CN107705574A - A kind of precisely full-automatic capturing system of quick road violation parking - Google Patents

Abstract

A kind of precisely full-automatic capturing system of quick road violation parking provided by the invention, is made up of front end control point part, internet transmission of virtual laboratory and central manage-ment unit.Vehicle testing techniques mainly include the dynamic detection and vehicle tracking technology of vehicle.It make use of the method for testing motion of time-based motion history image and the wireless vehicle tracking of feature-point optical flow estimation matching.In order to reach more preferable motion detection effect, the method for removing shade is used, has more accurately detected moving vehicle, employed multi-characteristic points light stream detecting and tracking mode, with reference to occlusion solution, significantly improve tracking accuracy.Systemic-function is powerful, be sufficient for each application scenario, it is main include carrying out the ball machine of front end control point remote control, the function such as the preview of video monitoring, illegal parking automatic detection, rock mass displacements, high definition video recording and the association of illegal activities video, information of vehicles storage browse, system administration and remote maintenance.

Description

A kind of precisely full-automatic capturing system of quick road violation parking

Technical field

The present invention relates to a kind of parking offense automatic snapshot system, more particularly to a kind of precisely quickly road violation parking Full-automatic capturing system, belong to road traffic law violation behavior intelligent processing method technical field.

Background technology

With the development of social economy and the raising of living standards of the people, Urban vehicles poputation rapidly increases, quickly The supply of the parking demand and parking space of growth is incompatible, parking supply and demand contradiction；Simultaneously as motor vehicle is driven The traffic safety consciousness of people is thin, and motor vehicle illegal parking turns into city persistent ailment, and traffic is gathered around caused by illegal parking every year Situations such as stifled, rear-end collision, is millions of, drastically influence the overall traffic environment in city, and can cause people people The loss of body property.People have appreciated that to rely solely on road construction eternal while accelerating road infrastructure and building It can not meet growing traffic needs, it is necessary to which the higher means of content of depending on science and technology change existing communications and transportation system System, carries out efficient management talent and is fundamentally solving traffic problems.

Motor vehicle number, which is continuously increased, causes municipal highway traffic pressure to increase, energy resource consumption is huge, traffic accident increases Add, a series of problems, such as traffic congestion is serious, environmental pollution.It is to be overcome to have become a urgency for city parking offense as can be seen here Serious problems.But for the illegal activities, the road violation monitoring system that prior art uses does not provide effective solution Method, there is the blind area of electronics law enforcement.Traffic police is typically taken directly to go on patrol the measure of patch penalty note at present, this not only adds volume Outer administrative expenditure cost, and the comprehensive and promptness of law enforcement can not be ensured.Traffic based on video, which is disobeyed, stops capturing system Parking offense event on road can be monitored in real time, automatic capture information of vehicles, advise the traffic of driver's voluntary observance Then, traffic accident and the generation of road congestion are reduced, and alleviates the workload of road administration personnel, before there is extraordinary application Scape and market value.In order to solve Railway Project above, it is very necessary to develop the parking offense capturing system of intelligent and high-efficiency, With security, reduction energy resource consumption, the traffic capacity for improving highway network, the raising Automobile Transportation benefit for improving public transport With the advantage of economic benefit etc..

There are some widely used road traffic "smart" products, prior art in each metropolitan traffic system at present In have some commercial products launch arrived in market, improve urban transportation.Automatic snapshot system improves human eye monitor video Pattern, employ modern advanced computer technology vision technique, communication network technology realizes round-the-clock automatic management, have Effect reduces traffic administration cost, to reducing traffic accident, ensureing that urban transportation operation plays an important role.At present, state Inside and outside scholar and company are all increased to automatic snapshot system R＆D intensity, and achieve larger achievement.In the prior art Automatic system Successful utilization based on video analysis on major traffic main artery, can detect to make a dash across the red light, exceed the speed limit, press yellow line, not by The violations of rules and regulations such as lanes, retrograde.Such as the camera system of making a dash across the red light of Australian Image Application companies, its System uses video analysis, unmanned control design case, full-automatic aperture regulation, focus adjustment and rotation regulation, effective capture The red light running behavior of vehicle.U s company have developed TrafficIII-G types and make a dash across the red light automatic monitoring camera, and use is special High-resolution traffic cameras, captured from headstock and the tailstock.Some Chinese security protection companies also have developed the product of oneself, such as sea Kang Wei makes a dash across the red light capturing system depending on the all-in-one released, and the big magnificent bayonet socket detecting system etc. of making a dash across the red light in Zhejiang, these products are in reality Preferable effect is achieved in border.

But the application in this field is captured for parking offense, prior art is in traffic actual scene using less. Prior art is difficult there is some on parking offense candid photograph is realized, such as：

When in actual monitored scene, there is the interference of many extraneous factors change, such as the illumination variation of environment, bag Include illumination gradual change and illuminance abrupt variation etc.；Motion change, the background such as camera position shake and leaf, ripple caused by bad weather Cycle movement of object etc.；The change of background form, new object stops in the scene or background object leaves scene, this point Showed in the traffic scene of parking offense detection especially prominent.

Second, many difficult points also still be present in vehicle detection, include the influence of dynamic scene, field is monitored in actual traffic Jing Zhong, background image often and nonstatic, such as, illumination variation in scene, the flag trees shaken and due to video camera Shake caused by image shift etc., these dynamic backgrounds are easy to be mistaken as moving target, so as to cause on a large scale Complex jamming, seriously add the difficulty of vehicle detection；Block the influence of adhesion, blocking between object can cause object each other Adhesion, these all can cause difficulty to the characteristic matching in target following, when video image has multiple objects, can exist unavoidably To each other block adhesion；The influence of shade is moved, in traffic scene, shade is frequently accompanied by moving vehicle appearance.And move Motion shade can be mistakenly considered moving vehicle by algorithm of target detection, and this has not only had a strong impact on the precision of vehicle detection, and respectively Individual moving target can be projected shade and be sticky, and then vehicle tracking is interfered.

Third, carry out target following in Video Supervision Technique, Major Difficulties are the influence of extraneous factor, in background image Various interference be present, easily produce false target, moreover, because cast shadow and shading one another overlapping can be sent out between moving target The phenomenon of raw adhesion, very big difficulty is caused to target following；The requirement of real-time of tracking, because the video data volume is huge, if not Suitable algorithm policy is selected, is extremely difficult to requirement of real-time.

Fourth, realizing that the global design difficulty of parking offense automatic snapshot system is larger, architecture design, the summary of system are set Meter, software function are realized all complex.The system expandability, function opposite independent, stability requirement are high.Evidence obtaining requires vehicle Feature can resolution it is high, system requirements high-definition network ball machine, the number-plate number, ball machine control speed can be distinguished after being furthered to target vehicle Degree, response speed, capture operation fluency and all have high requirements.

The content of the invention

In view of the shortcomings of the prior art, a kind of precisely quick road violation parking provided by the invention is full-automatic captures system System, is made up of front end control point part, internet transmission of virtual laboratory and central manage-ment unit three parts.Control point part in front end includes height Clear network ball machine, video analysis unit two parts composition.Systemic-function is powerful, it is sufficient to meet each application scenario, it is main to include pair Ball machine progress remote control, the preview of video monitoring, illegal parking automatic detection, the number-plate number of front end control point are known automatically Not, high definition video recording and the association of illegal activities video, information of vehicles storage browse, the function such as system administration and remote maintenance.This is Unite with the obvious advantage, mainly there is liberation police strength, improve violation snap-shooting efficiency, collected evidence using double ball machines, meet national standard, million pictures The evidence obtaining of plain level high-definition image, vehicle characteristics can resolution it is high, easy for installation, can Gao Xinggao, installation, maintenance cost be low, special evidence obtaining Management platform, variation are compared and informed, the six big advantages such as hommization law enforcement.With reference to the control protocol of high-speed ball-forming machine, height is developed Fast ball machine zone location and the algorithm rotated, realize and interesting target in monitoring scene are accurately positioned.Positioned to improve Accuracy, devise multiple bearing method.First positioning obtains the positional information of vehicle and rotated and furthered control for ball machine, Ensure to obtain clearer vehicle data, ensure that car plate can be recognized accurately in Car license recognition module, improve rate of accurateness.

