CN105225249B - High ferro platform crosses the border detection method - Google Patents

High ferro platform crosses the border detection method Download PDF

Info

Publication number
CN105225249B
CN105225249B CN201410715458.5A CN201410715458A CN105225249B CN 105225249 B CN105225249 B CN 105225249B CN 201410715458 A CN201410715458 A CN 201410715458A CN 105225249 B CN105225249 B CN 105225249B
Authority
CN
China
Prior art keywords
image
platform
detection
pattern
coordinate
Prior art date
Application number
CN201410715458.5A
Other languages
Chinese (zh)
Other versions
CN105225249A (en
Inventor
俞大海
王敬华
岳明
舒明
Original Assignee
天津光电高斯通信工程技术股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 天津光电高斯通信工程技术股份有限公司 filed Critical 天津光电高斯通信工程技术股份有限公司
Priority to CN201410715458.5A priority Critical patent/CN105225249B/en
Publication of CN105225249A publication Critical patent/CN105225249A/en
Application granted granted Critical
Publication of CN105225249B publication Critical patent/CN105225249B/en

Links

Abstract

A kind of high ferro platform crosses the border detection method, can be arranged as required to monitoring regional extent, monitors the red floor tile region on the outside of white line, when passenger is when platform is waited, if into platform monitoring area, system sends out prompting alarm automatically, and prompting picks up attendant and passengers please far from danger zone.Additionally, due to the technology for employing real-time background updating, the present invention can overcome illumination variation round the clock and the external environments such as daily Changes in weather influence throughout the year, and natural conditions variation is avoided to interfere detection.

Description

High ferro platform crosses the border detection method

Technical field

The present invention relates to the technical field of picture recognition, specifically a kind of high ferro platform crosses the border detection method.

Background technology

The detection of high ferro platform is mainly monitored the red floor tile region of platform edge, since high ferro platform is than in platform High-speed railway rail will height, so close to the region be the easy danger for occurring to be slipped to from platform on track when there is passenger, meanwhile, When high ferro is arrived at a station, passenger also be easy to cause risk of collision into this region.High ferro platform from the platform region that passenger waits to Inner orbit region of standing is divided into, and platform is waited area, and yellow sidewalk for visually impaired people area, cross the border warning white line, waits and forbids entering red floor tile Region and platform track for a train region.

Invention content

It crosses the border detection method the technical problem to be solved in the present invention is to provide a kind of high ferro platform.

The present invention is adopted the technical scheme that solve technical problem present in known technology:

The high ferro platform of the present invention crosses the border detection method, includes the following steps:

A, detection starts, loading system parameter;

B, image is loaded into, camera is carried out and self-adjusting judgement is carried out to image;

C, image preprocessing establishes image buffers;

D, background modeling is carried out to image, determines the detection range in platform region:The region position detected according to actual needs It puts and clicks corresponding tie point on the video images with size;Selected dot sequency is connected, confirms marked region;Mark Determine the safe line position of white on platform, prediction drop shadow spread of the calibration people on platform, since there are illumination, the shadows in environment Range needs be excluded from image procossing, so need predict drop shadow spread;By region labeling point, white safety line and people The image coordinate of prediction drop shadow spread is combined, as the systematic parameter of detection zone, when reloading systematic parameter, according to The parameter coordinate of detection zone automatically generates the image template of detection zone, realizes the real image extraction to detection zone;

E, foreground extraction is carried out to image;

F, judge whether the foreground object in image is detection target, i.e., whether is people on platform;

G, during the non-inlet parking of train, the detection target in tracking prospect enters above-mentioned detection when there is target in image In region, i.e., when someone crosses the border on platform, system carries out geofence;When pull in stopping after, detection zone switchs to can be into Enter region, detection of crossing the border stops, and system is no longer alarmed.

The present invention can also use following technical measures:

The detection zone parameter calibration formula is:I (x, y)=α * I ' (x, y)+δ,

I (x, y) is coordinate in image, and I ' (x, y) is practical platform coordinate, and α is transformation coefficient, and δ is empirical parameter.

Assuming that the real space coordinate in prosecution region, in the plane of coordinate system Z '=0, detection zone is by a group echo point Connection composition, is denoted as I 'i(xi, yi), while corresponding mark point I is found in camera image coordinate system Zi(xi, yi) and record Coordinate value calculates coordinate transform factor alpha by coordinate transform.

Adaptive Gauss background modeling is carried out to detection zone, by the foreground and background in perimeter detection area image point It opens, while is analyzed for prospect.

