CN109359544A - A kind of portrait search method and device - Google Patents
A kind of portrait search method and device Download PDFInfo
- Publication number
- CN109359544A CN109359544A CN201811091048.2A CN201811091048A CN109359544A CN 109359544 A CN109359544 A CN 109359544A CN 201811091048 A CN201811091048 A CN 201811091048A CN 109359544 A CN109359544 A CN 109359544A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- recognition result
- video
- network model
- gait
- 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.)
- Granted
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
- G06V40/23—Recognition of whole body movements, e.g. for sport training
- G06V40/25—Recognition of walking or running movements, e.g. gait recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of portrait search method and device, method includes: to obtain video to be detected；Motion detection is carried out to video, obtains the gait feature sequence of pedestrian；Gait sequence network model is obtained, by gait feature sequence inputting to gait sequence network model, obtains the recognition result and identification probability for pedestrian；Judge whether identification probability is greater than preset threshold；If more than determining that resulting recognition result is correct, using recognition result as search result；If being not more than, determine that resulting recognition result is incorrect, returns and execute acquisition video to be detected.Using the embodiment of the present invention, portrait effectiveness of retrieval and accuracy are improved.
Description
Technical field
The present invention relates to field of data retrieval more particularly to a kind of portrait search methods and device.
Background technique
With the development of internet technology, various video datas are in explosive growth, in order to quickly from massive video number
The relevant information of some personage is retrieved in, various portrait search methods are applied and given birth to.
Currently, portrait search method mainly uses manual identified method or face recognition technology, to the portrait in video
It is retrieved, obtains search result.But these methods have that efficiency is lower or accuracy rate is not high, are still difficult to meet
User's actual need.For example, the size of video to be retrieved may have several hundred T (Trillionbyte, terabyte), by artificial
Mode identify may one or two months, workload is huge and takes a long time；And although face recognition technology processing speed is very fast,
It is to be easy to be interfered by scene due to the technology and stringent to the size requirements of face, and the scene in video usually changes greatly,
And the face size of same personage in video may change, therefore the accurate of portrait is retrieved using the technology in video
Rate is not high.
It is therefore desirable to design a kind of new portrait search method, to overcome the above problem.
Summary of the invention
It is an object of the invention to overcome the defect of the prior art, a kind of portrait search method and device are provided, with reality
Now improve portrait effectiveness of retrieval and accuracy.
The present invention is implemented as follows:
In a first aspect, the present invention provides a kind of portrait search method, which comprises
Obtain video to be detected；Motion detection is carried out to the video, obtains the gait feature sequence of pedestrian；
Gait sequence network model is obtained to obtain by the gait feature sequence inputting to the gait sequence network model
To the recognition result and identification probability for being directed to the pedestrian；
Judge whether the identification probability is greater than preset threshold；
If more than determining that resulting recognition result is correct, using the recognition result as search result；
If being not more than, determine that resulting recognition result is incorrect, returns and execute the step of obtaining video to be detected.
Optionally, motion detection is carried out to pedestrian movement's video, obtains the gait feature sequence of pedestrian, comprising:
Using preset motion detection algorithm, each picture frame comprising pedestrian is detected from video, and from including pedestrian
Each picture frame in extract pedestrian gait feature；Extracted each gait feature is merged, the gait feature sequence of pedestrian is obtained
Column.
Optionally, the gait sequence network model is target nerve network model, obtains gait sequence network model, packet
It includes:
With the preset initial neural network model of training sample set training, the target nerve network model is obtained.
Optionally, the recognition result includes identification result and movement recognition result, and the preset threshold includes pre-
If first threshold and default second threshold, the identification probability includes identification probability and movement identification probability；Described in judgement
Whether identification probability is greater than preset threshold, if more than determining that resulting recognition result is correct, using the recognition result as retrieval
As a result, comprising:
When the identification probability is greater than preset first threshold value and movement identification probability is greater than default second threshold, sentence
Fixed resulting recognition result is correct, using the identification result and the movement recognition result as search result.
Optionally, if identification probability is not more than preset threshold, determine that resulting recognition result is incorrect, return executes acquisition
The step of video of pedestrian, comprising:
When the identification probability is not more than default second threshold no more than preset first threshold value or movement identification probability
When, the step of determining that resulting recognition result is incorrect, return to the video for executing acquisition pedestrian.
