CN108171187A - A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point - Google Patents
A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point Download PDFInfo
- Publication number
- CN108171187A CN108171187A CN201810007029.0A CN201810007029A CN108171187A CN 108171187 A CN108171187 A CN 108171187A CN 201810007029 A CN201810007029 A CN 201810007029A CN 108171187 A CN108171187 A CN 108171187A
- Authority
- CN
- China
- Prior art keywords
- bone point
- abnormal behaviour
- bone
- video
- confidence map
- 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
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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- 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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of abnormal behaviour automatic identifying methods and device based on the extraction of bone point, the present invention to the personnel in video by carrying out bone point extraction and detection, then line trace is clicked through to bone, finally classified using trained grader to the bone point position of different frame, so as to fulfill the identification to abnormal behaviour.Fast and accurately the abnormal behaviour of the personnel in video can be identified by the present invention and early warning, the problem of so as to avoid traditional abnormal behaviour recognition methods accuracy rate is low, real-time is poor.
Description
Technical field
The present invention relates to field of communication technology more particularly to a kind of abnormal behaviour automatic identifying methods based on the extraction of bone point
And device.
Background technology
In recent years, the accident and anomalous event to take place frequently has seriously affected the public safety of society.For taking place frequently
Accident and the behavior that endangers public security such as anomalous event, countries in the world are using abnormal behaviour identification technology as peace
A kind of important means taken precautions against entirely.
Human body abnormal behaviour refers to the irregular improper behavior of human body, running, fight, fall ill, hesitate suddenly such as human body
It wanders, personnel's largely aggregation etc..Abnormal behaviour identification technology is that personnel's exception row is detected and identified from collected video sequence
For a kind of technology, which can effectively prevent the generation of the accidents such as insurgent violence, tread event, can be monitoring
Personnel provide timely warning information, effectively related personnel are assisted to sentence in real time to the abnormal conditions occurred in monitoring scene
Break and take measures.
It, in the prior art still cannot be accurately to burst but currently for more and more accidents and anomalous event
The behavior of people is identified in event and anomalous event.
Invention content
In view of above-mentioned analysis, the present invention is intended to provide a kind of abnormal behaviour automatic identifying method based on the extraction of bone point and
Device, to solve the problems, such as that the prior art cannot accurately be identified the behavior of people in accident and anomalous event.
To solve the above problems, the present invention is mainly achieved through the following technical solutions：
The present invention provides a kind of abnormal behaviour automatic identifying method based on the extraction of bone point, this method includes：Acquisition regards
The confidence map of people and the affine field part affine field in position in frequency image carry out bone point detection to the personnel in video,
And determine bone point ownership；Using optical flow method to each bone point position of different people into line trace；By continuous videos in the predetermined time
The bone point position of frame is classified, to identify abnormal behaviour.
Further, the affine field of confidence map and position for obtaining people in video image includes：Pass through two-way multistage
Join convolution deep neural network extraction confidence map and the affine field in position.
Further, it is described that confidence map and the affine field packet in position are extracted by two-way multi-cascade convolution deep neural network
It includes：Confidence map is extracted by multi-cascade convolution deep neural network all the way to predict proprietary bone point, obtains owner
The position of bone point；The affine field in position of each bone point is extracted by multi-cascade convolution deep neural network all the way, bone is clicked through
Row ownership judges.
Further, it is described using optical flow method to each bone point position of different people carry out tracking include：
Any one video frame to one in continuous sequence of frames of video, by extracting the affine word of its confidence map and position
Two features of section determine the bone point position of human body, to any two adjacent video frames later, find the bone occurred in previous frame
The optimum position of point in the current frame obtains the position coordinates of bone point in the current frame, and such iteration carries out, with to human body bone point
Into line trace.
Further, the bone point position of successive video frames in the predetermined time is classified, to identify that abnormal behaviour includes：
One section of training video stream for including a type abnormal behaviour is obtained, makees the detection of bone point to every frame image in this section of training video stream
With the operation of bone point tracking, the position of the human body bone point of multiple image is obtained；The bone point position of all video frame is formed one
Training sample；It repeats the above steps, obtains multiple training samples in a kind of abnormal behaviour, composing training sample set；Using height
This mixed model carries out the bone point position that training sample is concentrated modeling structure grader, by the grader to abnormal behaviour
Classify.
