CN109068105A - A kind of prison video monitoring method based on deep learning - Google Patents
A kind of prison video monitoring method based on deep learning Download PDFInfo
- Publication number
- CN109068105A CN109068105A CN201811091834.2A CN201811091834A CN109068105A CN 109068105 A CN109068105 A CN 109068105A CN 201811091834 A CN201811091834 A CN 201811091834A CN 109068105 A CN109068105 A CN 109068105A
- Authority
- CN
- China
- Prior art keywords
- prison
- criminal
- monitoring
- target detection
- deep learning
- 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.)
- Withdrawn
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Signal Processing (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Alarm Systems (AREA)
Abstract
The prison video monitoring method based on deep learning that the present invention provides a kind of, the function that can be realized facilitate prison officer to check first is that filter out the target monitoring there are potential danger from monitoring all in prison；There is the picture of the undisciplined violation of criminal second is that automatically saving in the monitoring of prison.Specific method is to monitor image data by mark prison and assemble for training to practice the target detection network model based on deep learning；Prison video monitoring is grouped, present frame is read to each group of circulation and is detected；Setting respective rule judges whether to be potentially dangerous and the undisciplined violation of criminal.Comprehensively consider the practicability of detection accuracy, speed and judgment rule, the present invention can provide a kind of economy, efficiently scheduling and monitoring method for existing prison video monitoring system.
Description
Technical field
The present invention relates to prison video surveillance management fields more particularly to a kind of based on deep learning algorithm of target detection
The scheduling of prison video monitoring and criminal's unlawful practice grasp shoot method.
Background technique
Background of related of the invention is illustrated below, but these explanations might not constitute it is of the invention existing
Technology.
The prison of broad sense refers to the place for putting in prison all convicts, including prison, detention house, minor reformatory etc., criminal
The behaviors such as escape from prison, commit suiside, fighting, assaulting police seriously affect social security and the safety of jails.
To ensure supervision safety, the monitoring camera inside prison passes through dedicated Internet access video throughout each corner
Equipment is stored, real time monitoring over multiple terminals is realized by stream media technology and has access to the functions such as video recording.However, in prison
Monitoring camera quantity is likely to be breached several hundred or even thousands of, almost impossible by all monitoring of manpower real time inspection；Moreover,
Monitored picture in most times be it is normal, that criminal is not undisciplined or potential danger, checks be easy to cause work for a long time
The visual fatigue of personnel.
Existing monitoring method is to randomly select or several or more than ten according to the sequential selection being previously set mostly
Monitoring display, then gone to judge whether there is the undisciplined violation of criminal or potential danger by staff.It is this to be supervised in real time by manpower
The monitoring limited amount that the method for control is checked causes many dangerous situations to send out well below the total amount of monitoring camera in prison
Now not in time.
Application No. is CN201711231668.7, the date of application is a kind of intelligence of disclosure of the invention on November 30th, 2017
Video analysis and monitoring system judge that method of the foreground target in warning region can use the videos such as enclosure wall circumference at the prison
It in monitoring, realizes and detects and alarm close to the criminal of enclosure wall or other objects to crossing power grid, improve anti-escaping ability；But
Be it is internal at the prison, criminal life area, site for labour and road etc. do not have warning region, but occur criminal's suicide,
Fight, the main place of undisciplined violation, it is therefore desirable to a kind of method can be realized to all areas in prison occur it is dangerous into
Row alarm.
Application No. is CN201610758302.4, the date of application is a kind of prison of the disclosure of the invention on the 30th of August in 2016
Video monitoring system, the system need to wear wrist strap for each criminal, judge danger by wrist strap data, utilize less radio-frequency
Then the three-dimensional coordinate of locating criminals calls video monitoring to carry out real time inspection.The system needs to consider wrist in practical applications
The problems such as band charging, system complexity and cost.
In recent years, convolutional neural networks and deep learning were in computer vision field acquirement important breakthrough, wherein a system
Such as Faster-RCNN of the algorithm of target detection based on deep learning is arranged, all progress is significant in speed and precision by YOLO-V3 etc.,
Can rapidly and accurately identify each object category for including in picture, and with the coordinate of rectangle frame positioning object, security protection,
The fields such as automatic Pilot have practical value.