The Gaussian Background modeling includes the following steps:

A, each new pixel value XtIt is compared as the following formula with current K model, point until finding matching new pixel value Cloth model, i.e., with the mean bias of the model in 2.5 σ

|XtI, t-1|≤2.5σI, t-1

If b, the matched pattern of institute meets context request, which belongs to background, otherwise belongs to prospect;

C, each schema weight updates as follows, and wherein α is learning rate, for matched pattern MK, t=1, otherwise MK, t=0, then the weight of each pattern be normalized;

wK, t=(1- α) * wK, t-1+α*MK, t

D, the mean μ of non-match pattern and standard deviation sigma are constant, and the parameter of match pattern is updated according to equation below:

ρ=α * η (Xtk, σk)

μt=(1- ρ) * μt-1+ρ*Xt

If e, not having any pattern match in a steps, the pattern of weight minimum is replaced, i.e. the mean value of the pattern is Current pixel, standard deviation are initialization higher value, and weight is smaller value;

F, each pattern is according to w/ α2It arranges in descending order, weight is big, and the small pattern arrangement of standard deviation is forward;

G, for B pattern as background, B meets following formula before selecting, and parameter T represents background proportion,

The invention has the advantages and positive effects that:

The high ferro platform of the present invention crosses the border detection method, can be arranged as required to monitoring regional extent, and main monitoring exists Red floor tile region on the outside of white line, when passenger is when platform is waited, if into platform monitoring area, system sends out prompting automatically Alarm, prompting pick up attendant and passengers please far from danger zone.Additionally, due to the technology for employing real-time background updating, this hair It is bright to overcome illumination variation round the clock and the throughout the year influence of the external environments such as daily Changes in weather, avoid natural conditions variation pair Detection interferes.

Description of the drawings

The high ferro platform that Fig. 1 is the present invention crosses the border the flow chart of detection method.

Specific embodiment

The high ferro platform of the present invention crosses the border detection method, includes the following steps:

A, detection starts, loading system parameter;

B, image is loaded into, camera is carried out and self-adjusting judgement is carried out to image, is sentenced by carrying out Image Adjusting to infrared image It is disconnected, it avoids detecting mistake caused by carrying out image pixel adjustment due to camera, loading figure again after then adjusting if you need to self-adjusting Picture;

C, image preprocessing establishes image buffers, and infrared image Gaussian smoothing is pre-processed, and reduces picture noise to detection The influence of arithmetic result;

D, background modeling is carried out to image, determines the detection range in platform region:The region position detected according to actual needs It puts and clicks corresponding tie point on the video images with size;Selected dot sequency is connected, confirms marked region;Mark Determine the safe line position of white on platform, prediction drop shadow spread of the calibration people on platform;By region labeling point, white safety line The image coordinate for predicting drop shadow spread with people is combined, as the systematic parameter of detection zone, when reloading systematic parameter, The image template of detection zone is automatically generated according to the parameter coordinate of detection zone, realizes and the real image of detection zone is carried It takes;

E, foreground extraction is carried out to image;

F, judge whether the foreground object in image is detection target, i.e., whether is people on platform;

G, during the non-inlet parking of train, tracing detection target, calculate target position and warning zone with the presence or absence of sky Between on coincidence, when there is target to enter in above-mentioned detection zone in image, i.e., when someone crosses the border on platform, system is crossed the border Alarm;When pull in parking after, detection zone switchs to accessible area, and detection of crossing the border stops, and system no longer alarms.

The detection zone parameter calibration formula is:I (x, y)=α * I ' (x, y)+δ,

I (x, y) is coordinate in image, and I ' (x, y) is practical platform coordinate, and α is transformation coefficient, and δ is empirical parameter.

Assuming that the real space coordinate in prosecution region, in the plane of coordinate system Z '=0, detection zone is by a group echo point Connection composition, is denoted as I 'i(xi, yi), while corresponding mark point I is found in camera image coordinate system Zi(xi, yi) and record Coordinate value calculates coordinate transform factor alpha by coordinate transform.

Adaptive Gauss background modeling is carried out to detection zone, by the foreground and background in perimeter detection area image point It opens, while is analyzed for prospect.

Gaussian Background modeling includes the following steps:

A, each new pixel value XtIt is compared as the following formula with current K model, point until finding matching new pixel value Cloth model, i.e., with the mean bias of the model in 2.5 σ

|XtI, t-1|≤2.5σI, t-1

If b, the matched pattern of institute meets context request, which belongs to background, otherwise belongs to prospect;

C, each schema weight updates as follows, and wherein α is learning rate, for matched pattern MK, t=1, otherwise MK, t=0, then the weight of each pattern be normalized;

wK, t=(1-α)*wK, t-1+α*MK, t

D, the mean μ of non-match pattern and standard deviation sigma are constant, and the parameter of match pattern is updated according to equation below:

ρ=α * η (Xtk, σk)