On the other hand, the present invention also provides a kind of abnormal behaviour automatic identification equipment based on the extraction of bone point, the device packets
It includes：Bone point detection unit for obtaining the affine field of the confidence map of people and position in video image, carries out the personnel in video
Bone point detects, and determines bone point ownership；Bone point tracking cell, for being carried out using optical flow method to each bone point position of different people
Tracking；Abnormal behaviour recognition unit, for the bone point position of successive video frames in the predetermined time to be classified, to identify exception
Behavior.
Further, the bone point detection unit is additionally operable to, and is put by the extraction of two-way multi-cascade convolution deep neural network
Letter figure and the affine field in position.
Further, the bone point detection unit is additionally operable to, and is put by the extraction of multi-cascade convolution deep neural network all the way
Letter figure predicts proprietary bone point, obtains the position of owner's bone point；Pass through multi-cascade convolution depth nerve net all the way
Network extracts the affine field in position of each bone point, and ownership judgement is carried out to bone point.
Further, the bone point tracking cell is additionally operable to, any one to one in continuous sequence of frames of video regards
Frequency frame by extracting its confidence map and position two features of affine field, determines the bone point position of human body, to arbitrary two later
A adjacent video frames find the optimum position of the bone point that occurs in the current frame in previous frame, obtain bone point in the current frame
Position coordinates, such iteration carry out, to click through line trace to human body bone.
Further, the abnormal behaviour recognition unit is additionally operable to, and obtains one section of training for including a type abnormal behaviour
Video flowing makees every frame image in this section of training video stream the operation of the detection of bone point and the tracking of bone point, obtains multiple image
The position of human body bone point；The bone point position of all video frame is formed into a training sample；It repeats the above steps, obtains the first kind
Multiple training samples of abnormal behaviour, composing training sample set；The bone point position concentrated using gauss hybrid models to training sample
It puts and carries out modeling structure grader, classified by the grader to abnormal behaviour.
The present invention has the beneficial effect that：
Then the present invention clicks through bone line trace, last profit by carrying out bone point extraction and detection to the personnel in video
Classified with trained grader to the bone point position of different frame, so as to fulfill the identification to abnormal behaviour.The present invention can
Fast and accurately the abnormal behaviour of the personnel in video to be identified and early warning, know so as to avoid traditional abnormal behaviour
The problem of other method accuracy rate is low, real-time is poor.
Other features and advantages of the present invention will illustrate in the following description, and partial become from specification
It is clear that understood by implementing the present invention.The purpose of the present invention and other advantages can by the specification write,
Specifically noted structure is realized and is obtained in claims and attached drawing.
Description of the drawings
Fig. 1 is a kind of flow signal of abnormal behaviour automatic identifying method extracted based on bone point of the embodiment of the present invention
Figure；
Fig. 2 is a kind of structural representation of abnormal behaviour automatic identification equipment extracted based on bone point of the embodiment of the present invention
Figure；
Fig. 3 is the flow signal of another abnormal behaviour automatic identifying method extracted based on bone point of the embodiment of the present invention
Figure；
Fig. 4 is the structural representation of abnormal behaviour automatic identification equipment that the another kind of the embodiment of the present invention is extracted based on bone point
Figure.
Specific embodiment
The preferred embodiment of the present invention is specifically described below in conjunction with the accompanying drawings, wherein, attached drawing forms the application part, and
It is used to illustrate the principle of the present invention together with embodiments of the present invention.For purpose of clarity and simplification, when it may make the present invention
Theme it is smudgy when, illustrating in detail for known function and structure in device described herein will be omitted.
Then the embodiment of the present invention clicks through line trace by carrying out bone point extraction and detection to the personnel in video to bone,
Finally classified using trained grader to the bone point position of different frame, so as to fulfill the identification to abnormal behaviour.This
Fast and accurately the abnormal behaviour of the personnel in video can be identified for invention and early warning, so as to avoid traditional abnormal
The problem of Activity recognition method accuracy rate is low, real-time is poor.Below in conjunction with attached drawing and several embodiments, to the present invention into traveling
One step is described in detail.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, this hair is not limited
It is bright.