Summary of the invention
To solve the problems, such as that above-mentioned prison video monitoring exists, the workload of relevant staff is reduced, is realized to prison
The real time monitoring of interior potential danger and criminal's unlawful practice, the present invention propose a kind of prison video prison based on deep learning
Prosecutor method.
First is that being screened in all monitoring using the algorithm of target detection based on deep learning, there are the monitoring of potential danger
Picture is shown that automatically and efficiently schedule video monitors by the department for enabling checking monitoring and personnel.
Second is that judging automatically according to the state of common items and criminal with the presence or absence of criminal's unlawful practice, and save phase
Foundation of the picture answered as supervisor's notification.
To complete above-mentioned two function, technical solution of the present invention the following steps are included:
Step 1) establishes data set: intercepting the frame in multiple prison video monitorings, is manually marked, the class of mark includes crime
Criminal, people's police, clothes, lorry, emphasis tool (crowbar, abrasive machine, all kinds of cutters etc.), safety cap, books and other it is common or
The object that person should be concerned；
Step 2 training objective detection model: building the target detection convolutional neural networks based on deep learning, with previous step mark
The data set training of the note network, obtains the target detection model for being suitable for prison；
Step 3) is grouped detection to monitoring camera: according to place prison, installation site etc. by monitoring camera in prison
It is grouped, to each group of monitoring, circulation reads the present frame of each camera, is sent into target detection model and is detected；
Step 4) is post-processed according to testing result: first is that judging whether to be potentially dangerous, if any then by prison belonging to the frame
It controls camera and is labeled as target monitoring；Second is that judge whether there is criminal's unlawful practice, if any then saving the frame；
Step 5) circulation terminates, and target monitoring is drawn in prison monitoring room and prison command centre (monitoring center) by screen more
Face is shown；By the criminal of preservation violate disciplines behavior frame be used as video supervise and examine notification.
As alternative dispensing means of the invention, step 1) and step 2 target detection model obtained can be by one kind
The target detection model increased income on the internet at present replaces.
Potential danger Rule of judgment.1) criminal's quantity is 1 people or 2 people: prison is as the place for putting in prison criminal, it is necessary to energy
It enough monitors criminal in real time, especially in individual criminal's independent action, the serious supervision safety such as commit suiside, run away, fighting easily occurs
Therefore accident counts criminal's number in post-processing stages, setting there are potential danger a adequate condition is criminal as 1 people
Or 2 people.2) there are lorries in testing result: because prison production and living need, often having lorry to pass in and out prison, mentions to some criminals
For the condition for lorry flight of hiding oneself, therefore, setting in testing result there are lorry is allowed there are another adequate condition of potential danger
Prison officer grasps wagon perimeter situation by monitor video in real time.3) other customized special cases.
Criminal's unlawful practice Rule of judgment.1) clothes is detected other than room drying: disobeying corresponding to what is disorderly hung clothes
Travel notes are.2) criminal and books are detected after prison is gone to bed: is late for the activities against discipline going to bed corresponding to criminal.3) it is requiring
The labour on-site supervision picture to wear a safety helmet detects that the rectangle frame of criminal and safety cap is handed over and is zero than (IOU), shows the criminal
It does not wear a safety helmet: corresponding to the unlawful practice not worn a safety helmet.4) criminal but no people's police are detected in people's police office space:
Enter the unlawful practice of people's police office without authorization corresponding to criminal.5) other detectable unlawful practices.
The invention adopts the above technical scheme compared with prior art, has following technical effect that
1. guaranteeing that system is stablized without changing existing monitoring system；
2. it realizes simply, it is low in cost, it is only necessary to which that training pattern and some calculating equipment of increase can be in existing video monitorings
System increases repertoire of the present invention；
3. considerably reducing the workload of relevant staff；
4. realize to the real time inspections of monitoring all in prison, effectively prevents to commit suiside in prison, escape from prison, fight, undisciplined violation etc.
Behavior；
5. can be customized according to user demand about the post-processing module for judging potential danger and undisciplined violation judgement, have fine
Flexibility and scalability.
Detailed description of the invention
Attached drawing 1 is in monitoring method of the present invention to the target detection of each monitoring present frame and post-processing flow chart.
Attached drawing 2 is the system construction drawing that the present invention realizes monitoring and scheduling and undisciplined candid photograph.
Attached drawing 3 is the pseudo-code of the algorithm that the present invention carries out circulation target detection to monitoring present frame.
Specific embodiment
In order to keep the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, pseudocode and
Attached drawing is further elaborated.
The present invention is examined using real-time pictures of the algorithm of target detection based on deep learning to monitor video in prison
Survey, in conjunction with testing result precision and different scenes actual conditions be arranged respective rule, judge whether there is potential danger and
Criminal violates disciplines behavior, and by potentially dangerous target monitoring prison duty room and command centre's (monitoring center) into
The frame for having the behavior that violates disciplines is saved the picture foundation as supervisor's notification by row display.
When establishing data set, it is labeled using previous video supervisor's notification picture, it is also possible under several scenes
Monitoring video cutting framing obtains.
Model training stage, can be in the algorithm of target detection (Faster-RCNN, YOLO etc.) arbitrarily based on deep learning
(fine-tuning) is finely adjusted on the model increased income；If the data set of mark is sufficiently large, from the beginning building network can open
Begin to train；If not establishing data set, part risk judgement (number can be equally directly completed using the model increased income
Statistics, lorry detection) and unlawful practice judgement (clothes detection etc.).
Purpose to monitoring camera grouping is first is that the mode with prison two-level management is adapted, second is that the monitoring of different groups
It can be performed in parallel in each calculating equipment, greatly the time used in shortening one cycle.The detection and post-processing of frame
Process is shown in Fig. 1.
Assuming that 50 first group of monitoring camera, totally 20 groups (specific number can appoint according to system functional requirement and time requirement
Meaning setting).
Attached drawing 3 is shown in the pseudo-code realization of each group of video cycle detection.
After circulation terminates, will be monitored in target monitoring list in this prison monitoring room and prison command centre (in monitoring
The heart) it is shown to operator on duty；The frame of preservation is used as video supervisor's notification.The system knot of video monitoring scheduling and undisciplined illegal behaviour capturing
Structure is shown in Fig. 2.
A picture is detected under graphics processor (GPU) hardware environment currently based on the algorithm of target detection of deep learning
One second even shorter time is only needed, while guaranteeing detection accuracy, therefore one cycle (50 monitoring) can be completed in 1 minute,
I.e. the refresh time of monitored picture is one minute, can satisfy the requirement of real time monitoring.
It was found that the Sample Scenario of potential danger: being stayed outside police's sight assuming that certain criminal is detached from Criminal Group, although prison
Control camera can photograph but monitoring center does not open the monitored picture, and the danger probably will not under conventional method
It is found；Under deep learning object detection method provided by the invention, within one minute, which is detected to only have
1 criminal is marked as target monitoring immediately and is shown to operator on duty, therefore, as long as criminal's independent action is super after one minute, claps
Monitoring to this situation is inherently shown in monitoring center.
It was found that the example of criminal's unlawful practice: certain criminal hangs over the head of a bed in order to block monitoring camera, by clothes, when
Target detection model gets the monitoring when the current frame, detects clothes, current to supervise according to unlawful practice judgment rule
It must not hang clothes under control, so the frame is saved as the picture foundation of video supervisor's notification, the target detection based on deep learning
Algorithm can effectively hit the idea of leaving things to chance of criminal to the judgement near real-time of such unlawful practice almost without omission.
Although referring to illustrative embodiments, invention has been described, but it is to be understood that the present invention does not limit to
The specific embodiment that Yu Wenzhong is described in detail and shows, without departing from claims limited range, this
Field technical staff can make various changes to the illustrative embodiments.
Claims (4)
1. a kind of prison video monitoring method based on deep learning, by the algorithm of target detection in computer vision technique from
The monitoring to merit special attention and frame are screened in the prison video monitoring data of magnanimity, it is characterised in that are included the following steps, walked
Rapid one: the present frame in the monitor video of prison being detected using the target detection model based on deep learning；Step 2: root
Potential danger and criminal's unlawful practice are judged whether there is according to testing result and setting respective rule；Step 3: it will be present
The target monitoring of potential danger is shown in prison monitoring room and prison command centre (monitoring center), and the undisciplined violation of criminal will be present
The frame of behavior automatically saves the foundation as video supervise and examine notification.