μt=(1- ρ) * μt-1+ρ*Xt

If e, not having any pattern match in a steps, the pattern of weight minimum is replaced, i.e. the mean value of the pattern is Current pixel, standard deviation are initialization higher value, and weight is smaller value;

F, each pattern is according to w/ α2It arranges in descending order, weight is big, and the small pattern arrangement of standard deviation is forward;

G, for B pattern as background, B meets following formula before selecting, and parameter T represents background proportion;

Due to there are the influence that scene illumination changes, for example, the cloudy day is different with brightness in fine day platform;Daytime is due to the sun Position is different, so sunlight is different in the range of platform;Night exists since platform opens headlamp, and brightness becomes high ring Border changes.The method that we employ real-time background updating carries out Adaptive Gauss model background modeling for detection zone, will The foreground and background that platform crosses the border in detection zone image separates, meanwhile, it is analyzed for prospect.Meanwhile with reference to detection zone Learning rate parameter during the brightness of image variation adjustment background modeling in domain, the wherein variation range of parameter 0.01~ Between 0.001, algorithm is made to greatly improve the Stability and veracity of target detection.

The above described is only a preferred embodiment of the present invention, not make limitation in any form to the present invention, though The right present invention has been described by way of example and in terms of the preferred embodiments, however, the present invention is not limited to, any technology people for being familiar with this profession Member without departing from the scope of the present invention, can utilize the technology contents disclosed to make a little change or modification certainly, into For the equivalent embodiment of equivalent variations, as long as being the content without departing from technical solution of the present invention, technical spirit according to the present invention To any simple modification, equivalent change and modification that above example is made, belong in the range of technical solution of the present invention.

Claims (3)

  1. The detection method 1. a kind of high ferro platform crosses the border, which is characterized in that include the following steps:
    A, detection starts, loading system parameter;
    B, image is loaded into, camera is carried out and self-adjusting judgement is carried out to image;
    C, image preprocessing establishes image buffers;
    D, background modeling is carried out to image, determines the detection range in platform region:The regional location that detects according to actual needs and Size clicks corresponding tie point on the video images;Selected dot sequency is connected, confirms marked region;Calibration station The safe line position of white on platform, prediction drop shadow spread of the calibration people on platform;By region labeling point, white safety line and people The image coordinate of prediction drop shadow spread is combined, as the systematic parameter of detection zone, when reloading systematic parameter, according to The parameter coordinate of detection zone automatically generates the image template of detection zone, realizes the real image extraction to detection zone;
    E, foreground extraction is carried out to image;
    F, judge whether the foreground object in image is detection target, i.e., whether is people on platform;
    G, during the non-inlet parking of train, the detection target in tracking prospect enters above-mentioned detection zone when there is target in image In, i.e., when someone crosses the border on platform, system carries out geofence;When pull in stopping after, detection zone switchs to that area can be entered Domain, detection of crossing the border stop, and system is no longer alarmed;
    Wherein, the parameter calibration formula of detection zone is:I (x, y)=α * I ' (x, y)+δ,
    I (x, y) is coordinate in image in this formula, and I ' (x, y) is practical platform coordinate, and α is transformation coefficient, and δ is empirical parameter;
    In the plane of coordinate system Z '=0, detection zone is made of the real space coordinate of detection zone group echo point connection, It is denoted as I 'i(xi, yi), while corresponding mark point I is found in camera image coordinate system Zi(xi, yi), and coordinate value is recorded, it passes through It crosses coordinate transform and calculates coordinate transform factor alpha.
  2. The detection method 2. high ferro platform according to claim 1 crosses the border, it is characterised in that:Detection zone is carried out adaptive Gaussian Background models, and the foreground and background in perimeter detection area image is separated, while is analyzed for prospect.
  3. The detection method 3. high ferro platform according to claim 2 crosses the border, it is characterised in that:Adaptive Gauss background modeling packet Include following steps:
    A, each new pixel value XtIt is compared as the following formula with current K model, the distributed mode until finding matching new pixel value Type, i.e., with the mean bias of the model in 2.5 σ
    |XtI, t-1|≤2.5σI, t-1
    If b, the matched pattern of institute meets context request, which belongs to background, otherwise belongs to prospect;
    C, each schema weight updates as follows, and α is learning rate in formula, for matched pattern MK, t=1, otherwise MK, t =0, then the weight of each pattern be normalized;
    wK, t=(1- α) * wK, t-1+α*MK, t
    D, the mean μ of non-match pattern and standard deviation sigma are constant, and the parameter of match pattern is updated according to equation below:
    ρ=α * η (Xtk, σk)
    μt=(1- ρ) * μt-1+ρ*Xt
    If e, not having any pattern match in a steps, the pattern of weight minimum is replaced, i.e. the mean value of the pattern is current Pixel, standard deviation are initialization higher value, and weight is smaller value;
    F, each pattern is according to w/ α2It arranges in descending order, weight is big, and the small pattern arrangement of standard deviation is forward;
    G, for B pattern as background, B meets following formula before selecting, and parameter T represents background proportion,
CN201410715458.5A 2014-12-01 2014-12-01 High ferro platform crosses the border detection method CN105225249B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410715458.5A CN105225249B (en) 2014-12-01 2014-12-01 High ferro platform crosses the border detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410715458.5A CN105225249B (en) 2014-12-01 2014-12-01 High ferro platform crosses the border detection method