To reach above technique effect, the technical solution adopted in the present invention is as follows：

A kind of precisely full-automatic capturing system of quick road violation parking, including front end control point part, network transmission Part and central manage-ment unit part；

Control point part in front end includes high-definition network ball machine and video analysis unit, and high-definition network ball machine collection high definition regards Frequently, it will be handled in live video stream incoming video analytic unit, from motion tracking vehicle, carry out vehicle amplification and rotation, video The hardware platform of analytic unit is industrial computer, and video analysis unit is divided into Computer Vision algoritic module, high-speed ball camera Candid photograph and Car license recognition module, vehicle peccancy evidence processing module, software security maintenance module, Computer Vision algoritic module The video of collection is monitored in real time using image processing algorithm, judges whether vehicle stops to disobey and stops region, regarded in real time Polygon be configured in frequency scene disobey stop region, judge entering the vehicle detection in this region, tracking, identification and act of violating regulations, Vehicle checking method includes time-based motion history image algorithm and vehicle shadow minimizing technology, and wireless vehicle tracking uses Matching pursuitalgorithm is estimated in distinguished point based light stream；

Internet transmission of virtual laboratory includes interchanger, single-mode fiber, fiber optical transceiver；

Central manage-ment unit includes two parts of server end and management client, and server end includes Web application services Device, management server, data storage server, the request of Web Application Server processing client and processing business, management service Device manages the ball machine of each video analysis unit and each control point, and data server completes the storage of parking offense data and standby Part.

A kind of precisely full-automatic capturing system of quick road violation parking, further, the collection of high-definition network ball machine is high Clear video flowing, real-time video is streamed to video analysis unit, video analysis unit using the TCP/IPRTP network transmission protocols Receive and video is decoded into the data transfer of yuv format to Computer Vision using H.264 standard agreement after video code flow Algoritic module simultaneously controls the rotation of ball machine to generate parking offense data, and central manage-ment unit is sent to regard to parking offense data.

A kind of precisely full-automatic capturing system of quick road violation parking, further, time-based motion history Image algorithm carries out moving meshes and Activity recognition under certain application scenarios, and specific method is：In the video sequence Moving target, under the change of time, corresponding moving target is using maximum weights in present frame, and historical frames above Should constantly it reduce according to the length factorindex of time interval, final disappear is zero；

The history image that time-based motion history image moves in current time composition floating-point matrix description, under use Formula is updated：

Wherein mhi (x, y) is motion history image pixel value, and sihouette (x, y) is to be obtained just using frame differential method Value behind beginning moving region, sihouette (x, y) value are arranged to 0 in non-athletic area, and the pixel of moving region is arranged to 1, Duration is the period being manually set, and timestamp is current system time；

Unit object is split from the time-based motion history image sequence of real-time update first, to segmentation Target out is labelled, and specific requirement in the application is just reached after the completion of mark, the car using following methods to traveling Carry out image segmentation：

The first step carries out linearization process by following equation to each pixel of motion history image：

Mhi (x, y)=mhi (x, y) * 255/duration

Each pixel value is lifted after processing, when wherein the pixel in motion history image is current time stamp, The pixel value of mark is maximum；

Second step carries out the cavity that Morphological scale-space is removed in noise spot and blank map picture to obtained image；

3rd step carries out contour detecting to gained image, and independent region is represented with rectangle, is partitioned into motion mesh Vehicle is marked, corresponding threshold value is set to rectangle to remove interference of other targets to vehicle.

A kind of precisely full-automatic capturing system of quick road violation parking, further, vehicle shadow minimizing technology is adopted With the cast shadow suppressing method based on color notation conversion space, can not effectively split shade and moving target in RGB color Hsv color spatial manipulation is converted to, using the cast shadow suppressing method of color notation conversion space, when pixel is covered by shade, the picture The brightness V of vegetarian refreshments is dimmed, and saturation degree S can decline, and colourity H changes less, by the color for analyzing foreground point and background dot Color difference, shadow templates are defined as follows to each pixel of motion target area：

Wherein, Sh_k ^(C)(x, y) is image coordinate point (x, y) place direct-shadow image vegetarian refreshments mask, V_k ^F(x, y) S_k ^F(x, y), H_k ^F (x, y) and V_k-1 ^B(x, y), S_k-1 ^B(x, y), H_k-1 ^B(x, y) difference denotation coordination point (x, y) place's current pixel value and previous moment V, S, H component of background pixel value, if pixel is judged as shade, pixel mask Sh_k ^(C)(x, y) is 1, otherwise is 0, threshold alpha is factually set, if the cast shadow in background is stronger, α settings are smaller, and threshold value beta is used to ensure current pixel point Brightness can not be got too close to image background, T_SAnd T_HThe threshold value of saturation degree difference and colour difference respectively between shade and background, is obtained After shadow templates, replace moving target to be mistaken for the pixel value of shade with the pixel value of background image.

A kind of precisely full-automatic capturing system of quick road violation parking, further, distinguished point based light stream estimation The light stream of Matching pursuitalgorithm is the apparent motion of gradation of image pattern, and optical flow field is two-dimentional instantaneous velocity field, two dimension therein Velocity is projection of the three dimensional velocity vectors in imaging surface of visible point in scenery, and light stream carries three-dimensional vehicle structure letter Breath includes movable information；

The central point of vehicle characteristics point selection vehicle, the initial time for stopping region is disobeyed when vehicle target enters, detects car And information of vehicles is depicted with rectangle, obtain size, the positional information of vehicle, the central point of rectangle be set to vehicle center Point, in follow-up video sequence, light stream estimation is carried out to this central point and is tracked matching.When find this feature from Open to disobey and stop region, then discharge the resource of distribution, carry out subsequent treatment, it is accurate according to two aspect setting matchings during tracking Then：The distance criterion of central point movement, central point moving direction criterion.

A kind of precisely full-automatic capturing system of quick road violation parking, further, distinguished point based light stream estimation The idiographic flow of Matching pursuitalgorithm is as follows：

Step 1, detect whether that vehicle enters to disobey and stopped region, it is special if then using vehicle detecting algorithm to obtain vehicle Reference ceases；

Step 2, make light stream estimation, matched after trying to achieve central point according to matching criterior, such as match unsuccessful explanation step Vehicle could not be accurately detected in rapid one, goes to step five, if successful match, goes to step three；

Step 3, judges whether vehicle stops stopping in region separated, two is gone to step if vehicle does not stop, if car Stopping goes to step four, stops region if vehicle leaves to disobey and goes to step five；

Step 4, into the process part to parking offense vehicle；

Step 5, cancel vehicle characteristic information, free system resources.

It is a kind of that precisely quickly road violation is stopped full-automatic capturing system, further, high-speed ball camera capture and Car license recognition module captures the coordinate information and license board information of vehicles peccancy, when having detected that act of violating regulations occurs for vehicle, catches Current video image is caught, this image is the distant view image of vehicle, and rotation ball machine, which furthers, is accurately positioned target, captures a car Close shot picture, Car license recognition is carried out to this pictures after acquisition success, preserves the car plate identified, if Car license recognition loses Lose and identification is repeated, abandon after multiple recognition failures and this time capture, after Car license recognition, sending ball machine control command makes ball Machine returns to monitoring presetting bit, after being spaced setting time, then another Zhang Yuanjing vehicle pictures is captured, by three pictures of acquisition A pictures are combined as, and are superimposed picture evidence of the watermark information as parking offense.

A kind of precisely full-automatic capturing system of quick road violation parking, further, vehicle peccancy evidence processing mould Block by parking offense place, disobey between the stopping time, the number-plate number, disobey and stop behavior, picture path, information storage into database, and press Place is first class catalogue, and the time is that second-level directory preserves to picture in disk.

A kind of precisely full-automatic capturing system of quick road violation parking, further, software security maintenance module pair Running software is safeguarded and ensures the stability of software systems operation free of discontinuities, and system journal has comprising main contents：Stopping is grabbed At the beginning of clapping time, startup candid photograph time and serial dog board communications Time To Failure, software startup time, software shut-in time, capture card The beginningization time.

A kind of precisely full-automatic capturing system of quick road violation parking, further, is serviced in central management system Device end is the centralized management storage center of the whole network, manages the separated of the whole network concentratedly and stops capturing point, and stores the separated of the whole network and stop capturing card According to, management client is the interface of user and whole system, and multiple users access whole system simultaneously by different clients, in Heart management system includes four subsystems：

First, real-time video monitoring subsystem, user checks that the whole network disobeys the real-time video for stopping capturing point by management system；

Second, control subsystem of setting up defences, user carries out disobeying the setting for stopping capturing algorithm, deployed troops on garrison duty, operation of withdraw, is arranged on certain A little periods deploy troops on garrison duty, and other time section is then in the air；

Stop Query Subsystem third, disobeying, User Defined condition query, which is disobeyed, to stop alerting, and checks that certain is disobeyed the alarm for stopping alarm Picture and alarm process video recording；

Fourth, alarm export subsystem, user will disobey the form for stopping alarm and exporting as txt, html, excel form.

Compared with prior art, the advantage of the invention is that：

The full-automatic capturing system 1. a kind of precisely quick road violation provided by the invention is stopped, mainly for city road The public transport no-parking zone scene such as road, square, parking lot, it is the technology whole new set of applications such as vehicle detection, car plate detection.This System replaces artificial vision by computer vision, embodies the continuous operational advantages of machine and people's sector of breakdown cognitive ability.Can Detection record is accurately and rapidly carried out to the behavior of motor vehicle illegal parking, integration, which is delivered, is easy to construction and installation, efficiently solves Artificial detection human cost is high, fatiguability, easily the drawback such as carelessness, effectively suppresses the generation of parking offense act of violating regulations.

The full-automatic capturing system 2. a kind of precisely quick road violation provided by the invention is stopped, is examined for parking offense The concrete application of survey, employ the motion detection algorithm of time-based motion history image.Table of this algorithm to moving target Up to unrelated with the service behaviour of processing system, taking full advantage of video, entrained information is entered to the target in motion process in itself Row detection, quickly detects to the information of vehicles in video, possesses compared with hard real-time；When eliminating the shade of moving vehicle Employ the minimizing technology of shade characteristic.Information of vehicles can be accurately obtained by this method of test proof and stores vehicle Details.

The full-automatic capturing system 3. a kind of precisely quick road violation provided by the invention is stopped, has used characteristic point light Stream estimation Matching pursuitalgorithm.This algorithm carries out light stream estimation to the eight neighborhood of vehicle center point, is calculated and taken by mathematical formulae After averagely obtaining vehicle center point coordinates value, matched jamming is carried out using direction and distance criterion.This track algorithm is easy to Vehicle is judged by moving to this static process in ground, and anti-interference is stronger, meets specific application requirement.

The full-automatic capturing system 4. a kind of precisely quick road violation provided by the invention is stopped, realizes parking offense Automatic snapshot system.System that employs multithreading, modular design method, to the architecture design of system, Outline Design, soft Part function, which is realized, has all carried out detailed discussion and design.System is entered under the different weather situation such as fair weather or bad weather Test is gone, test result shows that system captures accuracy rate more than 90%, and the system expandability is strong, function opposite independent, steady It is qualitative strong.

The full-automatic capturing system 5. a kind of precisely quick road violation provided by the invention is stopped, effectively liberates police strength, Improve violation snap-shooting efficiency.System can realize the automatic candid photograph to illegal parking behavior, it is not necessary to which inspection is held on people's police to road Diligent, the unlawful data that institute a little position is shot only needs a people to carry out verification examination, has liberated police strength significantly.Secondly, presently, there are Some drivers handle thing for convenience, and just the place around processing thing carries out parking offense, when people's police are on the scene When, driver just at once out sails out of vehicle, and vehicle is parked in into roadside again after people's police walk, so " all with people's police on duty Rotation ", management difficulty is larger, and high definition violation snap-shooting system then can effectively solve problems, plays effectively fright effect.

The full-automatic capturing system 6. a kind of precisely quick road violation provided by the invention is stopped, using mega pixel level High definition is collected evidence, vehicle characteristics can resolution it is high, system uses high-definition network ball machine, more Zoom Lens, can monitoring range it is big, to mesh Mark vehicle can distinguish the number-plate number after furthering, ball machine control speed is fast, fast response time, and it is smooth to capture operation.System installation side Just, reliability is high, and the low system of installation maintenance cost is designed using integrated ball machine, and degree of protection is high, is different from monopod video camera Pattern, it is simple in construction, it is convenient for installation and maintenance.

Brief description of the drawings

Fig. 1 is the structural representation of the full-automatic capturing system of accurate quickly road violation parking of the present invention.

Fig. 2 is the Computer Vision algoritic module structural representation of the present invention.

The high-speed ball camera that Fig. 3 is the present invention is captured and Car license recognition modular structure schematic diagram.

Embodiment

Below in conjunction with the accompanying drawings, stop full-automatic capturing system to a kind of precisely quick road violation provided by the invention Technical scheme is further described, and those skilled in the art is better understood from the present invention and can be practiced.

Referring to Fig. 1 to Fig. 3, a kind of precisely full-automatic capturing system of quick road violation parking provided by the invention, bag Include front end control point part, internet transmission of virtual laboratory and central manage-ment unit part；

Control point part in front end includes high-definition network ball machine and video analysis unit, and high-definition network ball machine collection high definition regards Frequently, it will be handled in live video stream incoming video analytic unit, from motion tracking vehicle, carry out vehicle amplification and rotation, video The hardware platform of analytic unit is industrial computer, and video analysis unit is divided into Computer Vision algoritic module, high-speed ball camera Candid photograph and Car license recognition module, vehicle peccancy evidence processing module, software security maintenance module, Computer Vision algoritic module The video of collection is monitored in real time using image processing algorithm, judges whether vehicle stops to disobey and stops region, regarded in real time Be configured in frequency scene polygon disobey stop region, to enter this region vehicle detection and vehicle tracking, vehicle checking method bag Time-based motion history image algorithm and vehicle shadow minimizing technology are included, wireless vehicle tracking uses distinguished point based light stream Estimate Matching pursuitalgorithm；

Internet transmission of virtual laboratory includes interchanger, single-mode fiber, fiber optical transceiver；Mainly it is responsible for high definition video steaming the part Surveillance center is transferred to by fiber link, ensures transmission bandwidth and reduces network delay.

Central manage-ment unit includes two parts of server end and management client, and server end includes Web application services Device, management server, data storage server, the request of Web Application Server processing client and processing business, management service Device manages the ball machine of each video analysis unit and each control point, and data server completes the storage of parking offense data and standby Part.

High-definition network ball machine gathers high definition video steaming, using the TCP/IPRTP network transmission protocols by real-time video streaming To video analysis unit, video analysis unit receives and video is decoded into YUV lattice using H.264 standard agreement after video code flow The data transfer of formula is to Computer Vision algoritic module and controls the rotation of ball machine to generate parking offense data, with regard to parking offense Data are sent to central manage-ment unit.