2. the target detection model in claim 1 described in step 1 based on deep learning refers to that computer vision field is applicable in
In the convolutional neural networks of target detection (object detection) task, it is characterised in that can be obtained by internet
The target detection network model of open source, or by mark prison video monitoring image data collection, build target detection network come
Training, which obtains, is more suitable the target detection network model in prison.
3. the judgment rule in claim 1 described in step 2 there are potential danger be may occur escape from prison, commit suiside, beating
The condition of the events such as frame is characterized in that including: 1 to 2 criminal of discovery in testing result；It was found that lorry；In specific time crime
Violate the region discovery criminal that should not occur；Customized potential danger situation under other special screnes.
4. in claim 1 described in step 2 there are criminal's unlawful practice judgment rule be according to prison about criminal manage
The rules and regulations of reason are able to detect in target detection model to be arranged in range, characterized by comprising: in sunning room with outgoing
Existing clothes；The labour scene with safety cap is being required to find that criminal does not wear a safety helmet；In bedtime section discovery criminal's reading；Crime
Criminal enters people's police office when people's police are absent from the scene；Other detectable criminal's unlawful practices.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811091834.2A CN109068105A (en) | 2018-09-20 | 2018-09-20 | A kind of prison video monitoring method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811091834.2A CN109068105A (en) | 2018-09-20 | 2018-09-20 | A kind of prison video monitoring method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109068105A true CN109068105A (en) | 2018-12-21 |
Family
ID=64763155
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811091834.2A Withdrawn CN109068105A (en) | 2018-09-20 | 2018-09-20 | A kind of prison video monitoring method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109068105A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490126A (en) * | 2019-08-15 | 2019-11-22 | 成都睿晓科技有限公司 | A kind of safety cabinet security management and control system based on artificial intelligence |
| CN110516538A (en) * | 2019-07-16 | 2019-11-29 | 广州中科凯泽科技有限公司 | The double violation assessment method of leaving the post in prison based on deep learning target detection |
| CN111027463A (en) * | 2019-12-06 | 2020-04-17 | 江西洪都航空工业集团有限责任公司 | Wall turning detection method based on video analysis |
| CN111091098A (en) * | 2019-12-20 | 2020-05-01 | 浙江大华技术股份有限公司 | Training method and detection method of detection model and related device |
| CN111274872A (en) * | 2020-01-08 | 2020-06-12 | 哈尔滨融智爱科智能科技有限公司 | Template matching-based video monitoring dynamic irregular multi-supervision-area distinguishing method |
| CN111881865A (en) * | 2020-08-03 | 2020-11-03 | 南京奥拓电子科技有限公司 | Self-adaptive dangerous behavior monitoring method and system and intelligent equipment |
| CN113392263A (en) * | 2021-06-24 | 2021-09-14 | 上海商汤科技开发有限公司 | Data labeling method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201315654Y (en) * | 2008-08-25 | 2009-09-23 | 云南正卓信息技术有限公司 | Special SkyEyes** intelligent monitoring system for prison |
| CN102867236A (en) * | 2012-08-30 | 2013-01-09 | 都伊林 | Integrated Internet of things application system for prison |
| CN106303438A (en) * | 2016-08-26 | 2017-01-04 | 无锡卓信信息科技股份有限公司 | A kind of indoor video monitoring system for Prison staff |
| CN106339734A (en) * | 2016-08-30 | 2017-01-18 | 无锡卓信信息科技股份有限公司 | Positioning based prison video monitoring system |
-
2018
- 2018-09-20 CN CN201811091834.2A patent/CN109068105A/en not_active Withdrawn
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201315654Y (en) * | 2008-08-25 | 2009-09-23 | 云南正卓信息技术有限公司 | Special SkyEyes** intelligent monitoring system for prison |
| CN102867236A (en) * | 2012-08-30 | 2013-01-09 | 都伊林 | Integrated Internet of things application system for prison |
| CN106303438A (en) * | 2016-08-26 | 2017-01-04 | 无锡卓信信息科技股份有限公司 | A kind of indoor video monitoring system for Prison staff |
