20230237682A1 (en) | Method and system for binocular ranging for children | |
| CN107688828A (en) | A kind of bus degree of crowding estimating and measuring method based on mobile phone sensor | |
| CN114639168B (en) | Method and system for recognizing running gesture |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |