CN106986272A - It is a kind of to prevent slinging method and system based on the container container car that machine vision is tracked - Google Patents

Abstract

Prevent slinging system based on the container container car that machine vision is tracked the invention discloses a kind of, including tyre crane, camera, container car, container, video alarm machine, Central Control Room control device, camera is installed on the tyre crane bottom bracket, height is concordant with the vehicle frame of the container car, and the visual field of camera is perpendicular to container car travel direction；Camera inputs the video image for tracking and obtaining to video alarm machine, video alarm machine is handled the video image of acquisition using the track algorithm based on fast Fourier transform on-line study, when judging that container car is lifted, alarm signal is sent to Central Control Room control device.The system can automatic detection container whether separated with container car, so as to avoid human error from slinging container container car by mistake, realize the safety pre-control to case area storage yard operation.

Description

It is a kind of to prevent slinging method and system based on the container container car that machine vision is tracked

Technical field

Intelligently prevent slinging technical field the invention belongs to container container car, and in particular to one kind is based on machine vision tracking Container container car independently tracking slings method for early warning with anti-.

Background technology

With developing rapidly that global container is transported, the operation in container terminal and stockyard is more and more busier, work Become increasingly complex as environment.The lifting of Nowadays, Container shipping generally utilizes straddle truck or tyre crane, due to container handling operation Particularity, occur often in cargo handling process due to container car lock pin not completely open and by container car by container in the lump band The event risen.When carrying out container hanging operation using tyre crane on stockyard, it can generally be divided into following three kinds of works Condition：(1), now, there is more uniform gap, be referred to as " complete in container and container car separation between container and container car Fully separating state "；(2) container and container car are partially separated, i.e. one end of container is separated with container car, and the other end does not divide From, now, between container and container car there is certain gap, but be in the small triangle gap in the big one side in one side, it is referred to as For " being not completely separated from state " or " unilateral released state "；(3) container and container car two ends are not separated, now container and Basic gapless between container car, is referred to as " not released state completely ".Operating mode is planted for (2) and (3), can be led when serious Cause to damage container and truck, or even can occur driver's casualty accident.

In order to which the accident for preventing container car and container from taking up in the lump occurs, harbour and stockyard are generally imaged using camera Technology, tyre crane driver monitors the separate condition of container and container car by drivers' cab display, and lifting operation person passes through words Business machine links up field condition with driver, and the generation of accident is avoided by manual operation, i.e., avoids accident by " people's air defense " completely Occur.Currently, container pier storage yard area is big, and operation car type is more and vehicle condition is complicated, for large-scale Active workings, machine The allotment and management of tool equipment are extremely important, and fatigue and carelessness are easily caused by people, and rely on conventional wireless intercom to assign There is inconvenience in job instruction and transmission production information, greatly affected operating efficiency.Obviously, current " people's air defense " measure The generation of above-mentioned accident can not be prevented effectively from.

Thus, in order to tackle increasingly heavy and complicated Container Transport industry, it is necessary to a kind of automatic based on machine vision Whether detection container separates with container car, and automatic in the case of not separating or prompting stops the " skill of lifting mechanism action It is anti-" measure, it is to avoid the accident that container car and container take up in the lump occurs.

The content of the invention

The technical problems to be solved by the invention are to research and propose a kind of container container car tracked based on machine vision Autonomous tracking slings method for early warning with anti-, and container car running region is monitored in real time, and the potential accident of slinging is sent Alarm signal, can effectively prevent storage yard container container car to be lifted the generation of accident, increasingly complicated and heavy to adapt to Container Transport industry.

