CN109409250A - A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning - Google Patents
A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning Download PDFInfo
- Publication number
- CN109409250A CN109409250A CN201811168809.XA CN201811168809A CN109409250A CN 109409250 A CN109409250 A CN 109409250A CN 201811168809 A CN201811168809 A CN 201811168809A CN 109409250 A CN109409250 A CN 109409250A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- identified
- deep learning
- video camera
- recognition methods
- 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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant 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
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
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)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to pedestrian identification technology fields again, more particularly to a kind of across the video camera pedestrian of non-overlapping ken recognition methods again based on deep learning, it trains pedestrian's attitude mode, background segment model and Feature Selection Model first, after the pretreatment for carrying out the filtering of pedestrian's posture, background removal and unitary of illumination using pedestrian's picture of the model to monitor video network, it recycles Feature Selection Model to automatically extract the further feature of pedestrian, completes pedestrian finally by characteristic matching and identify again.The method understands pedestrian's further feature have better robustness to changing factors such as illumination, shooting angle, scale, pedestrian's postures, can effectively realize pedestrian and identify again using the mechanism that deep learning network imitates human brain.
Description
Technical field
The invention belongs to pedestrian identification technology fields again, and in particular to a kind of no overlap ken based on deep learning is across taking the photograph
Camera pedestrian recognition methods again is mainly used for the process video monitoring of handling a case of local police station.
Background technique
As public security is handled a case process management standardization and the development of electronic trend and the general of area's video surveillance network of handling a case
And people's police need the monitoring from area's different zones of handling a case (such as information collection room, query room, marquis ask room) after the completion of handling a case
In video, activity trajectory video of same suspect during handling a case is picked out to form case videograph.This case
The generation of part videograph needs to handle the monitor video data of magnanimity, and existing hand picking method expends a large amount of manpowers, object
Power and time, it is difficult to meet the needs of handling a case process management.It handles a case the working efficiency of process, proposes to improve public security police
The automatic interception spelling of video realized by the tracking to suspect's activity trajectory for public security area's exploitation set of system of handling a case
It connects.The system includes pedestrian detection and pedestrian recognizer again, develops a pedestrian again primarily directed to this application scenarios herein
Recognizer.
Identification technology utilizes a given target pedestrian image to pedestrian again, by way of vision comparison, from non-overlap
Target pedestrian is identified in the video image of multiple video cameras of the ken.By pedestrian, identification technology, video surveillance network can again
Multiple-camera is realized to the cooperative monitoring of specific pedestrian, completes the view for belonging to same a group traveling together of the video camera shooting under different scenes
The auto-associating of frequency segment.Identification technology mainly passes through the shallow-layers features such as extraction pedestrian's clothing color, texture to existing pedestrian again
Similarity measurement analysis is carried out, pedestrian is completed and identifies again.Such method defect is more obvious, is easy to be done by local police station
The shooting visual angle of case area monitor camera, scale, ambient lighting, pedestrian's dress ornament, pedestrian's posture, the shadow for the changing factors such as blocking
It rings, causes pedestrian to identify poor robustness again, the requirement that local police station intercepts generation to case ancestor video automatically cannot be reached.
The invention of Publication No. " CN104268583A " proposes a kind of pedestrian based on the color region feature side of identification again
Method is easy to be influenced by ambient lighting, and when the clothing color of target pedestrian is similar to the clothing color of other pedestrians,
This method is easy other pedestrians being mistakenly identified as target pedestrian.
The patent of invention of Publication No. " CN106250870A " proposes a kind of joint part and global similarity measurement
3 kinds of color histogram features and 2 kinds of Texture similarity features are formed 6 kinds of color and vein spies by the pedestrian of habit recognition methods again
Sign, 6 kinds of color and vein features of joint part and global image obtain measuring similarity matrix by training, utilize the matrix
The similitude of pedestrian to be identified Yu target pedestrian are measured, pedestrian is completed and identifies again.This method is easy by pedestrian's posture and bat
The influence of angle is taken the photograph, robustness is insufficient.
Summary of the invention
Across the video camera pedestrian of the no overlap ken recognition methods again based on deep learning that the purpose of the present invention is to propose to a kind of,
With solve select the shallow-layer feature sensitive to environmental change to identify pedestrian again in the prior art, and not to pedestrian image into
The effective pretreatment of row, recognition effect are easy by illumination, shooting angle, scale, pedestrian's posture, the changing factors such as block
It influences, poor robustness causes to be difficult to the problem of accurately identifying pedestrian again.
The invention is realized by the following technical scheme:
A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning, this method are divided into model training
Stage and pedestrian identify test phase again, specifically comprise the following steps:
(1) pedestrian's attitude mode, background segment model and Feature Selection Model are trained first；
(2) then pedestrian's posture is filtered using pedestrian's attitude mode, acquisition includes the pedestrian at front or the back side
Image is removed background using background segment model；
(3) it is automatically extracted using further feature of the Feature Selection Model to pedestrian；
(4) pedestrian is completed finally by characteristic matching to identify again.
Further, the step (1) from more scene videos of the non-overlapping ken specifically, using largely obtaining
Pedestrian's picture sample collection training convolutional neural networks, obtain pedestrian's attitude mode, background segment model and Feature Selection Model.
Further, the step (4) by the COS distance between further feature vector specifically, evaluated to be identified
The similarity of pedestrian image and target pedestrian image realizes characteristic matching, and setting similarity threshold values to judge pedestrian to be identified is
No is target pedestrian, is identified again to complete pedestrian.
Wherein, model training stage further comprises step:
S1: from the monitor video that each scene in area is handled a case by local police station, each appearance under the various angles of all kinds of scenes is collected
The pedestrian image of state is divided into training set sample and test sample；
S2: life is standardized according to the difference of foreground/background to the sample image vegetarian refreshments of the S1 training set sample collected
Name and mark, obtain the foreground/background classification of each pixel in every pedestrian sample figure, stroke the first training set sample；
S3: the first full convolutional network of training set sample off-line training completed is handled using S2, obtains background segment model；
S4: name is labeled and standardized according to the type of pedestrian's posture to the S1 training sample set collected, is gone
The affiliated posture classification of proper manners sheet forms the second training set sample；
S5: the second training set sample off-line training AlexNet network completed is handled using S4, obtains pedestrian's posture mould
Type；
S6: from the monitor video that each scene in area is handled a case by local police station, more scene pedestrian images of each pedestrian are collected
Set, after carrying out background process and illumination compensation process to it using the background segment model that S3 is obtained, as third
Training set sample；
S7: the third training set sample off-line training convolutional neural networks completed are handled using S6, obtain feature extraction mould
Type.
Preferably, the scene includes that information collection room, marquis ask room, query room and access way.
Wherein, pedestrian identifies that test phase further comprises step again:
S8: from public security handle a case area's monitor video video recording in selection standard test video, selected target pedestrian figure；
S9: the standard testing video and target pedestrian figure that input S8 chooses；
S10: pedestrian detection is carried out to test video and is gone forward side by side line trace；
S11: the pedestrian's attitude mode obtained using S5 carries out posture processing to pedestrian's graphic sequence to be identified that S10 is obtained,
It will scheme to import pedestrian's picture library to be identified by posture treated pedestrian；
S12: the target pedestrian of pedestrian's figure and S9 input to be identified is obtained to S11 using the background separation model obtained to S3
Figure carries out background and pre-processes；
S13: the pedestrian to be identified that S12 is obtained is schemed to carry out unitary of illumination processing using algorithm for image enhancement.
Further comprise step: S14: the Feature Selection Model obtained using S7 respectively locates target pedestrian figure and S13 in advance
Pedestrian to be identified figure and target pedestrian after reason scheme to carry out the feature extraction of pedestrian's deep layer.
Further comprise step: S15: two 512 dimensional feature vectors obtained to S14 carry out characteristic matching, with cosine away from
From both measurements similarity, the similarity between pedestrian to be identified and target pedestrian is exported.
Further comprise step: S16: setting similarity threshold values, it is to be identified if the similarity that S15 is obtained is greater than threshold values
Pedestrian is identified as target pedestrian again, and otherwise pedestrian to be identified is identified as non-targeted pedestrian again, the row to be identified that S11 is obtained
People's picture library is divided into target pedestrian picture library and non-targeted pedestrian's picture library.
The invention also includes a kind of non-volatile memory mediums comprising one or more computer instruction, described one
Or a plurality of computer instruction realizes above-mentioned across the video camera pedestrian of the no overlap ken based on the deep learning side of identification again when being executed
Method.
Compared with prior art, the present invention at least has the following beneficial effects or advantage:
This across the video camera pedestrian of the no overlap ken recognition methods again based on deep learning provided by the invention, utilizes depth
The filtering model of pedestrian's posture, pedestrian's background segment model and the depth characteristic that degree learning training obtains are extracted model and are regarded to monitoring
Pedestrian's picture in frequency network carries out pretreatment and identifies again, to determine if being target pedestrian, and then meets case video
Segment intercepts the demand of splicing automatically.The method carries out pedestrian's further feature using the mechanism that deep learning network imitates human brain
Understand that there is better robustness to changing factors such as illumination, shooting angle, scale, pedestrian's postures, pedestrian can be effectively realized
It identifies again.
Detailed description of the invention
The present invention is described in further details below with reference to attached drawing；
Fig. 1 is the overall flow figure of pedestrian of the invention recognition methods again；
Fig. 2 is background separation model training and test process flow chart of the invention；
Fig. 3 is the training of pedestrian's attitude mode and test process flow chart of the invention；
Fig. 4 is Feature Selection Model training and test process flow chart of the invention.
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 some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
For existing pedestrian, identification technology is difficult to complete the non-overlapping ken to identify problem, this hair again across the pedestrian of video camera again
It is bright to propose a kind of across the video camera pedestrian of non-overlapping ken recognition methods again based on deep learning, thought based on deep learning
Think, designs convolutional neural networks (Convolutional NeuralNetwork, CNN), using largely from the more of the non-overlapping ken
The pedestrian picture sample collection training CNN obtained in scene video, obtains pedestrian's attitude mode, background segment model and feature and mentions
Then modulus type successively carries out the filtering of pedestrian's posture to pedestrian's picture in monitor video network using these models, background is gone
Except operation and pedestrian's further feature automatically extract, row to be identified is evaluated finally by the COS distance between further feature vector
The similarity of people's image and target pedestrian image realizes characteristic matching, sets similarity threshold values whether to judge pedestrian to be identified
It is target pedestrian, is identified again to complete pedestrian.
According to being briefly discussed above, concrete scheme is as follows:
1.1 overall plan
A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning, overall plan can be divided into mould
Type training stage and pedestrian identify test phase again, as shown in Figure 1, itself specifically includes the following steps:
S1: from local police station handle a case each scene in area (information collection room, marquis ask room, query room, access way) prison
It controls in video, collects the pedestrian image of each posture under the various angles of all kinds of scenes, be divided into training set sample and test specimens
This；
S2: life is standardized according to the difference of foreground/background to the sample image vegetarian refreshments of the S1 training set sample collected
Name and mark, obtain the foreground/background classification of each pixel in every pedestrian sample figure, form the first training set sample；
S3: designing full convolutional network (Fully Convolutional Networks, FCN), handles completion using S2
First training set sample off-line training FCN network, obtains background segment model；
S4: name is labeled and standardized according to the type of pedestrian's posture to the S1 training sample set collected, is gone
The affiliated posture classification of proper manners sheet forms the second training set sample；
S5: the second training set sample off-line training AlexNet network completed is handled using S4, obtains pedestrian's posture mould
Type；
S6: from local police station handle a case each scene in area (information collection room, marquis ask room, query room, access way) prison
It controls in video, collects more scene pedestrian image set of each pedestrian, the background segment model obtained using S3 carry on the back to it
After scape processing and illumination compensation process, as third training set sample；
S7: the third training set sample off-line training convolutional neural networks completed are handled using S6, obtain feature extraction mould
Type；
S8: from public security handle a case area's monitor video video recording in selection standard test video, selected target pedestrian figure；
S9: the standard testing video and target pedestrian figure that input S8 chooses；
S10: pedestrian detection is carried out to test video and is gone forward side by side line trace；
S11: the pedestrian's attitude mode obtained using S5 carries out posture processing to pedestrian's graphic sequence to be identified that S10 is obtained,
It will scheme to import pedestrian's picture library to be identified by posture treated pedestrian；
S12: the target pedestrian of pedestrian's figure and S9 input to be identified is obtained to S11 using the background separation model obtained to S3
Figure carries out background and pre-processes；
S13: the pedestrian to be identified that S12 is obtained is schemed to carry out unitary of illumination processing using algorithm for image enhancement.
