CN109063667A - A kind of video identification method optimizing and method for pushing based on scene - Google Patents
A kind of video identification method optimizing and method for pushing based on scene Download PDFInfo
- Publication number
- CN109063667A CN109063667A CN201810921834.4A CN201810921834A CN109063667A CN 109063667 A CN109063667 A CN 109063667A CN 201810921834 A CN201810921834 A CN 201810921834A CN 109063667 A CN109063667 A CN 109063667A
- Authority
- CN
- China
- Prior art keywords
- video
- scene
- analyzed
- frequent
- correlation
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
-
- 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
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The present invention relates to video identification and processing technology fields, and in particular to a kind of video identification method optimizing and method for pushing based on scene.Include the following steps: S1: scene classification is carried out to video according to video content；S2: sorted transmission of video to the processing equipment to match with such video is handled, to obtain processing result.The present invention reaches intelligent recognition and calculates the optimization and push of power scheduling by the semi-supervised learning to video monitoring scene.By the automatic identification to video content, different video content is labeled, while different video contents is distributed into different processing equipments automatically and is handled.Recognition speed is fast, and resource occupation is few, while playing the role of peak load shifting, calculates power optimization.
Description
Technical field
The present invention relates to video identification and processing technology fields, and in particular to a kind of video identification mode based on scene is excellent
Change and method for pushing.
Background technique
Traditional video surveillance system intelligent degree is lower, does not have the ability of machine learning generally, is by imaging, passing
Defeated, control, display, the most of composition of record registration 5.Video camera is led transmission of video images to control by coaxial video cable
Vision signal is assigned to each monitor and video recording equipment again by machine, control host, while can be same by the voice signal for needing to transmit
Step is entered into video recorder.By controlling host, the capable of emitting instruction of operator, to the movements of the upper and lower, left and right of holder into
Row control and the operation that focusing and zooming is carried out to camera lens, and can be realized between multichannel video camera and holder by control host
Switching.Using special video record processing mode, the operation such as typing, playback, processing can be carried out to image, reaches video recording effect most
It is good.
Monitor video processing technique can be divided into three levels, from low to high respectively video processing, video analysis and intelligence
Video analysis, three levels are all critically important, have the space for continuing development, but the technical maturity of intellectual analysis this layer is opposite
Lower, Video Analysis Technology was in the stage developed to intelligent analysis at present.
Video frequency signal processing is foundation, deals with objects for pixel or as block, is not related to video content, is video analysis
Clear, continuous high quality information source is provided, in addition to the video acquisition of comparative maturity, filtering, compression, storage, denoising, enhancing, biography
Outside defeated equal Technology developments, in order to facilitate video analysis, there are also many processing work to improve, such as: it provides more
The equilibrium of high dynamic range (HDR) video image of image detail；Improve the super-resolution rebuilding of video image spatial resolution；
To greasy weather, half-light, the processing for a variety of damaged images such as blocking；Shake of interframe caused by removal is transmitted etc..
Monitor video analysis (VA) processing is related to video content.Due to the occasion and H target difference monitor video of application
The content that analysis includes is very many and diverse, such as:
1) scene cut, feature extraction, front and back scape separation etc.；
2) target detection and tracking, face/Car license recognition etc.；
3) Activity recognition, unusual checking, group behavior identification；
4) stream of people/wagon flow statistics, people invade detection etc..
Intelligent monitoring video analysis, it is not only related with video content, it is also related with semanteme expressed by video, it is desirable to from view
The information such as the meaning of scene state, target category, movement or scene are obtained in frequency content analysis.In general it is desired that passing through intelligence point
Analysis, the semantic conclusion of video content is independently obtained by computer, in other words manually intelligent method for people provide it is a variety of " depending on
Feel service ".There are many ways to intelligent video analysis technology belongs to the scope of artificial intelligence, realizes intellectual analysis, machine learning
Method be the analysis method generallyd use.
Machine learning model includes a large amount of learning algorithms, such as KMeans, naive Bayesian, artificial neural network etc.,
However, which kind of learning model all can be extensive at a universal model.
Four elements of machine learning are input, learning process, knowledge store and knowledge evaluation.Input is also known as
Environment is machine learning algorithm oracle to be processed, and predominantly machine-learning process provides training data and test specimens
This, input data level of abstraction has larger impact to model calculated performance and model usage scenario, under normal circumstances high toolization
Information is suitable for specific toolization and analyzes scene, and very high level conceptual information is since to abstract degree higher for it, so suitable solution
Evolvement problem；And input data quality also determines learning effect and model performance, therefore usually can all carry out to input data
Pretreatment operation, to eliminate the obstacle datas such as missing values, noise of input information, if input data complete and accurate, thus mould
Type induction and conclusion knowledge process is relatively easy to, and can be achieved the desired results, if data format missing or abnormal, effect
It is poor.
Learning process is learnt by external input information, and model finds that induction and conclusion goes out to know in learning process
Know, in systems by knowledge store, and knowledge is constantly improved by knowledge evaluation feedback.Learning process is with input data
Based on, by constructing model analysis, conclusion can obtain training pattern, and learning model training process is in machine learning step
Mostly important link is related to quantity of parameters configuration and parameter adjustment.
Knowledge store is to solidify the knowledge of learning process study, is mainly concerned with knowledge base creation, knowledge store
And the representation of knowledge, having larger impact to knowledge base and knowledge store is the representation of knowledge, modifies difficulty or ease, expression energy according to knowledge
Power, expansion capacity and inferential capability determine representation of knowledge form.
Knowledge evaluation model is machine learning overall model most researching value part, plays the role of coach, model
Performance quality is inseparable with knowledge evaluation.This link mainly solves the problems, such as currently to face practical, actual result comparison calculation
Method model result, enables arithmetic result to be fed back and model Automatic Optimal, is formed between actual conditions and operation result
Benign cycle.
The algorithm of video scene segmentation mainly has: based on clustering shots and Video Quality Metric nomography, be based on after to camera lens one
Cause property and scene dynamics algorithm are based on NCuts algorithm, are based on smooth visualization bag of words algorithm, are based on clustering shots and sequence alignment
Algorithm etc..However there are still some problems for these algorithms.One is in terms of camera lens description, or rely on global characteristics such as face
Color, texture etc., or local feature such as Scale invariant transformation and context contrast histogram etc. are relied on, it could not be by each of the two
It is combined from advantage and plays most effective descriptive power；The second is above method is all finding one always in terms of scene cut
The solution of kind global optimum, and rich and varied, the video production process of video genre and video content is considered in practice
Occupancy complicated and changeable processing hardware resource it is also more, while arithmetic speed is also relatively slow.
Summary of the invention
Mesh of the present invention provides a kind of video identification method optimizing and method for pushing based on scene, solves the prior art
Middle video content recognition occupies the problem that resource is more, recognition speed is slow.
The technical scheme adopted by the invention is as follows:
A kind of video identification method optimizing and method for pushing based on scene, includes the following steps:
S1: scene classification is carried out to video according to video content；
S2: sorted transmission of video to the processing equipment to match with such video is handled, to obtain processing knot
Fruit.
Further, the step S1 includes:
A1: dividing multiple and different video scene types previously according to video content, default frequent to each video type
Camera lens set and label；
A2: the video being analysed to carries out correlation comparison, the highest frequent camera collection of correlation with frequent camera set
The scene type of conjunction is the scene type of video to be analyzed；
A3: the label of selected video scene type is marked on video to be analyzed after correlation comparison.
Further, the step S2 includes:
B1: corresponding label is matched to different processing equipments in advance；
B2: after the signal of the video to be analyzed after receiving label, reading the label on video to be analyzed, marks corresponding
The video distribution to be analyzed of label is handled to corresponding processing equipment；
B3: corresponding processing equipment handles video to be analyzed, obtains processing result.
Further, the step a2 includes:
C1: Shot Detection first carries out video frame delineation to video to be analyzed, then establishes video continuous signal, then carry out
It is detected based on threshold boundaries；
C2: key-frame extraction carries out key-frame extraction to video to be analyzed, removes redundant content；
C3: scene cut is first described the video lens after key-frame extraction operates；Then between progress camera lens
Video content is carried out correlation with frequent camera set and compared, selects the highest frequent camera collection of correlation by correlation calculations
It closes；The video and frequent camera set are collectively constituted into new frequency again
Numerous camera lens set.