The design of system overall soft is broadly divided into Computer Vision algoritic module, high-speed ball camera is captured and car plate is known Four other module, vehicle peccancy evidence processing module and software security maintenance module modules.System global structure is as shown in Figure 1.

(1) Computer Vision algoritic module includes carrying out detecting and tracking and parking offense to disobeying the target stopped in region Activity recognition.In order to reduce operand, arithmetic speed is improved, the separated monitored space that stops for only cutting " interested " is handled.If Disobey that to stop monitor area be trapezoid area, input figure of the trapezoidal boundary rectangle of present invention cutting as parking offense detection module Picture.In addition, the analyze speed of image is limited by factors such as the real-time of algorithm, industrial computer CPU processing speeds, if at image The speed of reason is relatively low, and system will take the strategy of " frame-skipping ", that is, skips some frames and do not process.

(2) high-speed ball camera capture and Car license recognition module be responsible for when identify vehicle it is separated stop behavior after, according to rear The candid photograph flow of face part, by each presetting bit and the location matches of vehicles peccancy, control clipping the ball captures suitable preset point. Video image acquisition module, system have been completed the mapping relations of two-dimentional traffic image and high-speed ball camera presetting bit.Bag Include system parameter setting and video image obtains；The setting that system parameter setting includes system monitoring region (is usually set to one Individual rectangle or trapezoid area), the operation such as the demarcation of high-speed ball camera presetting bit, video image, which obtains, to be referred to read by capture card Take and show the video image disobeyed and stop monitor area.

(3) vehicle peccancy evidence processing module is that the close shot image that clipping the ball is captured is inputed into number plate identification software, Obtain license plate number to record, after Car license recognition, video camera returns to monitoring presetting bit, in the process, can obtain by near To four remote evidence pictures, four pictures are combined together as to the evidence of parking offense.Vehicle evidence includes carrying above The picture evidence and license plate number evidence arrived.Evidence is according to time violating the regulations, place, license plate number subhead address book stored.Meanwhile card violating the regulations According to also traffic control department host computer database will be given by network transmission, to facilitate inquiry.

(4) software security maintenance module records the daily record of work, device log, network state daily record of main runs software, Record main software is restarted or the event log such as the change of failure and equipment or network state.System carries out active school When, the error of equipment is no more than 1.0 seconds in 24 hours, calibrates the primary system time per hour.Equipment running status have alarm to carry Show, short message informing function, device history running status can inquire about, export form.System possesses role-security management function, right Different objects distribute different types of access right and can carry out security audit.System carries out automatic detection when breaking down, By being pre-designed to obtain automatic checkout system failure and fast quick-recovery normal function, possesses higher availability.Mainly run soft The working condition automatic error detection of part reports an error, powers off and restarts automatically and monitor capital equipment (technical grade video camera, server automatically Deng) etc. function.System uses network communications technology, enables administrative staff centrally through the remotely located headend equipment of management software Parameter, support remote upgrade and remote device reboots function.System need to provide the log recording of key operation, convenient inquiry and pipe Reason.System supports the remote maintenance functions such as management and the parameter setting to headend equipment, function when possessing system school.

First, front end control point part

1st, high-definition network ball machine

High-definition network ball machine gathers HD video, will be handled in live video stream incoming video analytic unit.Automatically Track vehicle, carry out vehicle amplification and rotation.

High-speed ball camera is the optimal video camera front end of current safety-security area general performance, and it is integrated with shooting simultaneously Machine, head and communication protocol interface.Video camera is commonly referred to as " movement ", and head refers to the rotating part that stepper motor drives, Communication system refers to the control to motor and the process part to vision signal.

The operation principle of high-speed ball-forming machine is specific as follows：(1) CPU first sends instruction and realizes that ball machine fast and accurately positions Rotation.(2) and then the image of video camera, the function of video camera are written into CPU.(3) when head is controlled, image is transmitted Out, the functions such as white balance, zoom, focusing are also achieved while.

High-speed ball camera is typically mounted above road, and visual field direction is roughly parallel to the direction of highway extension.

After separated parking is identified, is rotated using PELCO-D protocol integrated test systems high-speed ball-forming machine and lens focusing, candid photograph are specific Information of vehicles, as evidence violating the regulations.In current high-speed ball-forming machine, all comprising preset bit function.The present invention is in advance in monitor area In divide some presetting bit regions, coordinate is saved in configuration file.Each region correspond to ball machine one is specific preset Position.When capturing every time, three presetting bits nearest away from a separated parking target rectangle frame center are found out, control head is grabbed successively Target is clapped, obtaining as far as possible can be by close shot image that Car license recognition software effectively identifies.If number plate is identified, by the car Target is added in list；If continuous capture all fails effectively to identify car plate three times, then it is assumed that captures unsuccessful.

2nd, video analysis unit

Video analysis unit is the core of system, and hardware platform is industrial computer, and major function is to obtain video flowing to call first The parking offense video analysis algorithm entered, automatic detection disobey parking, and control ball machine captures evidence obtaining.Video analysis unit, which is divided into, to be regarded Frequency image processing algorithm module, high-speed ball camera is captured and Car license recognition module, vehicle peccancy evidence processing module, software peace Full maintenance module.

(1) Computer Vision algoritic module

Computer Vision unit is the key of system, and the video of collection is supervised in real time using image processing algorithm Control, judge whether vehicle stops to disobey and stop region.As shown in Fig. 2 the first step is to stop region to setting to disobey in real-time video scene, It is typically configured to polygonal region；After being provided with, the vehicle for entering this region is detected, is tracked, identified and broken rules and regulations Behavior judges.

(2) high-speed ball camera candid photograph and Car license recognition module

Parking is disobeyed as shown in figure 3, capturing and mainly completing to handle with Car license recognition module, captures the coordinate letter of vehicles peccancy Breath and license board information.When having detected that act of violating regulations occurs for vehicle, current video image is caught, this image is vehicle Distant view image；Rotation ball machine, which furthers, is accurately positioned target, the close shot picture of one vehicle of capture, to this pictures after acquisition success Car license recognition is carried out, the car plate identified is preserved, if identification is unsuccessfully repeated in Car license recognition, is abandoned after multiple recognition failures This time alert, after Car license recognition, sending ball control command makes ball machine return to monitoring presetting bit, then captures another Zhang Yuanjing Vehicle pictures.Three pictures of acquisition are combined as a pictures, and are superimposed figure of some watermark informations as parking offense Piece evidence, and record vehicle into region and leave two sections of videos in region, picture and video this two evidences can abundant boundaries Determine act of violating regulations.

(3) vehicle peccancy evidence processing module

The evidence of vehicles peccancy includes：Picture evidence and video evidence.The storage principle of evidence is convenient inquiry and statistics Analysis is, it is necessary to by parking offense place, disobey between the stopping time, the number-plate number, disobey and stop the information such as behavior, picture path, video path and deposit Store up in database.And by place be first class catalogue, the time is that second-level directory preserves to picture and video in disk.Ensure Real-time alarm data and can be checked in client, and after the incident, the separated of enquiring vehicle stops behavior at any time, and can ensure respectively to make With a certain place different time sections violating the regulations of personnel's statistics and analysis, the situation violating the regulations of different directions.

3rd, ball machine deployment scenario

The monitored area length of each control point will be depending on the performance of ball machine, and general control is within 150 meters, simultaneously The vertical rod of ball machine sets the specific orientation that consider capturing region.

2nd, TCP data segment

Mainly there is the equipment such as Digital Optical Terminal, single-mode fiber, network switch composition.Fiber plant or route, which will use, rents With the mode of operator's commercial operation passage.Digital video stream data is collected to command centre and accesses single Ethernet data friendship Change planes, avoid causing being significantly increased for whole central site network system load.