| CN106339734A (en) * | 2016-08-30 | 2017-01-18 | 无锡卓信信息科技股份有限公司 | Positioning based prison video monitoring system |
Non-Patent Citations (1)
| Title |
|---|
| 尹向兵，王宏群: "监狱视频识别报警***的研究与实现", 《电脑知识与技术》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516538A (en) * | 2019-07-16 | 2019-11-29 | 广州中科凯泽科技有限公司 | The double violation assessment method of leaving the post in prison based on deep learning target detection |
| CN110516538B (en) * | 2019-07-16 | 2022-10-11 | 广州中科凯泽科技有限公司 | Prison double off-duty violation assessment method based on deep learning target detection |
| CN110490126A (en) * | 2019-08-15 | 2019-11-22 | 成都睿晓科技有限公司 | A kind of safety cabinet security management and control system based on artificial intelligence |
| CN110490126B (en) * | 2019-08-15 | 2023-04-18 | 成都睿晓科技有限公司 | Safe deposit box safety control system based on artificial intelligence |
| CN111027463A (en) * | 2019-12-06 | 2020-04-17 | 江西洪都航空工业集团有限责任公司 | Wall turning detection method based on video analysis |
| CN111091098A (en) * | 2019-12-20 | 2020-05-01 | 浙江大华技术股份有限公司 | Training method and detection method of detection model and related device |
| CN111091098B (en) * | 2019-12-20 | 2023-08-15 | 浙江大华技术股份有限公司 | Training method of detection model, detection method and related device |
| CN111274872A (en) * | 2020-01-08 | 2020-06-12 | 哈尔滨融智爱科智能科技有限公司 | Template matching-based video monitoring dynamic irregular multi-supervision-area distinguishing method |
| CN111274872B (en) * | 2020-01-08 | 2023-08-22 | 哈尔滨融智爱科智能科技有限公司 | Video monitoring dynamic irregular multi-supervision area discrimination method based on template matching |
| CN111881865A (en) * | 2020-08-03 | 2020-11-03 | 南京奥拓电子科技有限公司 | Self-adaptive dangerous behavior monitoring method and system and intelligent equipment |
| CN113392263A (en) * | 2021-06-24 | 2021-09-14 | 上海商汤科技开发有限公司 | Data labeling method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109068105A (en) | A kind of prison video monitoring method based on deep learning | |
| Laufs et al. | Security and the smart city: A systematic review | |
| US7428314B2 (en) | Monitoring an environment | |
| CN105426820B (en) | More people's anomaly detection methods based on safety monitoring video data | |
| CN113298444A (en) | Private cloud-based intelligent power safety control platform | |
| CN107317718B (en) | A kind of O＆M service management and management platform | |
| CN106454250A (en) | Intelligent recognition and early warning processing information platform | |
| CN106650584B (en) | Flame detecting method and system | |
| CN109413369A (en) | A kind of monitor video intellectual analysis early warning platform | |
| US20090046147A1 (en) | Monitoring an environment | |
| CN103108159A (en) | Electric power intelligent video analyzing and monitoring system and method | |
| JP2013235329A (en) | Face identification monitoring/management method | |
| CN113011833A (en) | Safety management method and device for construction site, computer equipment and storage medium | |
| CN110012268A (en) | Pipe network AI intelligent control method, system, readable storage medium storing program for executing and equipment | |
| CN111507574B (en) | Security personnel deployment method and device, computer equipment and storage medium | |
| Song et al. | Real-time monitoring for crowd counting using video surveillance and GIS | |
| US20210364356A1 (en) | System and method for using artificial intelligence to enable elevated temperature detection of persons using commodity-based thermal cameras | |
| CN112257494A (en) | Behavior recognition method based on intelligent video analysis technology and application | |
| CN113538825A (en) | Campus wall-turning event alarm method and system | |
| CN203327163U (en) | Monitoring system | |
| CN103093177A (en) | Face identification, detection and monitoring method | |
| KR20200052418A (en) | Automated Violence Detecting System based on Deep Learning | |
| CN106130762A (en) | A kind of network training comprehensive analysis method based on finite automaton | |
| CN115567690A (en) | Intelligent monitoring system capable of automatically identifying dangerous points of field operation | |
| CN111083449A (en) | Campus security system |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20181221 |