Optionally, the initial neural network model is LSTM time recurrent neural networks model.
Optionally, when search result has it is multiple when, the method also includes:
According to the size of the identification probability of each search result, ascending order/descending is carried out to each search result and is arranged.
Second aspect, the present invention provide a kind of portrait retrieval device, and described device includes:
First obtains module, for obtaining video to be detected；Motion detection is carried out to the video, obtains the step of pedestrian
State characteristic sequence；
Second obtains module, for obtaining gait sequence network model, by the gait feature sequence inputting to the step
State sequence network model obtains the recognition result and identification probability for the pedestrian；
Judgment module, for judging whether the identification probability is greater than preset threshold；If more than the resulting identification knot of judgement
Fruit is correct, using the recognition result as search result；If being not more than, determine that resulting recognition result is incorrect, returns and execute
Obtain video to be detected.
Optionally, described first module is obtained to pedestrian movement's video progress motion detection, obtain the gait of pedestrian
Characteristic sequence, comprising:
Using preset motion detection algorithm, each picture frame comprising pedestrian is detected from video, and from including pedestrian
Each picture frame in extract pedestrian gait feature；Extracted each gait feature is merged, the gait feature sequence of pedestrian is obtained
Column.
Optionally, the gait sequence network model is target nerve network model, and the second acquisition module is walked
State sequence network model, specifically:
With the preset initial neural network model of training sample set training, the target nerve network model is obtained.
Optionally, the recognition result includes identification result and movement recognition result, and the preset threshold includes pre-
If first threshold and default second threshold, the identification probability includes identification probability and movement identification probability；The judgement
Module judges whether the identification probability is greater than preset threshold, if more than determining that resulting recognition result is correct, by the identification
As a result it is used as search result, specifically:
When the identification probability is greater than preset first threshold value and movement identification probability is greater than default second threshold, sentence
Fixed resulting recognition result is correct, using the identification result and the movement recognition result as search result.
Optionally, the judgment module determines resulting recognition result not just when identification probability is not more than preset threshold
Really, it returns to execute and obtains video to be detected, specifically:
When the identification probability is not more than default second threshold no more than preset first threshold value or movement identification probability
When, determine that resulting recognition result is incorrect, returns and execute acquisition video to be detected.
Optionally, the initial neural network model is LSTM time recurrent neural networks model.
Optionally, described device further include:
Sorting module, for when search result has multiple, according to the size of the identification probability of each search result, to each inspection
Hitch fruit carries out ascending order/descending arrangement.
The invention has the following advantages: obtaining video to be detected first using the embodiment of the present invention；To video into
Row motion detection obtains the gait feature sequence of pedestrian；In turn, gait sequence network model is obtained, gait feature sequence is defeated
Enter the recognition result and identification probability obtained to gait sequence network model for pedestrian；It is pre- to judge whether identification probability is greater than
If threshold value；If more than determining that resulting recognition result is correct, using recognition result as search result；If being not more than, gained is determined
Recognition result it is incorrect, return to execute and obtain video to be detected.
As it can be seen that obtaining gait feature sequence inputting to gait sequence network model for row using the embodiment of the present invention
The recognition result and identification probability of people improves recall precision, and when identification is general for existing manual identified mode
When rate is greater than preset threshold, using recognition result as search result；When identification probability is not more than preset threshold, then execution is returned
The step of obtaining video to be detected, to re-start retrieval to pedestrian, therefore, improve the accuracy rate of search result.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is a kind of flow diagram of portrait search method provided in an embodiment of the present invention；
Fig. 2 is a kind of structural schematic diagram that portrait provided in an embodiment of the present invention retrieves device.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts all other
Embodiment shall fall within the protection scope of the present invention.
It should be noted that portrait search method provided by the present invention can be applied to electronic equipment, wherein specific
In, which can be computer, PC, plate, mobile phone etc., this is all reasonable.