Prevent slinging method, including following step based on the container container car that machine vision is tracked the invention discloses a kind of Suddenly：

Step 1: camera gathers current time t video image, container container car corresponding region video image is obtained, Determine the width and height of the video image；

Step 2: image the latter half is tracking area-of-interest in selected camera visual field, area-of-interest will be tracked Middle to delimit several tracking sub-districts, tracking sub-district number is the odd number not less than 3；Each tracking sub-district is rectangle, its position Meet the topological relation of setting；

Step 3: camera collection subsequent time t+1 video image, video alarm machine is to each in the step 2 Sub-district uses the track algorithm based on fast Fourier transform on-line study, estimates the position of each sub-district target shape frame；

Step 4: judging the tracking result of each sub-district, if the upward vertical movement for having more than half sub-district exceedes N number of picture Element；If so, then judging that counter vehicle carriage is lifted；If it is not, then judging that counter vehicle carriage is not lifted, and is continued with T seconds time For interval, each sub-district is reset according to step 2 automatically, and repeats step 3 to step 4；

Step 5: if it is determined that counter vehicle carriage is lifted, then video alarm machine sends early warning letter to Central Control Room control device Number, remind hoistman pause to sling action；Otherwise, hoistman completes normal lifting according to program；

Wherein, the camera visual field is perpendicular to container car travel direction, the camera to video alarm machine input with The video image that track is obtained；T=1,2,3 ..., M, M be tracking frame of video sum, N is number of pixels, T for setting video figure As the time interval of collection.

Preferably, the tracking sub-district quantity delimited in the step 2 is 3.

Preferably, the method for selected tracking area-of-interest is in the step 2：

1) sequence of video images t two field pictures are obtained, represent to track area-of-interest, wherein T_ with rectangle frame T_Region Region four element representations, respectively upper left angle point abscissa T_Region.x, upper left angle point ordinate T_Region.y, Rectangle width of frame T_Region.width, rectangle frame height T_Region.height；

2) target area of container carrier frame section is included in tracking rectangle frame T_Region；The target area Selection rule is as follows：

Upper left angle point abscissa T_Region.x=0,

Upper left angle point ordinate

Rectangle width of frame T_Region.width=image.width,

Rectangle frame height

3) tracking sub-district box_1 central point is (x1, y1), and tracking sub-district box_2 central point is (x2, y2), tracking Sub-district box_3 central point is (x3, y3), and the width or height of each tracking sub-district take For tracking The height of ratio or tracking sub-district that the width of sub-district accounts for video image width accounts for the ratio of video image height；

4) three sub- zone position topological relation selection rules are：

X2=0.5*image.width,

Y2=T_Region.y+0.5*T_Region.height,

It is high that the height of the ratio or tracking sub-district that account for video image width for the width of tracking sub-district accounts for video image The ratio of degree；Q is the number of pixels of three tracking sub-district overlapping regions.

Preferably, the method for tracking target based on fast Fourier transform on-line study is in the step 3：

(1) tracking target area is initialized；

(2) according to t frame target following results, build t frames tracking target context prior probability model P (c (z) | o)；

(3) according to t frame target following results, the confidence level distribution map c (z) that t frames track target is built；

(4) the spatial context model that t frames track target is built

(5) according to t frame target following results, t+1 frame video images data is calculated and carry out target following, tracking is obtained The position coordinates of target in the current frameWherein t=1,2,3 ..., M-1, M for tracking video image frame number.

Preferably, the described pair of method that is initialized of tracking target area is：

(1) sequence of video images t two field pictures, the position of initialization tracking target area are obtained；

(2) tracking target context relevant range Context_Region is determined according to initialization result；

(3) Hanning window mouthful matrix M is defined_hmwindow；

(4) initialization size factor σ_tWith change of scale parameter s_t。

Preferably, the method for the position coordinates of the acquisition target in the current frame is：

(1) tracking target context prior probability model P (c (z) | o) is built according to t+1 two field pictures；

(2) the space-time context model that t+1 frames track target is built

(3) the confidence level distribution map c that t+1 frames track target is built_t+1(z)；

(4) the position point coordinates that t+1 frames track target is calculated

(5) size factor σ is updated_t；

(6) the spatial context model that t+1 frames track target is updated

Preferably, the tracking target context prior model P (c (z) | o) computational methods are：

Wherein I (z) represents tracking target area T_Region grey scale pixel value through past average value processing, and it is peaceful to be multiplied by the Chinese Window matrix is obtained, and z represents the pixel coordinate in T_Region,Representing matrix multiplying,

I (z)=I (z)-mean (I (z))

Wherein x^*For tracking target's center point coordinates, mean () represents image average, and a is normalized parameter, its value For

Preferably, the confidence level distribution map c (z) of target area computational methods are：

Wherein b is normalized parameter, and its value is

Preferably, the spatial context modelComputational methods be：

Preferably, the t+1 frames track the space-time context model of targetComputational methods be：

Wherein ρ is Studying factors.

Preferably, the t+1 frames track the confidence level distribution map c of target_t+1(z) computational methods are：

Preferably, the t+1 frames track the position point coordinates of targetComputational methods are：

Preferably, the renewal size factor σ_tComputational methods be：

WhereinIt is the confidence level distribution map that t frames track target, s '_tIt is to estimate according to continuous two frames tracing figure picture The target scale of meter, is represented with the ratio of t frames and t-1 frame objective degrees of confidence result of calculations；It is according to continuous n recently The average value of the target scale of frame tracking Image estimation, s_t+lIt is the target scale that Image estimation is tracked according to a new frame, λ ＞ 0 are The fixed value filtering parameter of setting.

Preferably, the spatial context model that t+1 frames track target is updatedComputational methods are：

Prevent slinging system, including tire based on the container container car that machine vision is tracked the invention also discloses a kind of Hang, camera, container car, container, video alarm machine, Central Control Room control device, it is characterised in that：The camera is installed on The tyre crane bottom bracket, the highly vehicle frame with the container car are concordant, and the visual field of the camera is perpendicular to counter garage Sail direction；The camera inputs the video image for tracking and obtaining to the video alarm machine, and the video alarm machine uses base The video image of acquisition is handled in the track algorithm of fast Fourier transform on-line study, when judgement container car is lifted When, send alarm signal to Central Control Room control device.

Beneficial effects of the present invention are as follows：

(1) independently tracking slings the pre- police to the container container car disclosed by the invention tracked based on machine vision with anti- Method, the track algorithm based on fast Fourier transform on-line study monitors, obtains and handled the video figure in storage yard operation area in real time Picture, realizes the significant improvement that " people's air defense " arrives " technical precaution ", can be prevented effectively from container container car quilt caused by artificial maloperation Sling the generation of accident；

(2) independently tracking slings the pre- police to the container container car disclosed by the invention tracked based on machine vision with anti- Method, using the track algorithm based on fast Fourier transform on-line study, will be tracked in stockyard area-of-interest be divided into it is multiple Sub-district, the tracking result for multiple sub-districts is compared, can realize to storage yard operation area it is comprehensive, without dead angle monitoring, together When ensure the accuracy of alarming result.

Brief description of the drawings

Fig. 1, which is that the container container car tracked based on machine vision in the present invention is anti-, slings system schematic；

Wherein, (a) is system front view, and (b) is system whole scene schematic diagram, and (c) is layout of system equipment's schematic diagram；

1-the first camera, 2-second camera, 3-container car, the 4-the first video alarm machine, the 5-the second video report Alert machine, 6-Central Control Room, 7-control device, 8-tyre crane, 9-tyre crane bottom bracket, 10-counter vehicle carriage, 11-collection Vanning, the right-hand member that 9-1 is the left end of tyre crane bottom bracket, 9-2 is tyre crane bottom bracket；

Fig. 2, which is that the container container car tracked based on machine vision in the present invention is anti-, slings method schematic diagram；

Fig. 3 is machine vision tracing area and multiple subarea video tracking schematic diagram in the present invention.

Embodiment

For the purposes of the present invention, technical scheme and advantage are more clearly understood, below in conjunction with the accompanying drawings and specific embodiment party Formula is described in further details to technical scheme.

As shown in Fig. 1 (a), 1 (b), 1 (c), Fig. 2, independently tracked based on the container container car that machine vision is tracked with preventing Method for early warning is sling to comprise the following steps：

Step 1: the first camera 1, second camera 2 gather current time t video image, container container car 3 is obtained The video image of corresponding visual field, wherein image.width are the video image width, and image.height is the video Picture altitude.