An embodiment of the present invention provides a kind of abnormal behaviour automatic identifying method based on the extraction of bone point, referring to Fig. 1, the party
Method includes：
S101, the confidence map of people and the affine field part affine field in position in video image are obtained, in video
Personnel carry out bone point detection, and determine bone point ownership；
S102, using optical flow method to each bone point position of different people into line trace；
S103, the bone point position of successive video frames in the predetermined time is classified, to identify abnormal behaviour.
That is, the embodiment of the present invention is to the personnel in video by carrying out bone point extraction and detection, then to bone point
Into line trace, finally classified using trained grader to the bone point position of different frame, so as to fulfill to abnormal behaviour
Identification.Fast and accurately the abnormal behaviour of the personnel in video can be identified by the present invention and early warning, so as to avoid
The problem of traditional abnormal behaviour recognition methods accuracy rate is low, real-time is poor.
Specifically, the embodiment of the present invention is when carrying out the detection of bone point, is owned by extracting confidence map and obtaining in video
Then the bone point position of personnel carries out ownership judgement by the affine field of extract part to bone point position, that is, judge the bone point category
In which individual, so as to finally determine everyone bone point position, and then multi-person interactive behavior is identified, such as row of fighting
For.
It should be noted that the prior art is such as fallen both for single independent behavior, squatting down is identified, without needle
It such as fights and is identified to multi-person interactive behavior, the method described in the embodiment of the present invention not only can be to single independent rows
To be identified, multi-person interactive behavior can also be identified.
That is, abnormal behaviour identification technology compared with prior art, the embodiment of the present invention is by human body in video image
The detection of bone point and the tracking of bone point to different video frame, automatically personnel's abnormal behaviour is identified and early warning, from
And greatly improve the accuracy rate and real-time of abnormal behaviour identification.
When it is implemented, the affine field packet of confidence map and position of people in video image is obtained described in the embodiment of the present invention
It includes：Confidence map and the affine field in position are extracted by two-way multi-cascade convolution deep neural network.
Further, confidence map and portion are extracted by two-way multi-cascade convolution deep neural network described in the embodiment of the present invention
Position is affine, and field includes：Confidence map is extracted by multi-cascade convolution deep neural network all the way to carry out in advance proprietary bone point
It surveys, obtains the position of owner's bone point；The position that each bone point is extracted by multi-cascade convolution deep neural network all the way is affine
Field carries out ownership judgement to bone point.
When it is implemented, tracking bag is carried out to each bone point position of different people using optical flow method described in the embodiment of the present invention
It includes：Any one video frame to one in continuous sequence of frames of video, by extracting the affine field two of its confidence map and position
A feature determines the bone point position of human body, to any two adjacent video frames later, finds the bone point occurred in previous frame and exists
Optimum position in present frame obtains the position coordinates of bone point in the current frame, and such iteration carries out, to be carried out to human body bone point
Tracking.
When it is implemented, the bone point position of successive video frames in the predetermined time is classified described in the embodiment of the present invention,
To identify that abnormal behaviour includes：One section of training video stream for including a type abnormal behaviour is obtained, in this section of training video stream
Every frame image make the operation of the detection of bone point and the tracking of bone point, obtain the position of the human body bone point of multiple image；By all videos
The bone point position of frame forms a training sample；It repeats the above steps, obtains multiple training samples in a kind of abnormal behaviour, structure
Into training sample set；Modeling structure grader is carried out to the bone point position that training sample is concentrated using gauss hybrid models, is passed through
The grader classifies to abnormal behaviour.
On the whole, the embodiment of the present invention by two-way multi-cascade convolution deep neural network model to people in video image
The bone point of body is detected, and then line trace is clicked through to the bone of different video frame using optical flow method, finally using grader to bone
Point position is classified, so as to fulfill the identification and early warning to personnel's abnormal behaviour.
Fig. 2 is a kind of structural representation of abnormal behaviour automatic identification equipment extracted based on bone point of the embodiment of the present invention
Figure, Fig. 3 are the flow diagrams of abnormal behaviour automatic identifying method that the another kind of the embodiment of the present invention is extracted based on bone point, under
Face will carry out the method described in the embodiment of the present invention detailed explanation and illustration with reference to Fig. 2 and Fig. 3：
Specifically, to solve the problems such as accuracy rate of the existing technology is low, real-time is poor, the embodiment of the present invention provides
A kind of abnormal behaviour identification device based on the extraction of bone point.Specifically as shown in Fig. 2, the Activity recognition device of taking pictures of the present invention is by regarding
Frequency data acquisition unit, bone point detection unit, bone point tracking cell, abnormal behaviour recognition unit, alarm unit composition.The present invention
Embodiment obtains video data first with video data acquiring unit, and the bone of human body is then carried out to collected video data
Point detection and the tracking of bone point, so as to accurately carry out abnormal behaviour identification using the bone point that traces into, finally to abnormal behaviour into
Row alarm.
The embodiment of the present invention additionally provides another abnormal behaviour automatic identifying method extracted based on bone point, specific as schemed
Shown in 3, in video data acquiring unit, video image is acquired using video acquisition equipment, it is single to be detected in bone point
Member analyzes it processing.People's puts in the video image got in bone point detection unit extraction video data acquiring unit
Letter figure and the affine field in position, so as to carry out bone point detection to the personnel in video.
It should be noted that the embodiment of the present invention need to carry out the pretreatments such as denoising to video image first, and then, the present invention
Confidence map and position two features of affine field are extracted by two-way multi-cascade convolution deep neural network.Specifically, it is of the invention
Embodiment predicts proprietary bone point by extracting confidence map using multi-cascade convolution deep neural network all the way, obtains
The position of owner's bone point；Using in addition multi-cascade convolution deep neural network is affine by the position for extracting each bone point all the way
Field carries out ownership judgement to bone point, that is, judges which individual any one bone point belongs to, so that it is determined that the bone point position of different people
It puts.
After bone point tracking cell determines the bone point position of different people, using optical flow method to each bone point position of different people
It puts into line trace.
The optical flow method of the embodiment of the present invention includes for the tracking of human body bone point：
1. any one video frame in a pair continuous sequence of frames of video, affine by extracting its confidence map and position
Two features of field determine the bone point position of human body；
2. for any two adjacent video frames after, find the bone point occurred in previous frame in the current frame most
Best placement, so as to obtain the position coordinates of bone point in the current frame；
3. such iteration carries out, the tracking of human body bone point can be realized.In abnormal behaviour recognition unit, the present invention utilizes instruction
The grader perfected classifies the bone point position of successive video frames in a period of time, so as to identify the abnormal behaviour of personnel.
The training process of grader of the embodiment of the present invention includes：
1. one section of training video stream for including certain type abnormal behaviour is obtained, to every frame image in this section of training video stream
Make the operation of the detection of bone point and the tracking of bone point, obtain the position of the human body bone point of multiple image；
2. the bone point position of all video frame is formed into a training sample；
3. repeating step 1 and 2, multiple training samples in certain class abnormal behaviour, composing training sample set are obtained；
4. modeling structure grader is carried out to the bone point position that training sample is concentrated using gauss hybrid models, to utilize
The trained grader classifies to abnormal behaviour.
If the output result of grader is abnormal, system provides abnormal alarm, does not otherwise alarm.
It should be noted that two-way multi-cascade convolution deep neural network and optical flow method are existing technologies, so herein
No longer this is described in detail.Also, the grader of the embodiment of the present invention is not limited to gauss hybrid models, those skilled in the art
Member can set other graders according to actual needs, and the embodiment of the present invention is not especially limited this.
The embodiment of the present invention additionally provides a kind of abnormal behaviour automatic identification equipment extracted based on bone point, should referring to Fig. 4
Device includes：
Bone point detection unit, for obtaining the confidence map of people and the affine field part in position in video image
Affinefield carries out the personnel in video bone point detection, and determines bone point ownership；
Bone point tracking cell, for using optical flow method to each bone point position of different people into line trace；
Abnormal behaviour recognition unit, for the bone point position of successive video frames in the predetermined time to be classified, with identification
Abnormal behaviour.
That is, the embodiment of the present invention is to the personnel in video by carrying out bone point extraction and detection, then to bone point
Into line trace, finally classified using trained grader to the bone point position of different frame, so as to fulfill to abnormal behaviour
Identification.Fast and accurately the abnormal behaviour of the personnel in video can be identified by the present invention and early warning, so as to avoid
The problem of traditional abnormal behaviour recognition methods accuracy rate is low, real-time is poor.