Referring to Fig. 1, the embodiment of the present invention provides a kind of portrait search method, and method includes the following steps:
S101, video to be detected is obtained；The video is handled, the gait feature sequence of pedestrian is obtained；
Video to be detected can be the video that video capture device acquires in real time, be also possible to be pre-stored within the present invention
The video of executing subject (such as electronic equipment) can also be the video that third party device provides.Video capture device can be
Video camera, video recorder etc., the present invention such as can be monocular-camera without limitation to the concrete model of video capture device,
It may be binocular camera.Video capture device, which can be fixedly mounted on, touches a position；Some motive objects can also be installed on
On body, such as on unmanned plane, automobile.
Video capture device can acquire data and obtain video, and video can be sent to electronic equipment, thus electronics
Equipment can obtain the video of video capture device acquisition, and can handle video, obtain the gait feature sequence of pedestrian
Column.
Specifically, the mode for being handled video to obtain the gait feature sequence of pedestrian can be with are as follows:
Using preset motion detection algorithm, each picture frame comprising pedestrian is detected from video, and from including pedestrian
Picture frame in extract pedestrian gait feature；Extracted each gait feature is merged, the gait feature sequence of pedestrian is obtained.
Video is made of continuous picture frame, it is believed that and it is continuous image frame sequence, using motion detection algorithm,
Each picture frame comprising pedestrian can be detected from video, and can extract the gait feature of pedestrian.Specific motion detection
Algorithm can be one of background image calculus of finite differences, frame differential method and optical flow method etc. or combination.
Gait feature is used to reflect feature when pedestrian movement, may include the features such as step-length, stride, cadence.It can be from
Gait feature is extracted in picture frame comprising pedestrian, extracted gait feature is merged, available gait feature sequence.Step
State characteristic sequence has uniqueness, can uniquely distinguish pedestrian and its motion state.The motion state of pedestrian includes hovering, slowly
It walks, normal walking, running, tumble etc..
In addition, improving recognition accuracy to eliminate the noise in video, motion detection is being carried out to video, is being gone
Before the gait feature sequence of people, method can also include:
Video is filtered.
Correspondingly, carrying out motion detection to video in step S101, the gait feature sequence of pedestrian is obtained, it can be with are as follows:
Motion detection is carried out to the video after filtering processing, obtains the gait feature sequence of pedestrian.
Since the video of video capture device acquisition is there may be noise jamming, motion detection is being carried out to video
Before, video is filtered, it is possible to reduce unnecessary noise jamming in video improves the clarity of video.
The embodiment of the present invention to the implementation of filtering processing without limitation, for example, can be using median filtering, linear filter
One of filtering algorithms such as wave, Kalman filtering or combination, are filtered video.
As it can be seen that using the embodiment of the present invention after being filtered to video, the noise in video can be removed, to filter
Wave treated video carries out motion detection, helps to improve the accuracy of gait feature sequence.
In another implementation, in order to improve the speed of feature extraction, detecting to include each of pedestrian from video
After picture frame, method can also include:
The processing of image Skeleton is carried out to each picture frame comprising pedestrian, obtains skeletonized images sequence；
Correspondingly, the gait feature of pedestrian is extracted from the picture frame comprising pedestrian, specifically:
The gait feature of pedestrian is extracted from skeletonized images sequence.
The refinement to picture frame may be implemented in the processing of image Skeleton, i.e., removes from original figure some unessential
Point, so as to obtain the skeleton of objects in images.The point removed does not influence the global shape of object, skeleton, it can be understood as
The axis of object, such as a rectangular skeleton are the central axes on its length direction；The skeleton of square is its center
Point；Round skeleton is its center of circle, and the skeleton of straight line is own, and the skeleton of isolated point is also itself.The bone of image is obtained
Frame is equivalent to the primary structure and shape information of prominent object, mentioning for pedestrian's gait feature may be implemented according to these information
It takes.
Using the embodiment of the present invention, due to eliminating redundant information, therefore feature extraction speed is improved.
S102, gait sequence network model is obtained, by the gait feature sequence inputting to the gait sequence network mould
Type obtains the recognition result and identification probability for the pedestrian；
Gait sequence network model can for neural network model, SVM (Support Vector Machine, support to
Amount machine) one of machine learning models such as model, genetic neural network model or combination.Gait sequence network model is preparatory instruction
Practice sample set training to convergent machine learning model, thus, after obtaining gait feature sequence, can export recognition result with
Identification probability.Identification probability is for evaluating the probability for identifying correct result.