S14: the Feature Selection Model obtained using S7 is respectively to target pedestrian figure and the pretreated pedestrian to be identified of S13
Figure and target pedestrian scheme to carry out the feature extraction of pedestrian's deep layer.
S15: carrying out characteristic matching to two 512 dimensional feature vectors that S14 is obtained, similar with both COS distance measurements
Degree, exports the similarity between pedestrian to be identified and target pedestrian.
S16: setting similarity threshold values, if the similarity that S15 is obtained is greater than threshold values, pedestrian to be identified is identified as mesh again
Pedestrian is marked, otherwise pedestrian to be identified is identified as non-targeted pedestrian again, and pedestrian's picture library to be identified that S11 is obtained is divided into target line
People's picture library and non-targeted pedestrian's picture library.
1.2 background segment model trainings and test process
It handles a case from local police station and collects and mark the pedestrian sample of each scene in area's monitor video, utilize what is marked
Sample off-line training and testing background parted pattern, detailed process such as Fig. 2.
1.3 pedestrian's attitude mode training and test process
It handles a case from local police station and collects and mark the pedestrian sample of each scene in area's monitor video, utilize what is marked
Sample off-line training and test pedestrian's attitude mode, detailed process such as Fig. 3.
The training of 1.4 Feature Selection Models and test process
It handles a case from local police station and collects and mark the pedestrian sample of each scene in area's monitor video, utilize what is marked
Sample off-line training and test feature extract model, detailed process such as Fig. 4.
1.5 experiments and result
This programme experiment case study using local police station handle a case each scene in area monitor video record a video, mentioned using the present invention
Across the video camera pedestrian of the no overlap ken recognition methods again based on deep learning out, to the row occurred in multiple scene videos
People is identified again, counts the result that pedestrian identifies again.Belong to mesh by being counted in the picture library for being identified as target pedestrian respectively
The pedestrian's map number for marking pedestrian and non-targeted pedestrian, obtains that the results are shown in Table 1:
The no overlap ken across video camera pedestrian of the table 1 based on deep learning recognition methods test result again
As shown in Table 1, recognition methods obtains 94% to across the video camera pedestrian of the no overlap ken based on deep learning again
Thus accuracy rate illustrates validity that process proposed herein is identified again in across the video camera pedestrian of the no overlap ken and accurate
Property；When the hardware configuration such as table 2 of algorithm, the pedestrian of every pedestrian's picture identifies that average time-consuming is 134ms again, substantially meets reality
The demand of scene application.
2 pedestrian of table identifies test hardware configuration used again
In conclusion automatically generating demand to meet local police station's area's case ancestor's videograph of handling a case, the present invention is proposed
A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning is handled a case each scene in area from local police station first
Under monitor video obtain pedestrian's picture samples of a large amount of more scene multi-poses, establish picture sample collection, utilize picture sample collection
Training CNN, obtains pedestrian's attitude mode, background segment model and Feature Selection Model, then successively right using these models
Pedestrian's picture in monitor video network carries out the pretreatment operations such as posture filtering, background removal and unitary of illumination and pedestrian is special
Sign automatically extracts, and evaluates pedestrian image to be identified and target pedestrian image finally by the COS distance between feature vector
Similarity realizes characteristic matching, sets similarity threshold values to judge whether pedestrian to be identified is target pedestrian, to efficiently accomplish
Across the video camera pedestrian of the no overlap ken identifies again.Method proposed by the invention can effectively realize the no overlap ken across video camera
Pedestrian identify again, greatly reduce the misrecognition of target pedestrian, and real-time and robustness are all higher, real field can be met
The application demand of scape.
A kind of recognition methods again of the pedestrian based on color region feature that prior art CN104268583A is mentioned is easy
It is influenced by ambient lighting, and is easy for target pedestrian to be mistakenly identified as pedestrian similar in clothing color.The prior art
Pedestrian's recognition methods again of a kind of joint that CN106250870A is mentioned part and global similarity measurement study, be easy by
The influence of pedestrian's posture and camera angle, robustness are insufficient, can not adapt to local police station and handle a case more shooting angle in area, more
The variation of pedestrian's posture.
This programme is based on deep learning thought, schemes training pedestrian across the pedestrian of scene multi-pose first with the no overlap ken
Attitude mode, background segment model and Feature Selection Model carry out pedestrian using pedestrian picture of the model to monitor video network
After posture filtering and the pretreatment of unitary of illumination, recycles Feature Selection Model to automatically extract the further feature of pedestrian, pass through
Characteristic matching is completed pedestrian and is identified again.The further feature that the present invention implicitly extracts is that CNN imitates the mechanism of human brain to pedestrian's feature
Automatic understanding, which is different from the shallow-layer feature invented explicitly extract in the prior art, has better adaptability
With feature representation ability.In practical applications, the accuracy rate and real-time that this method identifies pedestrian again are high, to illumination, shooting
The robustness of the factors such as angle, scale, pedestrian's posture variation is good, can there is that meet local police station's area's case ancestor's video of handling a case well automatic
Intercept the demand generated.
The invention also includes a kind of non-volatile memory mediums comprising one or more computer instruction, described one
Or a plurality of computer instruction realizes above-mentioned across the video camera pedestrian of the no overlap ken based on the deep learning side of identification again when being executed
Method.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
Describe in detail it is bright, it should be understood that the above is only a specific embodiment of the present invention, the guarantor being not intended to limit the present invention
Protect range.Without departing from the spirit and scope of the invention, any modification, equivalent substitution, improvement and etc. done also belong to this
Within the protection scope of invention.
Claims (10)
1. a kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning, which is characterized in that this method point
Test phase is identified again for model training stage and pedestrian, is specifically comprised the following steps:
(1) pedestrian's attitude mode, background segment model and Feature Selection Model are trained first；
(2) then pedestrian's posture being filtered using pedestrian's attitude mode, acquisition includes the pedestrian image at front or the back side,
Background is removed using background segment model；
(3) it is automatically extracted using further feature of the Feature Selection Model to pedestrian；
(4) pedestrian is completed finally by characteristic matching to identify again.
2. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 1 based on deep learning, special
Sign is, the step (1) specifically, utilize the pedestrian's picture largely obtained from more scene videos of the non-overlapping ken
Sample set training convolutional neural networks obtain pedestrian's attitude mode, background segment model and Feature Selection Model.
3. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 1 based on deep learning, special
Sign is, the step (4) specifically, by the COS distance between further feature vector evaluate pedestrian image to be identified with
The similarity of target pedestrian image realizes characteristic matching, sets similarity threshold values to judge whether pedestrian to be identified is target line
People identifies again to complete pedestrian.
4. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 1 based on deep learning, special
Sign is that model training stage further comprises step:
S1: from the monitor video that each scene in area is handled a case by local police station, each posture under the various angles of all kinds of scenes is collected
Pedestrian image is divided into training set sample and test set sample；
S2: being labeled the S1 training set sample collected according to the difference of foreground/background, obtains each in every pedestrian sample figure
The foreground/background classification of pixel forms the first training set sample；
S3: the first full convolutional network of training set sample off-line training completed is handled using S2, obtains background segment model；
S4: name is labeled and standardized according to the type of pedestrian's posture to the S1 training sample set collected, obtains pedestrian's sample
This affiliated posture classification forms the second training set sample；
S5: the second training set sample off-line training AlexNet network completed is handled using S4, obtains pedestrian's attitude mode；
S6: from the monitor video that each scene in area is handled a case by local police station, collecting more scene pedestrian image set of each pedestrian,
After the background segment model obtained using S3 carries out background process and illumination compensation process to the pedestrian image detected, by it
As third training set sample；
S7: the third training set sample off-line training convolutional neural networks completed are handled using S6, obtain Feature Selection Model.
5. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 4 based on deep learning, special
Sign is, the scene includes that information collection room, marquis ask room, query room and access way.
6. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 4 based on deep learning, special
Sign is that pedestrian identifies that test phase further comprises step again:
S8: from public security handle a case area's monitor video video recording in selection standard test video, selected target pedestrian figure；
S9: the standard testing video and target pedestrian figure that input S8 chooses；
S10: pedestrian detection is carried out to test video and is gone forward side by side line trace；
S11: the pedestrian's attitude mode obtained using S5, which will carry out posture processing to pedestrian's graphic sequence to be identified that S10 is obtained, to be passed through
Posture treated pedestrian schemes to import pedestrian's picture library to be identified；
S12: using the background separation model that S3 is obtained to S11 obtain the target pedestrian figure of pedestrian to be identified figure and S9 input into
Row goes background to pre-process；
S13: the pedestrian to be identified that S12 is obtained is schemed to carry out unitary of illumination processing using algorithm for image enhancement.
7. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 6 based on deep learning, special
Sign is, further comprises step:
S14: the Feature Selection Model obtained using S7 respectively pedestrian's figure to be identified pretreated to target pedestrian figure and S13 and
Target pedestrian schemes to carry out the feature extraction of pedestrian's deep layer.
8. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 7 based on deep learning, special
Sign is, further comprises step:
S15: carrying out characteristic matching to two 512 dimensional feature vectors that S14 is obtained, defeated with COS distance measurement similarity
Similarity between pedestrian to be identified and target pedestrian out.
9. across the video camera pedestrian of the no overlap ken recognition methods again according to claim 8 based on deep learning, special
Sign is, further comprises step:
S16: setting similarity threshold values, if the similarity that S15 is obtained is greater than threshold values, pedestrian to be identified is identified as target line again
People, otherwise pedestrian to be identified is identified as non-targeted pedestrian again, and pedestrian's picture library to be identified that S11 is obtained is divided into target pedestrian figure
Library and non-targeted pedestrian's picture library.
10. a kind of non-volatile memory medium, which is characterized in that including one or more computer instruction, described one or more
Computer instruction realizes the described in any item methods of the claims 1-9 when being executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811168809.XA CN109409250A (en) | 2018-10-08 | 2018-10-08 | A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811168809.XA CN109409250A (en) | 2018-10-08 | 2018-10-08 | A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109409250A true CN109409250A (en) | 2019-03-01 |
Family
ID=65466834
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811168809.XA Pending CN109409250A (en) | 2018-10-08 | 2018-10-08 | A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109409250A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919141A (en) * | 2019-04-09 | 2019-06-21 | 广东省智能制造研究所 | A kind of recognition methods again of the pedestrian based on skeleton pose |
| CN109934177A (en) * | 2019-03-15 | 2019-06-25 | 艾特城信息科技有限公司 | Pedestrian recognition methods, system and computer readable storage medium again |
| CN111460977A (en) * | 2020-03-30 | 2020-07-28 | 广东电网有限责任公司电力科学研究院 | Cross-vision person re-identification method, device, terminal and storage medium |
| CN111897993A (en) * | 2020-07-20 | 2020-11-06 | 杭州叙简科技股份有限公司 | Efficient target person track generation method based on pedestrian re-recognition |
| CN113435443A (en) * | 2021-06-28 | 2021-09-24 | 中国兵器装备集团自动化研究所有限公司 | Method for automatically identifying landmark from video |
| CN114820724A (en) * | 2022-06-28 | 2022-07-29 | 浙江树人学院 | Intelligent monitoring method and system for cross-scene tracking |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7711146B2 (en) * | 2006-03-09 | 2010-05-04 | General Electric Company | Method and system for performing image re-identification |
| CN105518744A (en) * | 2015-06-29 | 2016-04-20 | 北京旷视科技有限公司 | Pedestrian re-identification method and equipment |
| CN105574515A (en) * | 2016-01-15 | 2016-05-11 | 南京邮电大学 | Pedestrian re-identification method in zero-lap vision field |
| CN107153817A (en) * | 2017-04-29 | 2017-09-12 | 深圳市深网视界科技有限公司 | Pedestrian's weight identification data mask method and device |
| CN107704838A (en) * | 2017-10-19 | 2018-02-16 | 北京旷视科技有限公司 | The attribute recognition approach and device of destination object |