3rd, central management system

Central management system, including two parts of server and management client：Server is that the centralized management of the whole network is deposited Storage center, manage the separated of the whole network concentratedly and stop capturing point, and store the separated of the whole network and stop capturing evidence, so that statistical analysis is used.Pipe Reason client is the interface of user and whole system, and multiple users can access whole system simultaneously by different clients.In The function of heart management system includes：

(1) real-time video monitoring：User can check that the whole network disobeys the real-time video for stopping capturing point by management system.

(2) set up defences control：User can carry out disobeying the setting for stopping capturing algorithm, deploy troops on garrison duty, operation of withdraw, it might even be possible to set Deployed troops on garrison duty in some periods, and other time section is then in the air, meets the diversified demand of user.

(3) disobey and stop inquiring about：User can inquire about to disobey with custom condition to stop alerting, and checks that certain is disobeyed the alarm picture for stopping alarm With alarm process video recording.

(4) alarm export：User can will disobey the form for stopping alarm and exporting as the different-formats such as txt, html, excel.

4th, vehicle checking method

Vehicle target detection in motion is the important research content of the subjects such as computer graphics, Image Information Processing, Its purpose is to extract the vehicle of motion from background image from sequence of images.Vehicle detection is to video scene Analysis and the key understood, its essence is a kind of concrete application of image segmentation.Moving target is divided exactly from video flowing Cut out most important for post-processings such as target classification, tracking and identifications.Based on video analysis parking offense automatic system Concrete application scene analysis, used method for testing motion and the vehicle shadow removal side of time-based motion history image Method.

1st, the method for time-based motion history image

This application-specific is detected for parking offense, it is necessary to which the detection to the driving vehicle in monitor video follows necessarily Real-time, accuracy and robustness etc. require.Enter the no-parking zone of configuration in vehicle, just obtain to disobey immediately and stop letter Breath, it is desirable to which loss is low and small by noise and shadow effect.Present invention employs a kind of time-based motion history image to calculate Method.

Moving target in the video sequence, under the change of time, corresponding moving target uses maximum in present frame Weights, it is maximum to the effect of expressing mass motion process；And historical frames above should act on power according to the length of time interval Value constantly reduces, and final disappear is zero.Based on this method, time-based motion history image algorithm current time composition is floating The history image of dot matrix description motion, and be updated using following formula.

Wherein mhi (x, y) is motion history image pixel value, and sihouette (x, y) is to be obtained just using frame differential method Value behind beginning moving region, sihouette (x, y) value are arranged to 0 in non-athletic area, and the pixel of moving region is arranged to 1, Duration is the period being manually set, and timestamp is current system time.This definition mode to be based on the time Expression of the motion history image to moving target it is unrelated with the service behaviour of processing system.So IMAQ frame per second is based on not With processor performance system under same period, same parameters obtained by time-based motion history image history when Between scope be identical.This method make use of the information that video possesses in itself to be detected to moving vehicle, can be quick Information in video is reacted, real-time is very strong, meets that video analysis parking offense detects the specific requirement of this application.

The detecting and tracking system of traveling target is realized, first goes through unit object from the time-based motion of real-time update Split in history image sequence, and then target respectively out is labelled, just reached in the application after the completion of mark Specific requirement.The present invention carries out image segmentation using the method for detection profile to the vehicle of traveling.

The first step carries out linearization process by following mathematical formulae to each pixel of motion history image：

Mhi (x, y)=mhi (x, y) * 255/duration

Each pixel value is lifted after processing, when wherein the pixel in motion history image is current time stamp, The pixel value of mark is maximum.Noise spot is removed to obtained image progress Morphological scale-space in next step and blank map is as in Cavity.Final step carries out contour detecting to gained image, independent region is represented with rectangle, you can be partitioned into motion Target vehicle.Also need to set rectangle corresponding threshold value in specific application with interference of other targets of place to go to vehicle, Improve the accuracy of target detection.

2nd, vehicle shadow minimizing technology

Because the object moved in motion process is connected with shade, the change of body form profile, object space can be caused Correct estimation can be affected.Shade can connect into different objects one shape, and the correctness of detection is by serious shadow Ring.Because in moving body track and identifying system, the shade of motion can have a strong impact on the robustness of system, the detection of shade Just turn into moving body track and the key component in identification with removing.In intelligent transportation image identification system, it is necessary to carry out Shadow Detection, influence of the shade to real goal is removed, accurately to detect vehicle target.The present embodiment combination shade produces Principle design how shade to be suppressed based on color notation conversion space.

(1) generation of shade and feature

In actual scene, light can produce shade after being blocked by obstructions.Shade can be divided into dynamic shadow and static state is cloudy Shadow.Dynamic shadow is shade caused by vehicle target itself, so be divided into from shade and cast shadow again, the former is target itself Caused by the one side of backlight, the latter is projected by object under the conditions of light irradiation and formed in background, so it is divided into half again Shadow and shadow.In regional extent corresponding to shadow, its light thing that is blocked completely blocks, and penumbra is then the thing portion that is blocked That divides blocks.Shadow during object of which movement, is easily detected because being clearly distinguished from background, but would generally be by mistake Ground is detected as moving target.Static shade refers to the shade of building, trees etc., is generally taken as background.

(2) cast shadow suppressing based on color notation conversion space

When RGB color can not effectively split shade with moving target, other color spaces are typically converted into Handled.Using the cast shadow suppressing method of color notation conversion space, typically the rgb pixel value of the pixel extracted is transformed into Other color space models, the suppression of shade is carried out in the chromatic characteristic of new color space using shadows pixels value.For side Just detect shade, the colouring information of pixel is generally changed into HSV space from RGB color.HSV space reflects The vision system of people can effectively identify shade, H represents colourity (Hue), S in hsv color space to the perceptive mode of colour Saturation degree (Saturation) is represented, V represents brightness (Value).HSV space more meets sense of the human visual system to color Feel, can relatively accurately identify shade.Rgb space can be transformed into HSV space by below equation：

Learnt by experiment test, when the pixel of target is covered by shade, pixel colourity change is slow, color Brightness can be dimmed, and saturation degree would generally diminish.According to this conclusion, using such as making policy decision formula, judge that pixel value whether may For shade.

Wherein, Sh_k ^(C)(x, y) is image coordinate point (x, y) place direct-shadow image vegetarian refreshments mask, V_k ^F(x, y), S_k ^F(x, y), H_k ^F (x, y) and V_k-1 ^B(x, y), S_k-1 ^B(x, y), H_k-1 ^B(x, y) difference denotation coordination point (x, y) place's current pixel value and previous moment V, S, H component of background pixel value, if pixel is judged as shade, pixel mask Sh_k ^(C)(x, y) is 1, otherwise is 0, threshold alpha is factually set, if the cast shadow in background is stronger, α settings are smaller, and threshold value beta is used to ensure current pixel point Brightness can not be got too close to image background, TS and TH are respectively the threshold value of saturation degree difference and colour difference between shade and background, After obtaining shadow templates, moving target is replaced to be mistaken for the pixel value of shade with the pixel value of background image.

5th, wireless vehicle tracking

Tracking is the important evidence of traffic administration, and for video monitoring system, the detection of moving vehicle is first Step, in order to obtain vehicle continuous information in video scene to identify the behavior of vehicle, it is necessary to realize the real-time tracking of vehicle.This Embodiment elaborates the difficult point and solution method of the general principle and tracking problem matched during tracking, have studied and is based on regarding The tracking problem of the method for feel vehicle in video analysis parking offense system, with reference to parking offense systematic difference feature, if The method for having counted distinguished point based light stream estimation completes vehicle location and tracking.

1st, tracking general introduction and matching principle

For the parking offense traffic system of view-based access control model, although the result of moving vehicle detection and profile description carries Moving vehicle shape and the essential information of position in road are supplied, but these information are isolated, are not provided with image sequence Between contact, so these isolated information not can completely expression vehicle behavior.So orphan that moving vehicle is detected During vertical Information application can make the method that contacts in time to those, processing further to result is obtained to moving target The continuous description of one of information of vehicles.Moving vehicle is detected in vehicle tracking requirement video resulting from clipping the ball Coordinate position, and the detailed movement locus coordinate for stating out vehicle.

The core concept of most of track algorithm is to carry out characteristic matching to foreground target in continuous image.Target With one or several key features for referring to be directed to traveling target in video image, in present frame with being built in former frame or a few frames Found the matching relationship of these features.The premise of matching is the correct detection of vehicle target feature interested.Spy for matching Sign includes the feature in close relations with moving object such as angle, position, speed, size etc., and color and shape of target etc.. According to the real-time scene of concrete application during matching, various vehicle characteristics are set with different weight coefficients, so as to determine tracking Emphasis is simultaneously improved to tracking effect.

Because the target of motion is changed according to the change of time, clarification of objective description and dynamic change 's.For from time angle, foreground target feature of the moving target feature always than being matched with it falls a later frame.So as to Know, the feature of foreground target can be used for the estimation to target signature in present frame in present frame, and here it is the improvement based on matching The core concept of algorithm.Specifically method is：Its spy in the next frame is first predicted according to the motion history of moving target record Sign, the feature of moving target in present frame is obtained, both are compared and correct prediction result into " a matching-amendment-pre- The process of survey ".

2nd, distinguished point based light stream estimation matched jamming

Light stream is a kind of apparent motion of gradation of image pattern.Optical flow field is a kind of two-dimentional (2D) instantaneous velocity field, wherein 2D velocities be visible point in scenery projection of three-dimensional (3D) velocity in imaging surface.Light stream carries relevant scape The abundant information of thing three-dimensional structure contains the movable information of observed object.

The vehicle characteristics point that the present embodiment uses is the central point of vehicle.When vehicle target enter disobey stop region it is initial when Carve, detect vehicle and information of vehicles is depicted with rectangle.The information such as size, the position of vehicle are obtained, by the central point of rectangle It is set to vehicle center point.In follow-up video sequence, light stream estimation is carried out to this central point and is tracked matching.Work as hair This existing feature, which is left to disobey, stops region, then discharges the resource of distribution, carry out subsequent treatment.

To the point estimation light stream of single car center, detectability and antijamming capability to central point require high, central Heart point detection failure or drift may result in target following failure, cause to omit important information.To improve the robust of track algorithm Property, prevent system interference noise from being influenceed on characteristic point.During tracking, it is contemplated that the fortune of same vehicle in two continuous frames image Dynamic direction and move distance do not have obvious change, so can be according to the two aspect setting matching criterior：Central point movement Distance criterion, central point moving direction criterion.

The purpose for carrying out vehicle tracking is to judge that vehicle behavior provides foundation.Using the center point coordinate of vehicle come to vehicle The judgement of behavior.Feature-point optical flow matched jamming idiographic flow is as follows：

Step 1, detect whether that vehicle enters to disobey and stopped region, it is special if then using vehicle detecting algorithm to obtain vehicle Reference ceases；

Step 2, make light stream estimation, matched after trying to achieve central point according to matching criterior, such as match unsuccessful explanation step Vehicle could not be accurately detected in rapid one, goes to step five, if successful match, goes to step three；

Step 3, judges whether vehicle stops stopping in region separated, two is gone to step if vehicle does not stop, if car Stopping goes to step four, stops region if vehicle leaves to disobey and goes to step five；

Step 4, into the process part to parking offense vehicle；

Step 5, cancel vehicle characteristic information, free system resources.

Feature-point optical flow matching tracking method has following four advantage：

(1) tracking of vehicle center characteristic point is directed to, effective method is provided for vehicle Activity recognition.

(2) track algorithm that the present invention applies, has certain robustness, partial occlusion pair to the problems such as occlusion adhesion The influence for tracking accuracy is smaller；

(3) matching principle according to track algorithm of the present invention is understood, the process of tracking is also the process of an amendment, can be with The result of vehicle detection is modified, it is ensured that track the correctness of target；

(4) enter the separated initial time for stopping region in vehicle and detection acquisition characteristic point is carried out to vehicle, vehicle is in Wei Ting areas Only are detected by the computing of vehicle, has ensured accuracy and the raising of tracking for feature point detection, the every frame of reduction during moving in domain The real-time of system.

6th, systemic-function

The system is powerful, it is sufficient to meet each application scenario, it is main to include carrying out remotely the ball machine of front end control point Control, the preview of video monitoring, illegal parking automatic detection, rock mass displacements, high definition video recording and illegal activities video The functions such as association, information of vehicles storage browse, system administration and remote maintenance.

1st, remote control function

System supports staff to carry out angle and far and near contracting to the ball machine of front end monitoring site by operating platform software Adjustment is put, chooses suitable candid photograph picture.

2nd, video monitoring preview and vehicle pictures capture function

System can be by the comprehensive real-time video for browsing front end monitored area of operating platform software, for monitor area Interior illegal parking, violate and forbid the behaviors such as graticule to realize candid photograph function, system can realize that automatic identification is captured and hand Dynamic candid photograph is combined function, and the image of collection meets ISO/IEC 15444:2000 require, have anti-tamper function.

System can realize that writing function is captured in a variety of illegal activities, and system captures 4 pictures (two headstock pictures, one Open tailstock picture, a vehicle license plate characteristic picture) synthesized, section situation, at least a photo where vehicle can be reflected Brand number can clearly be distinguished.Support that the additional information (such as number-plate number, time, place) of picture is audited and compiled Volume.Vehicle pass-through information writes linked database, and relevant information (time, place, direction, speed, brand number etc.) is folded It is added on picture.The original image information of system record all has anti-tamper function, prevents in processes such as transmission, storage, processing It is middle artificially to be changed.

3rd, illegal parking automatic detection and rock mass displacements function

System possesses the automatic detection that prohibitory sign instruction is violated to motor vehicle and captures function, no parking place, After region parking exceedes setting threshold values, system can record act of violating regulations automatically and provide 4 illegal process pictures.System is for machine The traffic safety illegal activities occurred in motor-car quiescing process, the interval time of two width motor vehicle feature images of shooting are more than 10s.For violating traffic safety illegal activities as defined in temporary parking, it shoots the interval of two width motor vehicle feature images The temporary parking time as defined in time greater than flag or graticule.System supports the picture automatic identification to candid photograph, recognition result branch Desk checking verification is held, is possessed to civilian vehicle number plate, police vehicle number plate, military vehicle number plate, 07 formula People's Armed Police's number plate of vehicle Automatic identification function.

4th, high definition video recording and illegal activities video correlation function

System provide set video recording evidence obtaining function, capture fragment video recording can in the way of panorama, middle scape, close shot, The whole video recording of illegal activities can be carried out in the way of close shot, middle scape, panorama.Size of recording a video is controllable, at resolution ratio Reason is set.Each crossing video recording holding time must not be less than 30.Each traffic violation data has corresponding dynamic to record a video Association, intercept illegal activities before and after about 10 seconds video recording (video time can reflect illegal process) stored, in case later disobey Method dispute is verified.

5th, information of vehicles storage function of browse

The traveling such as time that the vehicle characteristic informations such as number plate of vehicle feature, the number-plate number and vehicle pass through, place Time information memory is no less than 2 years；Monitoring pictorial information storage (uses JPEG coded formats, meets ISO/ no less than August IEC1544:2000 require).Unlawful data form will meet public security relevant industries standard.System should be able to vehicle pass-through information, The information such as vehicle illegal information system running log information, System Operation Log carry out query statistic by different condition.