Specifically, gait sequence network model can be target nerve network model, gait sequence network model, packet are obtained
It includes:
With the preset initial neural network model of training sample set training, target nerve network model is obtained.
Training sample set can be the sample set for training initial neural network model, and each sample standard deviation includes gait
Characteristic sequence recognition result corresponding with its, the gait feature sequence that training sample is concentrated are adopted before video capture device
The video of collection, either, the video that the pre-stored video of electronic equipment or other third party devices provide, training sample
The recognition result of this concentration can be what expert marked in advance, be also possible to what other machines learning model prior learning came out.
Each parameter in initial neural network model is initial default parameters, available by instructing after training
The target nerve network model that model parameter after practicing maturation is constituted, model parameter determine the identification of target nerve network model
Accuracy.Initial neural network model is LSTM (Long Short-Term Memory, time recurrent neural networks model),
LSTM is the Recognition with Recurrent Neural Network after a kind of improvement, can remember long-term information, to solve the problems, such as to rely on for a long time, for place
The biggish data of information content are managed, there is preferable learning effect.
Certainly, in other implementations, initial neural network model can also be convolutional neural networks model, circulation mind
Through network model etc..
S103, judge whether identification probability is greater than preset threshold；If more than then executing S104；If being not more than, execute
S105；
S104, determine that resulting recognition result is correct, using the recognition result as search result；
S105, determine that resulting recognition result is incorrect, return and execute the step of obtaining video to be detected.
Recognition result may include identification result and movement recognition result, correspondingly, preset threshold may include pre-
If first threshold and default second threshold, identification probability includes identification probability and movement identification probability；Judge identification probability
Whether preset threshold is greater than, if more than, determine that resulting recognition result is correct, it, can using the recognition result as search result
With are as follows:
When the identification probability is greater than preset first threshold value and movement identification probability is greater than default second threshold, sentence
Fixed resulting recognition result is correct, using the identification result and the movement recognition result as search result.
Preset first threshold value and default second threshold can be preset according to demand, and the two can be identical, can also not
Together, for example, can be respectively 0.75 and 0.65, it is, being greater than when identification probability is greater than 0.75 and moves identification probability
When 0.65, determine that resulting recognition result is correct, using identification result and movement recognition result as search result.
Certainly, in other implementations, preset first threshold value or movement identification can also be greater than when identification probability
When probability is greater than default second threshold, determines that identification success or movement identification are correct, identification result or movement are known
Other result is as search result.
For example, preset first threshold value and default second threshold are respectively 0.75 and 0.65, when identification probability is greater than
When 0.75, identification success is determined, using identification result as search result；When moving identification probability greater than 0.65,
Determine that movement identification is correct, using movement recognition result as search result.
Alternatively, in another implementation, the purpose of Identification of Images is only that the identity of identification pedestrian, then recognition result
It can only include identification as a result, when identification probability is greater than preset threshold, determine that resulting recognition result is correct, it will
Identification result is as search result.
Or in another implementation, the purpose of Identification of Images is only that the movement of identification pedestrian, without concern
The identity of pedestrian, then recognition result can only include movement recognition result, when moving identification probability greater than preset threshold, determine
Resulting recognition result is correct, using identification result as search result.
Preset threshold can be previously set, and can be 0.6,0.65,0.7,0.75 etc..
Identification result can be used for the identity of unique identification pedestrian, and the particular content of identification result of the present invention is not
Limit, for example, can be the combination of the information such as name, identification card number, address of pedestrian, can also only include identification card number and
Name.Movement recognition result may include hover, be careful, normal walking, running, one of tumble etc..
In addition, determining that resulting recognition result is incorrect if identification probability is not more than preset threshold, returns and execute acquisition row
It the step of video of people, can be with are as follows:
When identification probability is not more than default second threshold no more than preset first threshold value or movement identification probability, sentence
Fixed resulting recognition result is incorrect, returns and executes the step of obtaining video to be detected.
Alternatively, in other implementations preset first threshold value and movement can also be not more than when identification probability
When identification probability is not more than default second threshold, determine that resulting recognition result is incorrect, returns and execute acquisition view to be detected
The step of frequency.