It further, further include the processing of unallocated camera lens in the step c3；If by correlation calculations between camera lens
Afterwards, each frequent camera set and the correlation of video to be analyzed are below 50%, then by the video carry out special marking, with to
Artificial treatment.
Further, in the step S1, the classification of video content includes the scene for being mainly vehicle, mainly people and people
The sparse scenes of number, mainly people and the dense scene of number and it is rainy when because in the scene that face is blocked of holding up an umbrella one
Kind is a variety of.
Further, the processing equipment in the step S2 or b2 includes Baidu's face recognition device, Shen Mo vehicle identification
One of face recognition device of equipment, the face recognition device of everybody intelligence and Ge Ling depth pupil is a variety of.
Further, in the step a1 in preset frequent camera set, including identical scene type 500 with
On.
Further, periodically the new frequent camera set formed in step c3 is inspected by random samples, guarantees frequent camera collection
The accuracy of conjunction is 85% or more.
Further, in the step c1, border detection is carried out in the case where threshold value is 10-1000.
The invention has the benefit that
The present invention is reached intelligent recognition and is calculated the optimization of power scheduling and pushed away by the semi-supervised learning to video monitoring scene
It send.By the automatic identification to video content, different video content is labeled, while automatically by different video content point
The different processing equipment of dispensing is handled.Recognition speed is fast, and resource occupation is few, while playing peak load shifting, calculates power optimization
Effect.
Detailed description of the invention
Fig. 1 is video scene classification process figure of the present invention.
Specific embodiment
With reference to the accompanying drawing and specific embodiment does further explaination to the present invention.
Embodiment 1:
A kind of video identification method optimizing and method for pushing based on scene is present embodiments provided, video to be processed is
It is mounted on the picture that camera is shot at the one of certain tourist attractions, divides multiple and different video scene types in advance, to each
Video type presets frequent end set and label.Preset scene is as follows: 1, densely populated place scene, selects 500 personnel close
The scene of collection is named as 0001 to the scene as default frequent camera set；2, the sparse scene of personnel selects 500 people
The sparse scene of member is named as 0010 to the scene as default frequent camera set；3, environment obscures scene, selects 500
A scene not high because of visibility when rainy, cloudy day or greasy weather is used as default frequent camera set, and to this
Scape is named as 0011；4, there is the scene of shelter, select 500 to block face by umbrella because of pedestrian of raining or since it is desired that hide
It keeps off the sun and default frequent camera set is used as by the scene that parasol blocks face, and be named as 0100 to the scene；5, vehicle
Scene selects the scene of 500 mainly vehicles as default frequent camera set, and is named as 0101 to the scene.
The video being analysed to carries out correlation comparison with frequent camera set, the highest frequent camera set of correlation
Scene type is the scene type of video to be analyzed, and the label label of selected video scene type exists after correlation comparison
On video to be analyzed.When the picture that camera takes and default scene carry out correlation comparison, if people in the picture taken
Member sparse, the preset sparse scene of personnel and the picture correlation highest taken, then be labeled as 0010；When camera takes
Picture inside densely populated place, preset densely populated place scene and the picture correlation highest taken, then labeled as 0001；When
Cause visibility not high because of the greasy weather, when the picture taken is relatively fuzzy, preset environment obscures scene and takes
Picture correlation highest is then labeled as 0011；In the case where because most of rainy day pedestrian is equipped with umbrella, face is by umbrella
In the case where blocking, the preset scene for having shelter and the picture correlation highest taken are then labeled as 0100.
It is default that the Baidu's face recognition device for being suitble to the scene analysis more than people will be given to carry out labeled as 0001 video push
Processing；It will be handled labeled as 0010 video push to everybody the higher face recognition device of intelligence of accuracy；It will mark
The face recognition device that 0011 and 0100 video push is denoted as to the deep pupil of lattice spirit is handled；Video labeled as 0101 is pushed away
Shen Mo vehicle identification equipment is given to be handled.