6th, system administration and remote maintenance functions

System should have the daily record of work for recording main runs software, device log, network state daily record, moreover it is possible to which record is main Software is restarted or the event log such as the change of failure and equipment or network state.When system can carry out active school, 24 The error of equipment is no more than 1.0 seconds in hour, calibrates the primary system time per hour.Equipment running status should have alarm, Short message informing function, device history running status can inquire about, export form.System should possess role-security management function, energy It is enough that different objects are distributed with different types of access right and security audit can be carried out.System can be carried out automatically when breaking down Detection, by being pre-designed to obtain automatic checkout system failure and fast quick-recovery normal function, possesses higher availability.Should have The working condition automatic error detection of main runs software reports an error, power off and restarts automatically and monitor capital equipment automatically (technical grade is taken the photograph Camera, server etc.) etc. function.

System uses network communications technology, administrative staff is joined centrally through the remotely located headend equipment of management software Number, support remote upgrade and remote device reboots function.System need to provide the log recording of key operation, convenient inquiry and pipe Reason.System supports the remote maintenance functions such as management and the parameter setting to headend equipment, function when possessing system school.

7th, system performance index

The system has stronger specific aim, is mainly used in major city vehicle no-parking zone.This system is one Individual to disobey a concrete application for stopping region, system need to realize whole day 24 hours, full-automatic, vehicles peccancy free of discontinuities is captured and obtained Reliably evidence violating the regulations.The application is proposed higher requirement to the security of system, real-time, stability and reliability.

Specific performance indications requirement has at following 8 points：

(1) the stronger system expandability.The extension that system can be applied according to being actually needed.As safety belt detects Deng.

(2) system captures accuracy rate.Night candid photograph rate is more than 92%, is grabbed under the severe weather conditions such as snowy day, rainy day, cloudy Beat rate is more than 95%, and candid photograph rate when daytime, weather condition was preferable is more than 98%.

(3) accuracy rate is captured.Night candid photograph rate is more than 90%, candid photograph rate under the severe weather conditions such as cloudy, rainy day, snowy day Not less than 95%, candid photograph rate when daytime, weather condition was preferable is more than 98%.

(4) Car license recognition accuracy rate.It is more than 97% by Car license recognition software Car license recognition accuracy rate.

(5) stability of a system.24 hours monitoring system fail softs restart number no more than 2 times.

(6) detect and track accuracy rate.Night Detection accuracy is more than 90%, the bad weather shape such as cloudy, rainy day, snowy day Detection accuracy is not more than 95% under condition, and Detection accuracy when daytime, weather condition was preferable is more than 98%.

(7) real-time.The time handled using image processing algorithm a two field picture is not more than 40ms/f.

8th, system performance testing

1st, field experiment

System is tested using Shanghai Communications University Yifu Science and Technology Building doorway road as monitoring scene.Disobeyed when vehicle enters Stop in region, detect the unlawful practice of vehicle.In order to effectively collect the detection data of separated stop system, carry out in all weather Effective experiment.Experiment is carried out under light rain, cloudy, three kinds of weathers of fine day respectively.Arriving 18 points of evening with early 8 points, totally 10 hours are made For experimental period.Table 1 describes experiment:

The experimental result of table 1 describes table

2nd, performance evaluation

According to the application requirement of parking offense automatic snapshot system, the performance indications that system need to meet include：System is accurate Rate, system capture accuracy rate, Car license recognition accuracy rate, detection and tracking accuracy rate, the system expandability, the stability of a system, reality Shi Xing.It can show that system often handles time needed for a frame no more than 40ms from test result, fully meet the reality of system The requirement of when property.It can be drawn from serial plate data, when interference is compared under serious situation, system understands adjust automatically state, ensures Stability.In addition, detection and tracking accuracy rate (n1), Car license recognition accuracy rate (n2), candid photograph accuracy rate (n3), system are accurate The calculation formula of rate (n) is as follows, and other specification refers to table 1 in formula：

Table 2 gives the Performance Analysis for experimental data.Weather conditions considerable influence is easily drawn from table The performance of system, main cause are that atrocious weather makes have substantial amounts of interference letter in the realtime graphic of high-speed ball camera capture Breath, largely have impact on the detection and tracking of vehicle.The algorithm that the present invention refers to inhibits interference information to a certain extent, grabs Beat rate has all reached more than 90% under various conditions, and system meets the performance indications requirement of design.

The system performance index analytical table of table 2

The basic principles, principal features and advantages of the present invention have been shown and described above.While there has been shown and described that Embodiments of the invention, for the ordinary skill in the art, it is possible to understand that do not departing from the principle and essence of the present invention A variety of change, modification, replacement and modification can be carried out to these embodiments, the scope of the present invention is by appended right in the case of god It is required that and its equivalent restriction.

Claims (10)

1. a kind of, precisely quickly road violation is stopped full-automatic capturing system, it is characterised in that including front end control point part, Internet transmission of virtual laboratory and central manage-ment unit part；

Control point part in front end includes high-definition network ball machine and video analysis unit, and high-definition network ball machine collection HD video will Handled in live video stream incoming video analytic unit, the hardware platform of video analysis unit is industrial computer, video analysis unit Be divided into Computer Vision algoritic module, high-speed ball camera capture and Car license recognition module, vehicle peccancy evidence processing module, Software security maintenance module, Computer Vision algoritic module are monitored in real time to the video of collection, judge whether vehicle stops Only stop region separated, vehicle detection and vehicle tracking to entering this region, vehicle checking method include time-based motion History image algorithm and vehicle shadow minimizing technology, wireless vehicle tracking are calculated using distinguished point based light stream estimation matched jamming Method；

Internet transmission of virtual laboratory includes interchanger, single-mode fiber, fiber optical transceiver；

Central manage-ment unit includes two parts of server end and management client, and server end includes Web Application Server, pipe Manage server, data storage server.

The full-automatic capturing system 2. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that High-definition network ball machine gathers high definition video steaming, and real-time video is streamed into video point using the TCP/IPRTP network transmission protocols Unit is analysed, video analysis unit receives the data that video is decoded into after video code flow yuv format using H.264 standard agreement It is transferred to Computer Vision algoritic module and controls the rotation of ball machine to generate parking offense data, parking offense data is sent To central manage-ment unit.

The full-automatic capturing system 3. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that Time-based motion history image algorithm carries out moving meshes and Activity recognition under certain application scenarios, specific side Method is：Moving target in the video sequence, under the change of time, corresponding moving target uses maximum power in present frame Value, and historical frames above should constantly reduce according to the length factorindex of time interval, final disappear is zero；

The history image that time-based motion history image moves in current time composition floating-point matrix description, is entered using following formula Row renewal：

<mrow> <mi>m</mi> <mi>h</mi> <mi>i</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>t</mi> <mi>i</mi> <mi>m</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> <mi>a</mi> <mi>m</mi> <mi>p</mi> </mrow> </mtd> <mtd> <mrow> <mi>s</mi> <mi>i</mi> <mi>h</mi> <mi>o</mi> <mi>u</mi> <mi>e</mi> <mi>t</mi> <mi>t</mi> <mi>e</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>!</mo> <mo>=</mo> <mn>0</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>s</mi> <mi>i</mi> <mi>h</mi> <mi>o</mi> <mi>u</mi> <mi>e</mi> <mi>t</mi> <mi>t</mi> <mi>e</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>=</mo> <mn>0</mn> <mi>a</mi> <mi>n</mi> <mi>d</mi> <mi> </mi> <mi>m</mi> <mi>h</mi> <mi>i</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&lt;</mo> <mi>t</mi> <mi>i</mi> <mi>m</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> <mi>a</mi> <mi>m</mi> <mi>p</mi> <mo>-</mo> <mi>d</mi> <mi>u</mi> <mi>r</mi> <mi>a</mi> <mi>t</mi> <mi>i</mi> <mi>o</mi> <mi>n</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>m</mi> <mi>h</mi> <mi>i</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>