Specifically, the embodiment of the present invention is when carrying out the detection of bone point, is owned by extracting confidence map and obtaining in video
Then the bone point position of personnel carries out ownership judgement by the affine field of extract part to bone point position, that is, judge the bone point category
In which individual, so as to finally determine everyone bone point position, and then multi-person interactive behavior is identified, such as row of fighting
For.
It should be noted that the prior art is such as fallen both for single independent behavior, squatting down is identified, without needle
It such as fights and is identified to multi-person interactive behavior, the method described in the embodiment of the present invention not only can be to single independent rows
To be identified, multi-person interactive behavior can also be identified.
That is, abnormal behaviour recognition methods compared with prior art, the embodiment of the present invention is by human body in video image
The detection of bone point and the tracking of bone point to different video frame, automatically personnel's abnormal behaviour is identified and early warning, from
And greatly improve the accuracy rate and real-time of abnormal behaviour identification.
When it is implemented, bone point detection unit is additionally operable to described in the embodiment of the present invention, pass through two-way multi-cascade convolution depth
Neural network extracts confidence map and the affine field in position.
Further, bone point detection unit is additionally operable to described in the embodiment of the present invention, passes through the god of multi-cascade convolution depth all the way
Proprietary bone point is predicted through network extraction confidence map, obtains the position of owner's bone point；It is rolled up by multi-cascade all the way
Product deep neural network extracts the affine field in position of each bone point, and ownership judgement is carried out to bone point.
When it is implemented, bone point tracking cell is additionally operable to described in the embodiment of the present invention, to a continuous sequence of frames of video
In any one video frame, by extracting its confidence map and position two features of affine field, determine the bone point position of human body,
To any two adjacent video frames later, the optimum position of the bone point occurred in previous frame in the current frame is found, obtains bone
The position coordinates of point in the current frame, such iteration carry out, to click through line trace to human body bone.
Abnormal behaviour recognition unit described in the embodiment of the present invention is additionally operable to when specific implementation is implemented, and is obtained different comprising a type
One section of training video stream of Chang Hangwei makees every frame image in this section of training video stream the behaviour of the detection of bone point and the tracking of bone point
Make, obtain the position of the human body bone point of multiple image；The bone point position of all video frame is formed into a training sample；In repetition
Step is stated, obtains multiple training samples of first kind abnormal behaviour, composing training sample set；Using gauss hybrid models to training
Bone point position in sample set carries out modeling structure grader, is classified by the grader to abnormal behaviour.
On the whole, the embodiment of the present invention by two-way multi-cascade convolution deep neural network model to people in video image
The bone point of body is detected, and then line trace is clicked through to the bone of different video frame using optical flow method, finally using grader to bone
Point position is classified, so as to fulfill the identification and early warning to personnel's abnormal behaviour.
The related content of the embodiment of the present invention can refer to embodiment of the method part and be understood, not be described in detail herein.
Correspondingly, the embodiment of the present invention also provides a kind of computer readable storage medium, the computer-readable storage
There are one media storages or multiple programs, one or more of programs can be performed by one or more processor, with
Realize that any automatic identification that previous embodiment provides is taken pictures the method for behavior, therefore can also realize corresponding technology
Effect, relevant portion can refer to embodiment of the method and understood, in this not go into detail.
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited thereto,
Any one skilled in the art in the technical scope disclosed by the present invention, the change or replacement that can be readily occurred in,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the protection model of claims
Subject to enclosing.
Claims (10)
1. a kind of abnormal behaviour automatic identifying method based on the extraction of bone point, which is characterized in that including：
Obtain video image in the confidence map of people and the affine field part affine field in position, to the personnel in video into
Row bone point detects, and determines bone point ownership；
Using optical flow method to each bone point position of different people into line trace；
The bone point position of successive video frames in predetermined time is classified, to identify abnormal behaviour.
2. according to the method described in claim 1, it is characterized in that, described obtain the confidence map of people and position parent in video image
Include with field：
Confidence map and the affine field in position are extracted by two-way multi-cascade convolution deep neural network.
3. according to the method described in claim 2, it is characterized in that, described carried by two-way multi-cascade convolution deep neural network
The affine field of confidence map and position is taken to include：
Confidence map is extracted by multi-cascade convolution deep neural network all the way to predict proprietary bone point, obtains owner
The position of bone point；
The affine field in position of each bone point is extracted by multi-cascade convolution deep neural network all the way, ownership is carried out to bone point and is sentenced
It is disconnected.
4. according to the method described in any one in claim 1-3, which is characterized in that it is described using optical flow method to different people
Each bone point position carries out tracking and includes：
Any one video frame to one in continuous sequence of frames of video, by extracting the affine field two of its confidence map and position
A feature determines the bone point position of human body, to any two adjacent video frames later, finds the bone point occurred in previous frame and exists
Optimum position in present frame obtains the position coordinates of bone point in the current frame, and such iteration carries out, to be carried out to human body bone point
Tracking.
5. according to the method described in any one in claim 1-3, which is characterized in that by successive video frames in the predetermined time
Bone point position is classified, to identify that abnormal behaviour includes：
One section of training video stream for including a type abnormal behaviour is obtained, bone point is made to every frame image in this section of training video stream
Detection and the operation of bone point tracking obtain the position of the human body bone point of multiple image；
The bone point position of all video frame is formed into a training sample；
It repeats the above steps, obtains multiple training samples in a kind of abnormal behaviour, composing training sample set；
Modeling structure grader is carried out to the bone point position that training sample is concentrated using gauss hybrid models, passes through the grader
Classify to abnormal behaviour.
6. a kind of abnormal behaviour automatic identification equipment based on the extraction of bone point, which is characterized in that including：
Bone point detection unit, for obtaining the confidence map of people and the affine field part affine field in position in video image,
Personnel in video are carried out with bone point detection, and determines bone point ownership；
Bone point tracking cell, for using optical flow method to each bone point position of different people into line trace；
Abnormal behaviour recognition unit, for the bone point position of successive video frames in the predetermined time to be classified, to identify exception
Behavior.
7. device according to claim 6, which is characterized in that
The bone point detection unit is additionally operable to, and extracts confidence map by two-way multi-cascade convolution deep neural network and position is affine
Field.
8. device according to claim 7, which is characterized in that
The bone point detection unit is additionally operable to, and confidence map is extracted to proprietary by multi-cascade convolution deep neural network all the way
Bone point is predicted, obtains the position of owner's bone point；Each bone point is extracted by multi-cascade convolution deep neural network all the way
The affine field in position, ownership judgement is carried out to bone point.
9. according to the device described in any one in claim 6-8, which is characterized in that
The bone point tracking cell is additionally operable to, any one video frame to one in continuous sequence of frames of video passes through extraction
Its confidence map and position two features of affine field determine the bone point position of human body, to any two adjacent video frames later,
The optimum position of the bone point occurred in previous frame in the current frame is found, obtains the position coordinates of bone point in the current frame, so
Iteration carries out, to click through line trace to human body bone.
10. according to the device described in any one in claim 6-8, which is characterized in that
The abnormal behaviour recognition unit is additionally operable to, and one section of training video stream for including a type abnormal behaviour is obtained, to the section
Every frame image in training video stream makees the operation of the detection of bone point and the tracking of bone point, obtains the position of the human body bone point of multiple image
It puts；The bone point position of all video frame is formed into a training sample；It repeats the above steps, obtains the more of first kind abnormal behaviour
A training sample, composing training sample set；Modeling structure is carried out to the bone point position that training sample is concentrated using gauss hybrid models
Grader is built, is classified by the grader to abnormal behaviour.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810007029.0A CN108171187A (en) | 2018-01-04 | 2018-01-04 | A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810007029.0A CN108171187A (en) | 2018-01-04 | 2018-01-04 | A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108171187A true CN108171187A (en) | 2018-06-15 |