Publications (2)

Publication Number Publication Date
CN105225249A CN105225249A (en) 2016-01-06
CN105225249B true CN105225249B (en) 2018-06-12

Family

ID=54994198

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410715458.5A CN105225249B (en) 2014-12-01 2014-12-01 High ferro platform crosses the border detection method

Country Status (1)

Country Link
CN (1) CN105225249B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002197445A (en) * 2000-12-26 2002-07-12 Ced System Inc Detector for abnormality in front of train utilizing optical flow
JP2007317052A (en) * 2006-05-29 2007-12-06 Ced System Inc System for measuring waiting time for lines
CN101635833A (en) * 2008-07-22 2010-01-27 深圳市朗驰欣创科技有限公司 Method, device and system for video monitoring
CN102663405A (en) * 2012-05-14 2012-09-12 武汉大学 Prominence and Gaussian mixture model-based method for extracting foreground of surveillance video
JP5045449B2 (en) * 2008-01-17 2012-10-10 株式会社明電舎 Environment-adaptive intruder detection device by image processing
CN103488993A (en) * 2013-09-22 2014-01-01 北京联合大学 Crowd abnormal behavior identification method based on FAST

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002197445A (en) * 2000-12-26 2002-07-12 Ced System Inc Detector for abnormality in front of train utilizing optical flow
JP2007317052A (en) * 2006-05-29 2007-12-06 Ced System Inc System for measuring waiting time for lines
JP5045449B2 (en) * 2008-01-17 2012-10-10 株式会社明電舎 Environment-adaptive intruder detection device by image processing
CN101635833A (en) * 2008-07-22 2010-01-27 深圳市朗驰欣创科技有限公司 Method, device and system for video monitoring
CN102663405A (en) * 2012-05-14 2012-09-12 武汉大学 Prominence and Gaussian mixture model-based method for extracting foreground of surveillance video
CN103488993A (en) * 2013-09-22 2014-01-01 北京联合大学 Crowd abnormal behavior identification method based on FAST

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"行人异常智能视频监控系统研究与实现";赵凤娟;《中国优秀硕士学位论文全文数据库 信息科技辑》;20111215(第12期);论文第13-28、63页 *

Also Published As

Publication number Publication date
CN105225249A (en) 2016-01-06

Similar Documents

Publication Publication Date Title
CN100533482C (en) Image processing techniques for a video based traffic monitoring system and methods therefor
US7965866B2 (en) System and process for detecting, tracking and counting human objects of interest
JP2019067778A (en) Image sensor
US9158979B2 (en) Vehicle vicinity monitoring device
CN101918980A (en) Runway surveillance system and method
US20130229518A1 (en) Systems and methods that employ object recognition
US20150279051A1 (en) Image detection and processing for building control
US20130136307A1 (en) Method for counting objects and apparatus using a plurality of sensors
CN101739686B (en) Moving object tracking method and system thereof
WO2016014930A3 (en) A vision-based system for dynamic weather detection
CN102073851A (en) Method and system for automatically identifying urban traffic accident
US9456183B2 (en) Image processing occupancy sensor
US9988238B2 (en) Elevator dispatch using facial recognition
JP2013008460A (en) Lighting device
CN102819263A (en) Multi-camera visual perception system for UGV (Unmanned Ground Vehicle)
CN106144816B (en) Occupant detection based on depth transducer
CN104828664A (en) Automatic debugging system and method
CN101382497A (en) Visibility detecting method based on monitoring video of traffic condition
CN201425781Y (en) Traffic signal lamps based on visitors flowrate
MX2012014267A (en) Determining a pixel classification threshold for vehicle occupancy detection.
CN102946663A (en) Image recognition intelligent illumination control system and control method thereof
US10513415B2 (en) Depth sensor based passenger sensing for passenger conveyance control
CN103049787A (en) People counting method and system based on head and shoulder features
CN101847206A (en) Pedestrian traffic statistical method and system based on traffic monitoring facilities
CN101552910B (en) Remnant detection device based on comprehensive computer vision

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant