CN107864334A - A kind of intelligent camera lens image pickup method and system using deep learning - Google Patents
A kind of intelligent camera lens image pickup method and system using deep learning Download PDFInfo
- Publication number
- CN107864334A CN107864334A CN201711098878.3A CN201711098878A CN107864334A CN 107864334 A CN107864334 A CN 107864334A CN 201711098878 A CN201711098878 A CN 201711098878A CN 107864334 A CN107864334 A CN 107864334A
- Authority
- CN
- China
- Prior art keywords
- shooting
- fragment
- deep learning
- camera lens
- current
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/60—Control of cameras or camera modules
- H04N23/667—Camera operation mode switching, e.g. between still and video, sport and normal or high- and low-resolution modes
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
- Studio Devices (AREA)
Abstract
The invention discloses a kind of intelligent camera lens image pickup method using deep learning, comprise the following steps：Open camera lens to be shot, the destination object in camera lens is perceived using the method for deep learning, obtains the feature of destination object；Detection positioning is carried out to shooting fragment using the method for deep learning, if to detect current clip be fragment interested, shooting high speed mode is opened, if to detect current clip be non-fragment interested, common screening-mode is opened, the shooting speed of the common screening-mode is less than shooting high speed mode.Automatic switchover shooting high speed mode of the present invention and common screening-mode, it is not necessary to manual operation, while possess reaction speed more faster than people, avoid missing important moment.
Description
Technical field
The present invention relates to a kind of intelligent camera lens image pickup method and system using deep learning.
Background technology
People are often worth record using camera record slice of life, performing active, competitive sports etc. or repeated
The scene of viewing.When watching documentary film, can usually the method for slow play be used, it is excellent or important to examine some of which
Fragment.To these fragments, be very important using high-speed camera to carry out shooting, picture otherwise can be caused discontinuous or
Person leaks through the problems such as some important moments.And to the shooting of fragment in addition, if still carried out using high-speed camera
Shooting, then can take substantial amounts of memory space.
The content of the invention
The technical problem to be solved in the present invention is to provide a kind of intelligent camera lens image pickup method and system using deep learning,
Automatic switchover shooting high speed mode and common screening-mode, it is not necessary to manual operation, while possess reaction speed more faster than people,
Avoid missing important moment.
In order to solve the above-mentioned technical problem, the present invention takes following technical scheme：
A kind of intelligent camera lens image pickup method using deep learning, comprise the following steps：
Open camera lens to be shot, the destination object in camera lens is perceived using the method for deep learning, obtain
The feature of destination object；
Detection positioning is carried out to shooting fragment using the method for deep learning, if to detect current clip be fragment interested,
Shooting high speed mode is opened, if to detect current clip be non-fragment interested, opens common screening-mode, the common shooting
The shooting speed of pattern is less than shooting high speed mode.
When detecting positioning shooting fragment, the predetermined threshold value of a confidence level is defined to shooting fragment, if current shooting piece
The confidence level of section is more than threshold value, then it is fragment interested to show current shooting fragment, shooting high speed mode is opened, if current shooting
The confidence level of fragment is less than threshold value, then it is non-fragment interested to show current shooting fragment, opens common screening-mode.
It is described when being perceived to destination object, obtain destination object using convolutional neural networks and Recognition with Recurrent Neural Network
Space characteristics and temporal aspect, identify the scene around clarification of objective key element and destination object, and characteristic element includes position, appearance
State and behavior, the information that characteristic element is changed over time and changed is obtained, obtain the confidence level of current frame image.
Each two field picture in the shooting is all carried out the extraction of space characteristics by identical convolutional neural networks, then should
Convolutional neural networks export current timing information, and the current timing information forms sequential spy together with timing information before
Sign, so as to export the confidence level of the current clip of current frame image, while timing information by current timing information and before
Accumulated and exported.