For example, can be respectively 0.75 and 0.65, it is, when identification probability is no more than 0.75 and movement identification is general
When rate is not more than 0.65, determine that resulting recognition result is incorrect.
As it can be seen that obtaining gait feature sequence inputting to gait sequence network model for row using the embodiment of the present invention
The recognition result and identification probability of people improves recall precision, and when identification is general for existing manual identified mode
When rate is greater than preset threshold, using recognition result as search result；When identification probability is not more than preset threshold, then execution is returned
Video to be detected is obtained, to re-start retrieval to pedestrian, therefore, improve the accuracy rate of search result.
In addition, after determining that resulting recognition result is incorrect, prompt letter can also be provided in order to improve user experience
Breath, the prompt information is for prompting the resulting recognition result of user incorrect, alternatively, the prompt information is also used to prompt the user be
No to receive the recognition result, if user selects to receive, electronic equipment can be using recognition result as search result；If user selects
It selects and does not receive, then electronic equipment can return to the step of execution obtains video to be detected.
Using the embodiment of the present invention, user can independently choose whether to receive recognition result, thus according to the user's choice
Search result is obtained, improves user experience.
When search result has it is multiple when, method further include:
According to the size of the identification probability of each search result, ascending order/descending is carried out to each search result and is arranged.
When in video including multiple pedestrians, for every a group traveling together, available recognition result and identification probability,
Identification probability can be greater than to each recognition result of preset threshold as search result, so that search result has multiple, retrieval knot
The identification probability of fruit is are as follows: the identification probability of the recognition result as the search result.
Alternatively, the pedestrian may be in different movement shapes in varied situations when only including a pedestrian in video
Identification probability can be greater than each recognition result of preset threshold as search result so that recognition result may also be multiple by state,
To search result have it is multiple.
Using the embodiment of the present invention, each search result ascending order/descending can be arranged, consequently facilitating subsequent check and divide
Analysis.
Corresponding with above-mentioned embodiment of the method, the embodiment of the present invention also provides a kind of portrait retrieval device.
Referring to fig. 2, Fig. 2 is a kind of structural schematic diagram of portrait retrieval device, device packet provided by the embodiment of the present invention
It includes:
First obtains module 201, for obtaining video to be detected；Motion detection is carried out to the video, obtains pedestrian
Gait feature sequence；
Second obtains module 202, for obtaining gait sequence network model, by the gait feature sequence inputting to described
Gait sequence network model obtains the recognition result and identification probability for the pedestrian；
Judgment module 203, for judging whether the identification probability is greater than preset threshold；If more than determining resulting knowledge
Other result is correct, using the recognition result as search result；If being not more than, determine that resulting recognition result is incorrect, returns
It executes and obtains video to be detected.
As it can be seen that obtaining gait feature sequence inputting to gait sequence network model for row using the embodiment of the present invention
The recognition result and identification probability of people improves recall precision, and when identification is general for existing manual identified mode
When rate is greater than preset threshold, using recognition result as search result；When identification probability is not more than preset threshold, then execution is returned
Video to be detected is obtained, to re-start retrieval to pedestrian, therefore, improve the accuracy rate of search result.
Optionally, described first module is obtained to pedestrian movement's video progress motion detection, obtain the gait of pedestrian
Characteristic sequence, comprising:
Using preset motion detection algorithm, each picture frame comprising pedestrian is detected from video, and from including pedestrian
Each picture frame in extract pedestrian gait feature；Extracted each gait feature is merged, the gait feature sequence of pedestrian is obtained
Column.
Optionally, the gait sequence network model is target nerve network model, and the second acquisition module is walked
State sequence network model, specifically:
With the preset initial neural network model of training sample set training, the target nerve network model is obtained.
Optionally, the recognition result includes identification result and movement recognition result, and the preset threshold includes pre-
If first threshold and default second threshold, the identification probability includes identification probability and movement identification probability；The judgement
Module judges whether the identification probability is greater than preset threshold, if more than determining that resulting recognition result is correct, by the identification
As a result it is used as search result, specifically:
When the identification probability is greater than preset first threshold value and movement identification probability is greater than default second threshold, sentence
Fixed resulting recognition result is correct, using the identification result and the movement recognition result as search result.