It can be by video when needing the densely populated place degree by the camera real-time monitoring region, people is few
It is pushed to everybody face recognition device of intelligence to handle, when number is more than preset value, automatically by video push to Baidu
Face recognition device is handled, when Baidu's face recognition device detects that the area people concentration is more than safety value,
Operation of warning is carried out immediately.
When scenic spot has child to wander away, and needs to carry out by camera to carry out Automatic-searching by recognition of face, in the area
When domain personnel are more sparse, handled by everybody face recognition device of intelligence, faster, the probability found is more for recognition speed
Greatly.When the area people is more intensive, handled by Baidu's face recognition device, recognition speed faster, the probability found
It is bigger.When rainy, then handled by the deep pupil face recognition device of lattice spirit, faster, accuracy rate is higher for recognition speed, finds
Probability is bigger.
The present invention can be different according to different environmental selections processing equipment, make processing speed faster, resource occupation is more
It is small.
Embodiment 2:
Present embodiments provide a kind of video identification method optimizing and method for pushing based on scene, the present embodiment and implementation
Example 1 is essentially identical, and difference is: if video to be processed is the picture for being mounted on the camera shooting of crossroad on street
When, when in the video that some period takes being mainly vehicle, then distributes to Shen Mo vehicle identification equipment and handled.
Embodiment 3:
A kind of video identification method optimizing and method for pushing based on scene is present embodiments provided, is included the following steps:
S1: scene classification is carried out to video according to video content；
S2: sorted transmission of video to the processing equipment to match with such video is handled, to obtain
Processing result.
In the step S1, the classification of video content includes the scene for being mainly vehicle, is mainly people and number is sparse
Scene, mainly people and the dense scene of number and it is rainy when because of one of scene that face is blocked of holding up an umbrella or more
Kind.
The step S1 includes:
A1: dividing multiple and different video scene types previously according to video content, default frequent to each video type
Camera lens set and label；
A2: the video being analysed to carries out correlation comparison, the highest frequent camera collection of correlation with frequent camera set
The scene type of conjunction is the scene type of video to be analyzed；
A3: the label of selected video scene type is marked on video to be analyzed after correlation comparison.
In the step a1 in preset frequent camera set, including identical scene type 500 or more.
The step S2 includes:
B1: corresponding label is matched to different processing equipments in advance；
B2: after the signal of the video to be analyzed after receiving label, reading the label on video to be analyzed,
The video distribution to be analyzed of corresponding label is handled to corresponding processing equipment；
B3: corresponding processing equipment handles video to be analyzed, obtains processing result.
Processing equipment in the step S2 or b2 include Baidu's face recognition device, Shen Mo vehicle identification equipment, everybody
One of intelligent face recognition device of face recognition device and Ge Ling depth pupil is a variety of.
The step a2 includes:
C1: Shot Detection first carries out video frame delineation to video to be analyzed, then establishes video continuous signal,
It carries out detecting based on threshold boundaries again；
C2: key-frame extraction carries out key-frame extraction to video to be analyzed, removes redundant content；
C3: scene cut is first described the video lens after key-frame extraction operates；Then into
Video content is carried out correlation with frequent camera set and compared, selected by correlation calculations between row camera lens
The highest frequent camera set of correlation；The video and frequent camera set are collectively constituted into new frequency again
Numerous camera lens set.
In the step c1, border detection is carried out in the case where threshold value is 10.
It further include the processing of unallocated camera lens in the step c3；If each frequent after correlation calculations between camera lens
Camera lens set and the correlation of video to be analyzed are below 50%, then the video are carried out special marking, to artificial treatment.
The present invention carries out frequent camera set by the way of semi-supervised learning perfect, can make the scene markers of video
It is more accurate, and can adapt to more changeable environment.Semi-supervised learning, i.e. model training input data are blended data, this
In blended data be label data and non-label data, in actual operation, model training needs a large amount of training datas, but logical
Normal label data needs are manually labeled, and this kind of data are less, and nonstandard preceding data are more, this kind of method can alleviate mark
The disadvantages of signing data deficiencies, to carry out maturing training to model.