During the deep learning, a data set for including several video segments, each video segment are built first
All carry out fragment start time point interested and terminate the mark at time point, train to obtain for deep learning with the data set
Deep neural network model.
A kind of intelligent camera lens camera system using deep learning, the system include：Unit is perceived, for perceiving and knowing
Scene around other destination object, and the feature of destination object, the perception unit include human detection module, Attitude estimation mould
Block, scene perception module, human tracking module；Positioning unit is detected, for carrying out detection positioning to shooting fragment, is checked whether
For fragment interested；Control unit, for switching shooting high speed mode and common screening-mode.
Also include comparison unit, for comparing the confidence level of current clip and the magnitude relationship of threshold value.
It is current also to include long mnemon in short-term, timing information for current timing information and before, and output
The confidence level of the current clip of two field picture, while the timing information to receiving is accumulated and exported to next length and remembered in short-term
Recall unit.
Automatic switchover shooting high speed mode of the present invention and common screening-mode, it is not necessary to manual operation, while possess and compare people
Faster reaction speed, important moment is avoided missing, obtain comprehensive and accurate spatial information and timing information, catch sense exactly
The fragment of interest, real-time user behavior analysis.
Brief description of the drawings
Accompanying drawing 1 switches the schematic diagram of screening-mode for the present invention；
Accompanying drawing 2 is perceived for the present invention and the schematic diagram of the confidence level of output shooting fragment.
Embodiment
For the feature of the present invention, technological means and the specific purposes reached, function can be further appreciated that, with reference to
Accompanying drawing is described in further detail with embodiment to the present invention.
As shown in Figure 1, present invention is disclosed a kind of intelligent camera lens image pickup method using deep learning, including following step
Suddenly：
Open camera lens to be shot, the destination object in camera lens is perceived using the method for deep learning, obtain
The feature of destination object, the specially position of the scene around identification destination object and destination object, posture, behavior etc..It is deep
During degree study, a data set for including several video segments is built first, and each video segment carries out interested
Section start time point and the mark for terminating time point, are trained to obtain the deep neural network mould for deep learning with the data set
Type, can effectively it be perceived using the deep neural network model.
Detection positioning is carried out to shooting fragment using the method for deep learning, current clip is fragment interested if detecting
When, shooting high speed mode is opened, if to detect current clip be non-fragment interested, opens common screening-mode, this is common
The shooting speed of screening-mode is less than shooting high speed mode.Automatically switch in shooting high speed mode and common screening-mode, from
And ensure timely to carry out fragment interested high-speed capture, will not mistakes and omissions excessively important moment.
When detecting positioning shooting fragment, the predetermined threshold value of a confidence level is defined to shooting fragment, if current shooting piece
The confidence level of section is more than threshold value, then it is fragment interested to show current shooting fragment, shooting high speed mode is opened, if current shooting
The confidence level of fragment is less than threshold value, then it is non-fragment interested to show current shooting fragment, opens common screening-mode.If work as
Preceding screening-mode is common screening-mode, when it is fragment interested that detection, which navigates to current clip, by common screening-mode certainly
The dynamic shooting high speed mode that is switched to carries out high-speed capture；If current shooting pattern is shooting high speed mode, when detection navigates to
When current clip is non-fragment interested, common screening-mode is automatically switched to by shooting high speed mode.
When specific shooting perceives destination object, the sky of destination object is obtained using convolutional neural networks and Recognition with Recurrent Neural Network
Between feature and temporal aspect, identify the scene around clarification of objective key element and destination object, characteristic element includes position, posture
With the information such as behavior, the information that characteristic element is changed over time and changed is obtained, obtains the confidence level of current frame image.Clap
Each two field picture in taking the photograph all is carried out the extraction of space characteristics by identical convolutional neural networks, and then the convolutional neural networks are defeated
Go out current timing information, the current timing information forms temporal aspect together with timing information before, works as so as to export
The confidence level of the current clip of prior image frame, while timing information by current timing information and before is accumulated and defeated
Go out.With several length, mnemon, convolutional neural networks are output to long short-term memory list after obtaining target object information in short-term
Member, the length in short-term mnemon simultaneously also receive before timing information, combine and acquire temporal aspect, then will
Current timing information and timing information before, which are accumulated, is output to next length mnemon in short-term.