Wherein mhi (x, y) is motion history image pixel value, and sihouette (x, y) is to obtain initial fortune using frame differential method Value behind dynamic region, sihouette (x, y) value are arranged to 0 in non-athletic area, and the pixel of moving region is arranged to 1, Duration is the period being manually set, and timestamp is current system time；

Unit object is split from the time-based motion history image sequence of real-time update first, to splitting Target label, specific requirement in the application is just reached after the completion of mark, the vehicle of traveling is entered using following methods Row image is split：

The first step carries out linearization process by following equation to each pixel of motion history image：

Mhi (x, y)=mhi (x, y) * 255/duration

Each pixel value is lifted after processing, when wherein the pixel in motion history image is current time stamp, mark Pixel value it is maximum；

Second step carries out Morphological scale-space to obtained image, removes the cavity in noise spot and blank map picture；

3rd step carries out contour detecting to gained image, and independent region is represented with rectangle, is partitioned into moving target car , corresponding threshold value is set to rectangle to remove interference of other targets to vehicle.

The full-automatic capturing system 4. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that Vehicle shadow minimizing technology uses the cast shadow suppressing method based on color notation conversion space, can not be effective in RGB color Segmentation shade is converted to hsv color spatial manipulation with moving target, using the cast shadow suppressing method of color notation conversion space, pixel When being covered by shade, the brightness V of the pixel is dimmed, and saturation degree S can decline, and colourity H changes less, before analysis Sight spot and the heterochromia of background dot, shadow templates are defined as follows to each pixel of motion target area：

<mrow> <msup> <msub> <mi>Sh</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>C</mi> <mo>)</mo> </mrow> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <mi>&amp;alpha;</mi> <mo>&amp;le;</mo> <mfrac> <mrow> <msup> <msub> <mi>V</mi> <mi>k</mi> </msub> <mi>F</mi> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> </mrow> <mrow> <msup> <msub> <mi>V</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mi>B</mi> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>&amp;le;</mo> <msup> <msub> <mi>&amp;beta;andS</mi> <mi>k</mi> </msub> <mi>F</mi> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <msup> <msub> <mi>S</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mi>B</mi> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;le;</mo> <msub> <mi>T</mi> <mi>s</mi> </msub> <mi>a</mi> <mi>n</mi> <mi>d</mi> <mo>|</mo> <msup> <msub> <mi>H</mi> <mi>k</mi> </msub> <mi>F</mi> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <msup> <msub> <mi>H</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mi>B</mi> </msup> <mrow> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> </mrow> <mo>|</mo> <mo>&amp;le;</mo> <msub> <mi>T</mi> <mi>H</mi> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>v</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>

Wherein, Sh_k ^(C)(x, y) is image coordinate point (x, y) place direct-shadow image vegetarian refreshments mask, V_k ^F(x, y), S_k ^F(x, y), H_k ^F(x, y) With V_k-1 ^B(x, y), S_k-1 ^B(x, y), H_k-1 ^B(x, y) difference denotation coordination point (x, y) place's current pixel value and previous moment background picture V, S, H component of element value, if pixel is judged as shade, pixel mask Sh_k ^(C)(x, y) is 1, otherwise is 0, threshold value α is factually set, if the cast shadow in background is stronger, α settings are smaller, and threshold value beta is used for the brightness for ensureing current pixel point It can not be got too close to image background, T_SAnd T_HThe threshold value of saturation degree difference and colour difference respectively between shade and background, obtains shade After template, moving target is replaced to be mistaken for the pixel value of shade with the pixel value of background image.

The full-automatic capturing system 5. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that The light stream of distinguished point based light stream estimation Matching pursuitalgorithm is the apparent motion of gradation of image pattern, and optical flow field is two-dimentional instantaneous Velocity field, two-dimension speed vector therein are projection of the three dimensional velocity vectors in imaging surface of visible point in scenery, and light stream is taken The three-dimensional vehicle structural information of band includes movable information；

The central point of vehicle characteristics point selection vehicle, the initial time for stopping region is disobeyed when vehicle target enters, detects vehicle simultaneously Information of vehicles is depicted with rectangle, obtains size, the positional information of vehicle, the central point of rectangle is set to vehicle center point, In follow-up video sequence, light stream estimation is carried out to this central point and is tracked matching.When find this feature leave it is separated Stop region, then discharge the resource of distribution, carry out subsequent treatment, matching criterior is set according to two aspects during tracking：In The distance criterion of heart point movement, central point moving direction criterion.

The full-automatic capturing system 6. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that The idiographic flow of distinguished point based light stream estimation Matching pursuitalgorithm is as follows：

Step 1, detect whether that vehicle enters to disobey and stopped region, believed if then using vehicle detecting algorithm to obtain vehicle characteristics Breath；

Step 2, make light stream estimation, matched after trying to achieve central point according to matching criterior, such as match unsuccessful explanation step 1 In could not accurately detect vehicle, go to step five, if successful match, go to step three；

Step 3, judges whether vehicle stops stopping in region separated, two is gone to step if vehicle does not stop, if vehicle stops Rotation stop step 4, stop region if vehicle leaves to disobey and go to step five；

Step 4, into the process part to parking offense vehicle；

Step 5, cancel vehicle characteristic information, free system resources.

The full-automatic capturing system 7. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that High-speed ball camera is captured and the coordinate information and license board information of Car license recognition module capture vehicles peccancy, when having detected vehicle During generation act of violating regulations, current video image is caught, this image is the distant view image of vehicle, rotates ball machine and furthers and accurately determines Position target, the close shot picture of a vehicle is captured, Car license recognition is carried out to this pictures after acquisition success, preserves what is identified Car plate, if identification is unsuccessfully repeated in Car license recognition, abandons after multiple recognition failures and this time capture, after Car license recognition, hair Golf emitter control command makes ball machine return to monitoring presetting bit, after being spaced setting time, then captures another Zhang Yuanjing vehicle figure Piece, three pictures of acquisition are combined as a pictures, and are superimposed picture evidence of the watermark information as parking offense.

The full-automatic capturing system 8. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that Vehicle peccancy evidence processing module by parking offense place, disobey between the stopping time, the number-plate number, disobey stop behavior, picture path, information are deposited Store up in database, and be first class catalogue by place, the time is that second-level directory preserves to picture in disk.

The full-automatic capturing system 9. a kind of precisely quick road violation according to claim 1 is stopped, it is characterised in that Software security maintenance module is safeguarded to running software and ensures the stability of software systems operation free of discontinuities, and system journal includes Main contents have：When stopping capturing time, startup candid photograph time, software startup time, software shut-in time, collection card initialization Between.

The full-automatic capturing system 10. a kind of precisely quick road violation according to claim 1 is stopped, its feature exist In, server end is the centralized management storage center of the whole network in central management system, manages the separated of the whole network concentratedly and stops capturing point, and The separated of storage the whole network stops capturing evidence, and management client is the interface of user and whole system, and multiple users pass through different clients End accesses whole system simultaneously, and central management system includes four subsystems：

First, real-time video monitoring subsystem, user checks that the whole network disobeys the real-time video for stopping capturing point by management system；

Second, control subsystem of setting up defences, user carries out disobeying the setting for stopping capturing algorithm, deployed troops on garrison duty, operation of withdraw, when being arranged on some Between section deploy troops on garrison duty, and other time section is then in the air；

Stop Query Subsystem third, disobeying, User Defined condition query, which is disobeyed, to stop alerting, and checks that certain is disobeyed the alarm picture for stopping alarm With alarm process video recording；

Fourth, alarm export subsystem, user will disobey the form for stopping alarm and exporting as txt, html, excel form.