Embodiment 4:
Present embodiments provide a kind of video identification method optimizing and method for pushing based on scene, the present embodiment and implementation
Example 3 is essentially identical, and difference is:
This method in the process of processing, need periodically to the new frequent camera set formed in step c3 into
Row sampling observation guarantees the accuracy rate of frequent camera set 85% or more.The time interval of sampling observation is -15 days 7 days.If being sent out after sampling observation
Existing accuracy rate is lower than 85%, then replaces the frequency camera lens set.
Embodiment 5:
A kind of video identification method optimizing and method for pushing based on scene is present embodiments provided, is included the following steps:
S1: scene classification is carried out to video according to video content；
S2: sorted transmission of video to the processing equipment to match with such video is handled, to obtain
Processing result.
In the step S1, the classification of video content includes the scene for being mainly vehicle, is mainly people and number is sparse
Scene, mainly people and the dense scene of number and it is rainy when because of one of scene that face is blocked of holding up an umbrella or more
Kind.
The step S1 includes:
A1: dividing multiple and different video scene types previously according to video content, default frequent to each video type
Camera lens set and label；
A2: the video being analysed to carries out correlation comparison, the highest frequent camera collection of correlation with frequent camera set
The scene type of conjunction is the scene type of video to be analyzed；
A3: the label of selected video scene type is marked on video to be analyzed after correlation comparison.
In the step a1 in preset frequent camera set, including identical scene type 500 or more.
The step S2 includes:
B1: corresponding label is matched to different processing equipments in advance；
B2: after the signal of the video to be analyzed after receiving label, reading the label on video to be analyzed,
The video distribution to be analyzed of corresponding label is handled to corresponding processing equipment；
B3: corresponding processing equipment handles video to be analyzed, obtains processing result.
Processing equipment in the step S2 or b2 include Baidu's face recognition device, Shen Mo vehicle identification equipment, everybody
One of intelligent face recognition device of face recognition device and Ge Ling depth pupil is a variety of.
The step a2 includes:
C1: Shot Detection first carries out video frame delineation to video to be analyzed, then establishes video continuous signal,
It carries out detecting based on threshold boundaries again；
C2: key-frame extraction carries out key-frame extraction to video to be analyzed, removes redundant content；
C3: scene cut is first described the video lens after key-frame extraction operates；Then into
Video content is carried out correlation with frequent camera set and compared, selected by correlation calculations between row camera lens
The highest frequent camera set of correlation；The video and frequent camera set are collectively constituted into new frequency again
Numerous camera lens set.
In the step c1, border detection is carried out in the case where threshold value is 1000.
It further include the processing of unallocated camera lens in the step c3；If each frequent after correlation calculations between camera lens
Camera lens set and the correlation of video to be analyzed are below 50%, then the video are carried out special marking, to artificial treatment.
Embodiment 6:
A kind of video identification method optimizing and method for pushing based on scene is present embodiments provided, is included the following steps:
S1: scene classification is carried out to video according to video content；
S2: sorted transmission of video to the processing equipment to match with such video is handled, to obtain
Processing result.
In the step S1, the classification of video content includes the scene for being mainly vehicle, is mainly people and number is sparse
Scene, mainly people and the dense scene of number and it is rainy when because of one of scene that face is blocked of holding up an umbrella or more
Kind.
The step S1 includes:
A1: dividing multiple and different video scene types previously according to video content, default frequent to each video type
Camera lens set and label；
A2: the video being analysed to carries out correlation comparison, the highest frequent camera collection of correlation with frequent camera set
The scene type of conjunction is the scene type of video to be analyzed；
A3: the label of selected video scene type is marked on video to be analyzed after correlation comparison.
In the step a1 in preset frequent camera set, including identical scene type 500 or more.
The step S2 includes:
B1: corresponding label is matched to different processing equipments in advance；
B2: after the signal of the video to be analyzed after receiving label, reading the label on video to be analyzed,
The video distribution to be analyzed of corresponding label is handled to corresponding processing equipment；
B3: corresponding processing equipment handles video to be analyzed, obtains processing result.
Processing equipment in the step S2 or b2 include Baidu's face recognition device, Shen Mo vehicle identification equipment, everybody
One of intelligent face recognition device of face recognition device and Ge Ling depth pupil is a variety of.
The step a2 includes:
C1: Shot Detection first carries out video frame delineation to video to be analyzed, then establishes video continuous signal,
It carries out detecting based on threshold boundaries again；
C2: key-frame extraction carries out key-frame extraction to video to be analyzed, removes redundant content；
C3: scene cut is first described the video lens after key-frame extraction operates；Then into
Video content is carried out correlation with frequent camera set and compared, selected by correlation calculations between row camera lens
The highest frequent camera set of correlation；The video and frequent camera set are collectively constituted into new frequency again
Numerous camera lens set.
In the step c1, border detection is carried out in the case where threshold value is 500.
It further include the processing of unallocated camera lens in the step c3；If each frequent after correlation calculations between camera lens
Camera lens set and the correlation of video to be analyzed are below 50%, then the video are carried out special marking, to artificial treatment.
The present invention is not limited to above-mentioned optional embodiment, anyone can show that other are each under the inspiration of the present invention
The product of kind form.Above-mentioned specific embodiment should not be understood the limitation of pairs of protection scope of the present invention, protection of the invention
Range should be subject to be defined in claims, and specification can be used for interpreting the claims.
Claims (10)
1. a kind of video identification method optimizing and method for pushing based on scene, which comprises the steps of:
S1: scene classification is carried out to video according to video content；
S2: sorted transmission of video to the processing equipment to match with such video is handled, to obtain processing result.
2. a kind of video identification method optimizing and method for pushing based on scene according to claim 1, which is characterized in that
The step S1 includes:
A1: dividing multiple and different video scene types previously according to video content, presets frequent camera to each video type
Set and label；
A2: the video being analysed to carries out correlation comparison with frequent camera set, the highest frequent camera set of correlation
Scene type is the scene type of video to be analyzed；
A3: the label of selected video scene type is marked on video to be analyzed after correlation comparison.
3. a kind of video identification method optimizing and method for pushing based on scene according to claim 2, it is characterised in that:
In the step a1 in preset frequent camera set, including identical scene type 500 or more.
4. a kind of video identification method optimizing and method for pushing based on scene according to claim 2, which is characterized in that
The step a2 includes:
C1: Shot Detection first carries out video frame delineation to video to be analyzed, then establishes video continuous signal,
It carries out detecting based on threshold boundaries again；
C2: key-frame extraction carries out key-frame extraction to video to be analyzed, removes redundant content；
C3: scene cut is first described the video lens after key-frame extraction operates；Then related between progress camera lens
Property calculate, by video content and frequent camera set progress correlation compare, select the highest frequent camera set of correlation；Again
The video and frequent camera set are collectively constituted into new frequent camera set.
5. a kind of video identification method optimizing and method for pushing based on scene according to claim 4, it is characterised in that:
Periodically the new frequent camera set formed in step c3 is inspected by random samples, guarantee frequent camera set accuracy 85% with
On.
6. a kind of video identification method optimizing and method for pushing based on scene according to claim 4, it is characterised in that:
In the step c1, border detection is carried out in the case where threshold value is 10-1000.
7. a kind of video identification method optimizing and method for pushing based on scene according to claim 4, it is characterised in that:
It further include the processing of unallocated camera lens in the step c3；If after correlation calculations between camera lens, each frequent camera set
It is below 50% with the correlation of video to be analyzed, then the video is subjected to special marking, to artificial treatment.
8. a kind of video identification method optimizing and method for pushing based on scene according to claim 1, which is characterized in that
The step S2 includes:
B1: corresponding label is matched to different processing equipments in advance；
B2: after the signal of the video to be analyzed after receiving label, reading the label on video to be analyzed,
The video distribution to be analyzed of corresponding label is handled to corresponding processing equipment；
B3: corresponding processing equipment handles video to be analyzed, obtains processing result.