In addition, present invention further teaches a kind of intelligent camera lens camera system using deep learning, the system includes：Sense
Know unit, for perceiving and identifying the scene around destination object, and the feature of destination object, this feature is destination object
The information such as position, posture, behavior, the perception unit include human detection module, Attitude estimation module, scene perception module, people
Volume tracing module；Positioning unit is detected, for carrying out detection positioning to shooting fragment, is checked whether as fragment interested；Control
Unit, for switching shooting high speed mode and common screening-mode.Comparison unit, for comparing the confidence level and threshold of current clip
The magnitude relationship of value.
It is current also to include long mnemon in short-term, timing information for current timing information and before, and output
The confidence level of the current clip of two field picture, while the timing information to receiving is accumulated and exported to next length and remembered in short-term
Recall unit.
Convolutional neural networks include the deep neural network module of multiple different tasks, such as human detection module, posture
Estimation module, human tracking module etc., the scene of surrounding is identified, obtain the information such as the position of target, posture, behavior.
It should be noted that these are only the preferred embodiments of the present invention, it is not intended to limit the invention, although ginseng
The present invention is described in detail according to embodiment, for those skilled in the art, it still can be to foregoing reality
Apply the technical scheme described in example to modify, or equivalent substitution is carried out to which part technical characteristic, but it is all in this hair
Within bright spirit and principle, any modification, equivalent substitution and improvements made etc., protection scope of the present invention should be included in
Within.
Claims (8)
1. a kind of intelligent camera lens image pickup method using deep learning, comprise the following steps：
Open camera lens to be shot, the destination object in camera lens is perceived using the method for deep learning, obtain
The feature of destination object；
Detection positioning is carried out to shooting fragment using the method for deep learning, if to detect current clip be fragment interested,
Shooting high speed mode is opened, if to detect current clip be non-fragment interested, opens common screening-mode, the common shooting
The shooting speed of pattern is less than shooting high speed mode.
2. the intelligent camera lens image pickup method according to claim 1 using deep learning, it is characterised in that detect positioning
When shooting fragment, the predetermined threshold value of a confidence level is defined to shooting fragment, if the confidence level of current shooting fragment is more than threshold value,
It is fragment interested then to show current shooting fragment, shooting high speed mode is opened, if the confidence level of current shooting fragment is less than threshold
Value, then it is non-fragment interested to show current shooting fragment, opens common screening-mode.
3. the intelligent camera lens image pickup method according to claim 2 using deep learning, it is characterised in that described to target
When object is perceived, space characteristics and the sequential spy of destination object are obtained using convolutional neural networks and Recognition with Recurrent Neural Network
Sign, the scene around clarification of objective key element and destination object is identified, characteristic element includes position, posture and behavior, obtains special
The information that sign key element is changed over time and changed, obtain the confidence level of current frame image.
4. the intelligent camera lens image pickup method according to claim 3 using deep learning, it is characterised in that in the shooting
Each two field picture all by identical convolutional neural networks carry out space characteristics extraction, then the convolutional neural networks output work as
Preceding timing information, the current timing information forms temporal aspect together with timing information before, so as to export present frame
The confidence level of the current clip of image, while timing information by current timing information and before is accumulated and exported.
5. the intelligent camera lens image pickup method according to claim 4 using deep learning, it is characterised in that the depth
During habit, a data set for including several video segments is built first, and each video segment carries out fragment interested and risen
Begin time point and the mark for terminating time point, is trained to obtain the deep neural network model for deep learning with the data set.
6. a kind of intelligent camera lens camera system using deep learning, it is characterised in that the system includes：
Perceive unit, for perceiving and identifying the scene around destination object, and the feature of destination object, the perception unit bag
Include human detection module, Attitude estimation module, scene perception module, human tracking module；
Positioning unit is detected, for carrying out detection positioning to shooting fragment, is checked whether as fragment interested；
Control unit, for switching shooting high speed mode and common screening-mode.
7. the intelligent camera lens camera system according to claim 6 using deep learning, it is characterised in that also include contrast
Unit, for comparing the confidence level of current clip and the magnitude relationship of threshold value.
8. the intelligent camera lens camera system according to claim 7 using deep learning, it is characterised in that also including length
When mnemon, timing information for current timing information and before, and the current clip of output current frame image
Confidence level, while the timing information to receiving is accumulated and exported to next length mnemon in short-term.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711098878.3A CN107864334A (en) | 2017-11-09 | 2017-11-09 | A kind of intelligent camera lens image pickup method and system using deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711098878.3A CN107864334A (en) | 2017-11-09 | 2017-11-09 | A kind of intelligent camera lens image pickup method and system using deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107864334A true CN107864334A (en) | 2018-03-30 |
Family
ID=61700353
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711098878.3A Pending CN107864334A (en) | 2017-11-09 | 2017-11-09 | A kind of intelligent camera lens image pickup method and system using deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107864334A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110221626A (en) * | 2019-06-06 | 2019-09-10 | 睿魔智能科技（深圳）有限公司 | One kind is with clapping control method, device, computer equipment and storage medium |
| WO2020015333A1 (en) * | 2018-07-19 | 2020-01-23 | 北京微播视界科技有限公司 | Video shooting method and apparatus, terminal device, and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101529890A (en) * | 2006-10-24 | 2009-09-09 | 索尼株式会社 | Imaging device and reproduction control device |
| CN104618656A (en) * | 2015-02-15 | 2015-05-13 | 联想(北京)有限公司 | Information processing method and electronic equipment |
| CN104967803A (en) * | 2015-07-01 | 2015-10-07 | 广东欧珀移动通信有限公司 | Video recording method and video recording device |
| CN105323466A (en) * | 2015-06-25 | 2016-02-10 | 维沃移动通信有限公司 | Shooting method based on scene, and terminal |
| CN105654506A (en) * | 2014-11-24 | 2016-06-08 | 韩国电子通信研究院 | Apparatus and method for analyzing motion |
| CN105701460A (en) * | 2016-01-07 | 2016-06-22 | 王跃明 | Video-based basketball goal detection method and device |
| CN106028134A (en) * | 2015-03-31 | 2016-10-12 | 泽普实验室公司 | Detect sports video highlights for mobile computing devices |
| CN106845374A (en) * | 2017-01-06 | 2017-06-13 | 清华大学 | Pedestrian detection method and detection means based on deep learning |
| CN107087211A (en) * | 2017-03-30 | 2017-08-22 | 北京奇艺世纪科技有限公司 | A kind of anchor shots detection method and device |
| CN107122736A (en) * | 2017-04-26 | 2017-09-01 | 北京邮电大学 | A kind of human body based on deep learning is towards Forecasting Methodology and device |
-
2017
- 2017-11-09 CN CN201711098878.3A patent/CN107864334A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101529890A (en) * | 2006-10-24 | 2009-09-09 | 索尼株式会社 | Imaging device and reproduction control device |
| CN105654506A (en) * | 2014-11-24 | 2016-06-08 | 韩国电子通信研究院 | Apparatus and method for analyzing motion |
| CN104618656A (en) * | 2015-02-15 | 2015-05-13 | 联想(北京)有限公司 | Information processing method and electronic equipment |
| CN106028134A (en) * | 2015-03-31 | 2016-10-12 | 泽普实验室公司 | Detect sports video highlights for mobile computing devices |