Family
ID=62517285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810007029.0A Pending CN108171187A (en) | 2018-01-04 | 2018-01-04 | A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108171187A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543513A (en) * | 2018-10-11 | 2019-03-29 | 平安科技（深圳）有限公司 | Method, apparatus, equipment and the storage medium that intelligent monitoring is handled in real time |
| CN111046819A (en) * | 2019-12-18 | 2020-04-21 | 浙江大华技术股份有限公司 | Behavior recognition processing method and device |
| CN111091060A (en) * | 2019-11-20 | 2020-05-01 | 吉林大学 | Deep learning-based fall and violence detection method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787469A (en) * | 2016-03-25 | 2016-07-20 | 广州市浩云安防科技股份有限公司 | Method and system for pedestrian monitoring and behavior recognition |
| CN106485055A (en) * | 2016-09-22 | 2017-03-08 | 吉林大学 | A kind of old type ii diabetes patient moving training system based on Kinect sensor |
-
2018
- 2018-01-04 CN CN201810007029.0A patent/CN108171187A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787469A (en) * | 2016-03-25 | 2016-07-20 | 广州市浩云安防科技股份有限公司 | Method and system for pedestrian monitoring and behavior recognition |
| CN106485055A (en) * | 2016-09-22 | 2017-03-08 | 吉林大学 | A kind of old type ii diabetes patient moving training system based on Kinect sensor |
Non-Patent Citations (1)
| Title |
|---|
| ZHE CAO ET AL: "《Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields》", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543513A (en) * | 2018-10-11 | 2019-03-29 | 平安科技（深圳）有限公司 | Method, apparatus, equipment and the storage medium that intelligent monitoring is handled in real time |
| CN111091060A (en) * | 2019-11-20 | 2020-05-01 | 吉林大学 | Deep learning-based fall and violence detection method |
| CN111091060B (en) * | 2019-11-20 | 2022-11-04 | 吉林大学 | Fall and violence detection method based on deep learning |
| CN111046819A (en) * | 2019-12-18 | 2020-04-21 | 浙江大华技术股份有限公司 | Behavior recognition processing method and device |
| CN111046819B (en) * | 2019-12-18 | 2023-09-05 | 浙江大华技术股份有限公司 | Behavior recognition processing method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108197575A (en) | A kind of abnormal behaviour recognition methods detected based on target detection and bone point and device | |
| CN103246869B (en) | Method is monitored in crime based on recognition of face and behavior speech recognition | |
| CN108053427A (en) | A kind of modified multi-object tracking method, system and device based on KCF and Kalman | |
| CN109117721A (en) | A kind of pedestrian hovers detection method | |
| CN108062349A (en) | Video frequency monitoring method and system based on video structural data and deep learning | |
| CN108664608A (en) | Recognition methods, device and the computer readable storage medium of a suspect | |
| CN110443109A (en) | Abnormal behaviour monitor processing method, device, computer equipment and storage medium | |
| CN107204114A (en) | A kind of recognition methods of vehicle abnormality behavior and device | |
| CN106022296B (en) | A kind of fake-licensed car detection method of the probability polymerization based on vehicle hot spot region | |
| CN106529477B (en) | Video human Activity recognition method based on significant track and temporal-spatial evolution information | |
| CN105787472B (en) | A kind of anomaly detection method based on the study of space-time laplacian eigenmaps | |
| CN110136447A (en) | Lane change of driving a vehicle detects and method for distinguishing is known in illegal lane change | |
| CN108629282A (en) | A kind of smoking detection method, storage medium and computer | |
| CN108229407A (en) | A kind of behavioral value method and system in video analysis | |
| CN110425005A (en) | The monitoring of transportation of belt below mine personnel's human-computer interaction behavior safety and method for early warning | |
| CN102426757A (en) | Safe driving monitoring system and method based on pattern recognition | |
| CN104637246A (en) | Driver multi-behavior early warning system and danger evaluation method | |
| CN108171187A (en) | A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point | |
| CN108846332A (en) | A kind of railway drivers Activity recognition method based on CLSTA | |
| CN104504400A (en) | Detection method of driver's abnormal behavior modeled based on online behavior | |
| CN106127814A (en) | A kind of wisdom gold eyeball identification gathering of people is fought alarm method and device | |
| CN108986472A (en) | One kind turns around vehicle monitoring method and device | |
| CN103198296A (en) | Method and device of video abnormal behavior detection based on Bayes surprise degree calculation | |
| CN112071084A (en) | Method and system for judging illegal parking by utilizing deep learning | |
| CN109543577A (en) | A kind of fatigue driving detection method for early warning based on facial expression feature |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180615 |
|
| RJ01 | Rejection of invention patent application after publication |