9. a kind of video identification method optimizing and method for pushing based on scene according to claim 8, it is characterised in that:
Processing equipment in the step S2 or b2 includes the people of Baidu's face recognition device, Shen Mo vehicle identification equipment, everybody intelligence
Face identifies one of face recognition device of equipment and Ge Ling depth pupil or a variety of.
10. a kind of video identification method optimizing and method for pushing, feature based on scene according to claim 1 exists
In: in the step S1, the classification of video content include mainly the scene of vehicle, mainly people and the sparse scene of number,
Mainly people and the dense scene of number and it is rainy when because of one of scene that face is blocked of holding up an umbrella or a variety of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810921834.4A CN109063667B (en) | 2018-08-14 | 2018-08-14 | Scene-based video identification mode optimization and pushing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810921834.4A CN109063667B (en) | 2018-08-14 | 2018-08-14 | Scene-based video identification mode optimization and pushing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063667A true CN109063667A (en) | 2018-12-21 |
| CN109063667B CN109063667B (en) | 2021-02-19 |
Family
ID=64683897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810921834.4A Active CN109063667B (en) | 2018-08-14 | 2018-08-14 | Scene-based video identification mode optimization and pushing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063667B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109815936A (en) * | 2019-02-21 | 2019-05-28 | 深圳市商汤科技有限公司 | A kind of target object analysis method and device, computer equipment and storage medium |
| CN111209897A (en) * | 2020-03-09 | 2020-05-29 | 腾讯科技（深圳）有限公司 | Video processing method, device and storage medium |
| CN111401239A (en) * | 2020-03-16 | 2020-07-10 | 科大讯飞(苏州)科技有限公司 | Video analysis method, device, system, equipment and storage medium |
| WO2020248386A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技（深圳）有限公司 | Video analysis method and apparatus, computer device and storage medium |
| CN112799826A (en) * | 2019-11-14 | 2021-05-14 | 杭州海康威视数字技术股份有限公司 | Intelligent analysis algorithm selection method, device and system and electronic equipment |
| CN112911234A (en) * | 2021-01-26 | 2021-06-04 | 上海商米科技集团股份有限公司 | Power consumption control method suitable for intelligent IPC |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101021904A (en) * | 2006-10-11 | 2007-08-22 | 鲍东山 | Video content analysis system |
| CN101834837A (en) * | 2009-12-18 | 2010-09-15 | 北京邮电大学 | On-line landscape video active information service system of scenic spots in tourist attraction based on bandwidth network |
| CN102214222A (en) * | 2011-06-15 | 2011-10-12 | 中国电信股份有限公司 | Presorting and interacting system and method for acquiring scene information through mobile phone |
| CN102508246A (en) * | 2011-10-13 | 2012-06-20 | 吉林大学 | Method for detecting and tracking obstacles in front of vehicle |
| US20180032846A1 (en) * | 2016-08-01 | 2018-02-01 | Nvidia Corporation | Fusing multilayer and multimodal deep neural networks for video classification |
| CN108009518A (en) * | 2017-12-19 | 2018-05-08 | 大连理工大学 | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks |
-
2018
- 2018-08-14 CN CN201810921834.4A patent/CN109063667B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101021904A (en) * | 2006-10-11 | 2007-08-22 | 鲍东山 | Video content analysis system |
| CN101834837A (en) * | 2009-12-18 | 2010-09-15 | 北京邮电大学 | On-line landscape video active information service system of scenic spots in tourist attraction based on bandwidth network |
| CN102214222A (en) * | 2011-06-15 | 2011-10-12 | 中国电信股份有限公司 | Presorting and interacting system and method for acquiring scene information through mobile phone |
| CN102508246A (en) * | 2011-10-13 | 2012-06-20 | 吉林大学 | Method for detecting and tracking obstacles in front of vehicle |
| US20180032846A1 (en) * | 2016-08-01 | 2018-02-01 | Nvidia Corporation | Fusing multilayer and multimodal deep neural networks for video classification |
| CN108009518A (en) * | 2017-12-19 | 2018-05-08 | 大连理工大学 | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109815936A (en) * | 2019-02-21 | 2019-05-28 | 深圳市商汤科技有限公司 | A kind of target object analysis method and device, computer equipment and storage medium |