| CN108537136A (en) * | 2018-03-19 | 2018-09-14 | 复旦大学 | The pedestrian's recognition methods again generated based on posture normalized image |
-
2018
- 2018-10-08 CN CN201811168809.XA patent/CN109409250A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7711146B2 (en) * | 2006-03-09 | 2010-05-04 | General Electric Company | Method and system for performing image re-identification |
| CN105518744A (en) * | 2015-06-29 | 2016-04-20 | 北京旷视科技有限公司 | Pedestrian re-identification method and equipment |
| CN105574515A (en) * | 2016-01-15 | 2016-05-11 | 南京邮电大学 | Pedestrian re-identification method in zero-lap vision field |
| CN107153817A (en) * | 2017-04-29 | 2017-09-12 | 深圳市深网视界科技有限公司 | Pedestrian's weight identification data mask method and device |
| CN107704838A (en) * | 2017-10-19 | 2018-02-16 | 北京旷视科技有限公司 | The attribute recognition approach and device of destination object |
| CN108537136A (en) * | 2018-03-19 | 2018-09-14 | 复旦大学 | The pedestrian's recognition methods again generated based on posture normalized image |
Non-Patent Citations (1)
| Title |
|---|
| 杜久伦: ""多视图行人重识别算法研究与数据采集"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109934177A (en) * | 2019-03-15 | 2019-06-25 | 艾特城信息科技有限公司 | Pedestrian recognition methods, system and computer readable storage medium again |
| CN109919141A (en) * | 2019-04-09 | 2019-06-21 | 广东省智能制造研究所 | A kind of recognition methods again of the pedestrian based on skeleton pose |
| CN111460977A (en) * | 2020-03-30 | 2020-07-28 | 广东电网有限责任公司电力科学研究院 | Cross-vision person re-identification method, device, terminal and storage medium |
| CN111460977B (en) * | 2020-03-30 | 2024-02-20 | 广东电网有限责任公司电力科学研究院 | Cross-view personnel re-identification method, device, terminal and storage medium |
| CN111897993A (en) * | 2020-07-20 | 2020-11-06 | 杭州叙简科技股份有限公司 | Efficient target person track generation method based on pedestrian re-recognition |
| CN113435443A (en) * | 2021-06-28 | 2021-09-24 | 中国兵器装备集团自动化研究所有限公司 | Method for automatically identifying landmark from video |
| CN114820724A (en) * | 2022-06-28 | 2022-07-29 | 浙江树人学院 | Intelligent monitoring method and system for cross-scene tracking |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109409250A (en) | A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning | |
| CN104881637B (en) | Multimodal information system and its fusion method based on heat transfer agent and target tracking | |
| CN108256439A (en) | A kind of pedestrian image generation method and system based on cycle production confrontation network | |
| CN109190508B (en) | Multi-camera data fusion method based on space coordinate system | |
| CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
| CN110796079A (en) | Multi-camera visitor identification method and system based on face depth features and human body local depth features | |
| CN110414462A (en) | A kind of unsupervised cross-domain pedestrian recognition methods and system again | |
| WO2019127273A1 (en) | Multi-person face detection method, apparatus, server, system, and storage medium | |
| CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
| CN110458025B (en) | Target identification and positioning method based on binocular camera | |
| CN105701467A (en) | Many-people abnormal behavior identification method based on human body shape characteristic | |
| CN104915655A (en) | Multi-path monitor video management method and device | |
| CN106991370B (en) | Pedestrian retrieval method based on color and depth | |
| CN102521565A (en) | Garment identification method and system for low-resolution video | |
| CN113269091A (en) | Personnel trajectory analysis method, equipment and medium for intelligent park | |
| CN109784130A (en) | Pedestrian recognition methods and its device and equipment again | |
| Park et al. | Detection of construction workers in video frames for automatic initialization of vision trackers | |
| CN109344842A (en) | A kind of pedestrian's recognition methods again based on semantic region expression | |
| CN109492534A (en) | A kind of pedestrian detection method across scene multi-pose based on Faster RCNN | |
| CN108932509A (en) | A kind of across scene objects search methods and device based on video tracking | |
| CN111460884A (en) | Multi-face recognition method based on human body tracking | |
| CN112668557A (en) | Method for defending image noise attack in pedestrian re-identification system | |
| CN115620090A (en) | Model training method, low-illumination target re-recognition method and device and terminal equipment | |
| CN115100716A (en) | Intelligent community pedestrian tracking and positioning method and system | |
| CN109889773A (en) | Method, apparatus, equipment and the medium of the monitoring of assessment of bids room personnel |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190301 |
|
| RJ01 | Rejection of invention patent application after publication |