| CN105323466A (en) * | 2015-06-25 | 2016-02-10 | 维沃移动通信有限公司 | Shooting method based on scene, and terminal |
| CN104967803A (en) * | 2015-07-01 | 2015-10-07 | 广东欧珀移动通信有限公司 | Video recording method and video recording device |
| CN105701460A (en) * | 2016-01-07 | 2016-06-22 | 王跃明 | Video-based basketball goal detection method and device |
| CN106845374A (en) * | 2017-01-06 | 2017-06-13 | 清华大学 | Pedestrian detection method and detection means based on deep learning |
| CN107087211A (en) * | 2017-03-30 | 2017-08-22 | 北京奇艺世纪科技有限公司 | A kind of anchor shots detection method and device |
| CN107122736A (en) * | 2017-04-26 | 2017-09-01 | 北京邮电大学 | A kind of human body based on deep learning is towards Forecasting Methodology and device |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020015333A1 (en) * | 2018-07-19 | 2020-01-23 | 北京微播视界科技有限公司 | Video shooting method and apparatus, terminal device, and storage medium |
| GB2590204A (en) * | 2018-07-19 | 2021-06-23 | Beijing Microlive Vision Tech Co Ltd | Video shooting method and apparatus, terminal device, and storage medium |
| GB2590204B (en) * | 2018-07-19 | 2023-02-15 | Beijing Microlive Vision Tech Co Ltd | Method and apparatus for video shooting, terminal device and storage medium |
| US11611701B2 (en) | 2018-07-19 | 2023-03-21 | Beijing Microlive Vision Technology Co., Ltd | Method and apparatus for video shooting, terminal device and storage medium |
| CN110221626A (en) * | 2019-06-06 | 2019-09-10 | 睿魔智能科技（深圳）有限公司 | One kind is with clapping control method, device, computer equipment and storage medium |
| CN110221626B (en) * | 2019-06-06 | 2022-03-25 | 睿魔智能科技（深圳）有限公司 | Follow-shot control method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| RU2762142C1 (en) | Method and apparatus for determining the key point of the face, computer apparatus, and data storage | |
| CN111325089B (en) | Method and apparatus for tracking object | |
| JP2020194608A (en) | Living body detection device, living body detection method, and living body detection program | |
| CN110264493B (en) | Method and device for tracking multiple target objects in motion state | |
| EP3648448A1 (en) | Target feature extraction method and device, and application system | |
| TWI517716B (en) | Mobile computing device having video-in-video real-time broadcasting capability | |
| CN106559645B (en) | Monitoring method, system and device based on camera | |
| CN110830756A (en) | Monitoring method and device | |
| WO2005041579A3 (en) | Method and system for processing captured image information in an interactive video display system | |
| CN101188677A (en) | Imaging apparatus, image processing apparatus, image processing method and computer program for execute the method | |
| CN107395957B (en) | Photographing method and device, storage medium and electronic equipment | |
| KR20130027520A (en) | Viewpoint detector based on skin color area and face area | |
| US20190384965A1 (en) | A method for selecting frames used in face processing | |
| CN107105193B (en) | Robot monitoring system based on human body information | |
| CN107111371B (en) | method, device and terminal for displaying panoramic visual content | |
| CN109886995B (en) | Multi-target tracking method in complex environment | |
| US9323989B2 (en) | Tracking device | |
| CN105631422A (en) | Video identification method and video identification system | |
| WO2022040886A1 (en) | Photographing method, apparatus and device, and computer-readable storage medium | |
| US8817125B2 (en) | Gesture recognition using chroma-keying | |
| US20130308829A1 (en) | Still image extraction apparatus | |
| CN114332744A (en) | Transformer substation self-adaptive security method and system based on machine vision | |
| CN107864334A (en) | A kind of intelligent camera lens image pickup method and system using deep learning | |
| CN106650583A (en) | Face detection method, face detection device and terminal equipment | |
| WO2022041182A1 (en) | Method and device for making music recommendation |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180330 |