Optionally, the judgment module determines resulting recognition result not just when identification probability is not more than preset threshold
Really, it returns to execute and obtains video to be detected, specifically:
When the identification probability is not more than default second threshold no more than preset first threshold value or movement identification probability
When, determine that resulting recognition result is incorrect, returns and execute acquisition video to be detected.
Optionally, the initial neural network model is LSTM time recurrent neural networks model.
Optionally, described device further include:
Sorting module, for when search result has multiple, according to the size of the identification probability of each search result, to each inspection
Hitch fruit carries out ascending order/descending arrangement.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of portrait search method, which is characterized in that the described method includes:
Obtain video to be detected；Motion detection is carried out to the video, obtains the gait feature sequence of pedestrian；
It obtains gait sequence network model and obtains needle by the gait feature sequence inputting to the gait sequence network model
To the recognition result and identification probability of the pedestrian；
Judge whether the identification probability is greater than preset threshold；
If more than determining that resulting recognition result is correct, using the recognition result as search result；
If being not more than, determine that resulting recognition result is incorrect, returns and execute the step of obtaining video to be detected.
2. being obtained the method according to claim 1, wherein carrying out motion detection to pedestrian movement's video
The gait feature sequence of pedestrian, comprising:
Using preset motion detection algorithm, each picture frame comprising pedestrian is detected from video, and from including each of pedestrian
The gait feature of pedestrian is extracted in picture frame；Extracted each gait feature is merged, the gait feature sequence of pedestrian is obtained.
3. the method according to claim 1, wherein the gait sequence network model is target nerve network mould
Type obtains gait sequence network model, comprising:
With the preset initial neural network model of training sample set training, the target nerve network model is obtained.
4. method according to claim 1-3, which is characterized in that the recognition result includes identification result
With movement recognition result, the preset threshold includes preset first threshold value and default second threshold, and the identification probability includes body
Part identification probability and movement identification probability；Judge whether the identification probability is greater than preset threshold, if more than resulting knowledge is determined
Other result is correct, using the recognition result as search result, comprising:
When the identification probability is greater than preset first threshold value and movement identification probability is greater than default second threshold, institute is determined
The recognition result obtained is correct, using the identification result and the movement recognition result as search result.
5. according to the method described in claim 4, it is characterized in that, determining resulting if identification probability is not more than preset threshold
The step of recognition result is incorrect, returns to the video for executing acquisition pedestrian, comprising:
When the identification probability is not more than default second threshold no more than preset first threshold value or movement identification probability, sentence
The step of fixed resulting recognition result is incorrect, returns to the video for executing acquisition pedestrian.
6. according to the method described in claim 3, it is characterized in that, the initial neural network model is LSTM time recurrence mind
Through network model.
7. the method according to claim 1, wherein when search result has multiple, the method also includes:
According to the size of the identification probability of each search result, ascending order/descending is carried out to each search result and is arranged.
8. a kind of portrait retrieves device, which is characterized in that described device includes:
First obtains module, for obtaining video to be detected；Motion detection is carried out to the video, the gait for obtaining pedestrian is special
Levy sequence；
Second obtains module, for obtaining gait sequence network model, by the gait feature sequence inputting to the gait sequence
Column network model obtains the recognition result and identification probability for the pedestrian；
Judgment module, for judging whether the identification probability is greater than preset threshold；If more than determining resulting recognition result just
Really, using the recognition result as search result；If being not more than, determine that resulting recognition result is incorrect, return executes acquisition
Video to be detected.
9. device according to claim 8, which is characterized in that it is described first obtain module to pedestrian movement's video into
Row motion detection obtains the gait feature sequence of pedestrian, specifically:
Using preset motion detection algorithm, each picture frame comprising pedestrian is detected from video, and from including each of pedestrian
The gait feature of pedestrian is extracted in picture frame；Extracted each gait feature is merged, the gait feature sequence of pedestrian is obtained.