| CN109815936B (en) * | 2019-02-21 | 2023-08-22 | 深圳市商汤科技有限公司 | Target object analysis method and device, computer equipment and storage medium |
| WO2020248386A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技（深圳）有限公司 | Video analysis method and apparatus, computer device and storage medium |
| CN112799826A (en) * | 2019-11-14 | 2021-05-14 | 杭州海康威视数字技术股份有限公司 | Intelligent analysis algorithm selection method, device and system and electronic equipment |
| WO2021093625A1 (en) * | 2019-11-14 | 2021-05-20 | 杭州海康威视数字技术股份有限公司 | Intelligent analysis algorithm selection method, apparatus and system, and electronic device |
| CN111209897A (en) * | 2020-03-09 | 2020-05-29 | 腾讯科技（深圳）有限公司 | Video processing method, device and storage medium |
| CN111209897B (en) * | 2020-03-09 | 2023-06-20 | 深圳市雅阅科技有限公司 | Video processing method, device and storage medium |
| CN111401239A (en) * | 2020-03-16 | 2020-07-10 | 科大讯飞(苏州)科技有限公司 | Video analysis method, device, system, equipment and storage medium |
| CN111401239B (en) * | 2020-03-16 | 2021-04-20 | 科大讯飞(苏州)科技有限公司 | Video analysis method, device, system, equipment and storage medium |
| CN112911234A (en) * | 2021-01-26 | 2021-06-04 | 上海商米科技集团股份有限公司 | Power consumption control method suitable for intelligent IPC |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063667B (en) | 2021-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109063667A (en) | A kind of video identification method optimizing and method for pushing based on scene | |
| CN109948582B (en) | Intelligent vehicle reverse running detection method based on tracking trajectory analysis | |
| CN109376572B (en) | Real-time vehicle detection and trajectory tracking method in traffic video based on deep learning | |
| CN105574543B (en) | A kind of vehicle brand type identifier method and system based on deep learning | |
| Fregin et al. | The driveu traffic light dataset: Introduction and comparison with existing datasets | |
| US20230289979A1 (en) | A method for video moving object detection based on relative statistical characteristics of image pixels | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| CN110472467A (en) | The detection method for transport hub critical object based on YOLO v3 | |
| CN104915655A (en) | Multi-path monitor video management method and device | |
| CN108304798A (en) | The event video detecting method of order in the street based on deep learning and Movement consistency | |
| CN109829382B (en) | Abnormal target early warning tracking system and method based on intelligent behavior characteristic analysis | |
| CN108776974B (en) | A kind of real-time modeling method method suitable for public transport scene | |
| Amirgholipour et al. | A-CCNN: adaptive CCNN for density estimation and crowd counting | |
| CN104378582A (en) | Intelligent video analysis system and method based on PTZ video camera cruising | |
| CN107424171A (en) | A kind of anti-shelter target tracking based on piecemeal | |
| CN108182447A (en) | A kind of adaptive particle filter method for tracking target based on deep learning | |
| CN105989358A (en) | Natural scene video identification method | |
| CN111340855A (en) | Road moving target detection method based on track prediction | |
| CN112241969A (en) | Target detection tracking method and device based on traffic monitoring video and storage medium | |
| CN111524164B (en) | Target tracking method and device and electronic equipment | |
| CN111738218B (en) | Human body abnormal behavior recognition system and method | |
| CN103886089B (en) | Driving recording video concentration method based on study | |
| CN105512618A (en) | Video tracking method | |
| CN109948474A (en) | AI thermal imaging all-weather intelligent monitoring method | |
| WO2019043406A1 (en) | Anomaly detection from video data from surveillance cameras |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Optimization and Push Method of a Scene Based Video Recognition Method Effective date of registration: 20230822 Granted publication date: 20210219 Pledgee: Dongshan Branch of Guangzhou Bank Co.,Ltd. Pledgor: SHIYUN RONGJU (GUANGZHOU) TECHNOLOGY Co.,Ltd. Registration number: Y2023980053288 |