10. device according to claim 8, which is characterized in that the gait sequence network model is target nerve network
Model, described second, which obtains module, obtains gait sequence network model, specifically:
With the preset initial neural network model of training sample set training, the target nerve network model is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811091048.2A CN109359544B (en) | 2018-09-19 | 2018-09-19 | Portrait retrieval method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811091048.2A CN109359544B (en) | 2018-09-19 | 2018-09-19 | Portrait retrieval method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109359544A true CN109359544A (en) | 2019-02-19 |
| CN109359544B CN109359544B (en) | 2022-01-21 |
Family
ID=65351305
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811091048.2A Active CN109359544B (en) | 2018-09-19 | 2018-09-19 | Portrait retrieval method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109359544B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113821689A (en) * | 2021-09-22 | 2021-12-21 | 沈春华 | Pedestrian retrieval method and device based on video sequence and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140261887A1 (en) * | 2011-11-03 | 2014-09-18 | Thales Nederland B.V. | System for characterizing motion of an individual, notably a human individual, and associated method |
| CN105354548A (en) * | 2015-10-30 | 2016-02-24 | 武汉大学 | Surveillance video pedestrian re-recognition method based on ImageNet retrieval |
| CN107590452A (en) * | 2017-09-04 | 2018-01-16 | 武汉神目信息技术有限公司 | A kind of personal identification method and device based on gait and face fusion |
| CN108520216A (en) * | 2018-03-28 | 2018-09-11 | 电子科技大学 | A kind of personal identification method based on gait image |
-
2018
- 2018-09-19 CN CN201811091048.2A patent/CN109359544B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140261887A1 (en) * | 2011-11-03 | 2014-09-18 | Thales Nederland B.V. | System for characterizing motion of an individual, notably a human individual, and associated method |
| CN105354548A (en) * | 2015-10-30 | 2016-02-24 | 武汉大学 | Surveillance video pedestrian re-recognition method based on ImageNet retrieval |
| CN107590452A (en) * | 2017-09-04 | 2018-01-16 | 武汉神目信息技术有限公司 | A kind of personal identification method and device based on gait and face fusion |
| CN108520216A (en) * | 2018-03-28 | 2018-09-11 | 电子科技大学 | A kind of personal identification method based on gait image |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113821689A (en) * | 2021-09-22 | 2021-12-21 | 沈春华 | Pedestrian retrieval method and device based on video sequence and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109359544B (en) | 2022-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106339428B (en) | Suspect's personal identification method and device based on video big data | |
| CN106651916B (en) | A kind of positioning and tracing method and device of target | |
| CN105574543B (en) | A kind of vehicle brand type identifier method and system based on deep learning | |
| CN105488468B (en) | A kind of localization method and device of target area | |
| CN110136199A (en) | A kind of vehicle location based on camera, the method and apparatus for building figure | |
| CN104915351B (en) | Picture sort method and terminal | |
| CN106127788B (en) | A kind of vision barrier-avoiding method and device | |
| CN108009466B (en) | Pedestrian detection method and device | |
| CN102243687A (en) | Physical education teaching auxiliary system based on motion identification technology and implementation method of physical education teaching auxiliary system | |
| Chaudhry et al. | Design of a mobile face recognition system for visually impaired persons | |
| CN108960124B (en) | Image processing method and device for pedestrian re-identification | |
| CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
| CN107977656A (en) | A kind of pedestrian recognition methods and system again | |
| CN105808610A (en) | Internet picture filtering method and device | |
| CN106909886A (en) | A kind of high accuracy method for traffic sign detection and system based on deep learning | |
| CN109766873A (en) | A kind of pedestrian mixing deformable convolution recognition methods again | |
| CN115713715B (en) | Human behavior recognition method and recognition system based on deep learning | |
| CN110210550A (en) | Image fine granularity recognition methods based on integrated study strategy | |
| CN114445780A (en) | Detection method and device for bare soil covering, and training method and device for recognition model | |
| CN109670517A (en) | Object detection method, device, electronic equipment and target detection model | |
| CN111191531A (en) | Rapid pedestrian detection method and system | |
| CN109284700A (en) | The method of multiple Face datections, storage medium, equipment and system in image | |
| CN113762049B (en) | Content identification method, content identification device, storage medium and terminal equipment | |
| CN110163264A (en) | A kind of walking mode recognition methods based on machine learning | |
| CN109359544A (en) | A kind of portrait search method and device |
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 |