Step 2: as shown in figure 3, image the latter half is interested tracking to track area-of-interest in selected visual field Tracking sub-district box_1, box_2 and box_3 delimited in region.Three sub-districts are square, and its position meets the topology of setting Relation.Tracking target area is initialized.Specific method is：

(1) sequence of video images t two field pictures are obtained, represent to track area-of-interest with rectangle frame T_Region, wherein T_Region four element representations, respectively upper left angle point abscissa T_Region.x, upper left angle point ordinate T_ Region.y, rectangle width of frame T_Region.width, rectangle frame height T_Region.height.Rectangle frame T_Region should Include the target area of the part of Containerizable Cargo vehicle carriage 10.

(2) target area of the part of Containerizable Cargo vehicle carriage 10 should be included in tracking rectangle frame T_Region.The target The selection rule in region is as follows：

Upper left angle point abscissa T_Region.x=0,

Upper left angle point ordinateM takes 2 in the present embodiment；

Rectangle width of frame T_Region.width=image.width,

Rectangle frame heightN is taken to take 2 in the present embodiment.

(3) tracking sub-district box_1 central point is (x1, y1), and tracking sub-district box_2 central point is (x2, y2), tracking Sub-district box_3 central point is (x3, y3), and the width and height of each tracking sub-district take For tracking sub-district The height of ratio or tracking sub-district that width accounts for video image width accounts for the ratio of video image height.P is taken in the present embodiment =10.

(4) three sub- zone position topological relation selection rules are：

X2=0.5*image.width,

Y2=T_Region.y+0.5*T_Region.height,

It is high that the height of the ratio or tracking sub-district that account for video image width for the width of tracking sub-district accounts for video image The ratio of degree；Q is to take q=10 in the number of pixels of three tracking sub-district overlapping regions, the present embodiment.

Step 3: using the target following side based on fast Fourier transform on-line study to each sub-district in step 2 Method, estimates the position of three sub-district target shape frames.Illustrate the target based on fast Fourier transform on-line study by taking box1 as an example Tracking, box2 and box3 use identical tracking.Specific method is：

(1) sequence of video images t two field pictures are obtained, sub-district box1 is represented with rectangle frame Target_Region, wherein Target_Regio tetra- element representations of n, respectively upper left angle point abscissa Target_Region.x, upper left angle point is vertical to be sat Mark Target_Region.y, rectangle width of frame Target_Region, rectangle frame height Target_Region.height.

The target area center point coordinate is centerPoint, and its abscissa is：

CenterPoint.x=Target_Region.x+Target_Region.width*0.5

Ordinate is：

CenterPoint.y=Target_Region.y+Target_Region.height*0.5.

(2) tracking target context relevant range Context_Region is determined according to rectangle frame Target_Region.This Conte_xt R length and wide respectively 2 times of Target_Reg, central point and target rectangle frame Target_ is chosen in embodiment Region central points are overlapped.I.e. its is a width of：

Context_Region.width=Target_Region.width*2

It is a height of：

Context_Region.height=Target_Region.height*2

Then its upper left angle point abscissa is：

Context_Region.x=centerPoint.x-Context_Region.width*0.5

Its upper left angle point ordinate is：

Context_Region.y=centerPoint.y-Context_Region.height*0.5

(3) Hanning window mouthful matrix is defined, to reduce the frequency influence that image border is brought to Fourier transform.Hanning window with M_hmwindowRepresent, it is wide consistent with Context_Region with height, i.e.,

Hmwindow.width=Context_Region.width

Hmwindow.height=Context_Region.height

Each position is defined as follows in Hanning window matrix：

Hmwindow (i, j)=

(0.54-0.46*cos(2*π*i/hmwindow.height))*(0.54-0.46*cos(2*π*j/ hmwindow.width))

Wherein, i=0,1,2 ..., hmwindow.height-1, j=0,1,2 ..., hmwindow.width-1, π take 3.14。

(4) initialization size factor σ_t=(T_Region.width+T_Region.height) * 0.5, change of scale ginseng Number s_t=1.

(5) tracking target context prior probability model is obtained

Wherein I_t(z) by target area T_Region grey scale pixel value through past average value processing, and it is multiplied by Hanning window square Battle array is obtained：

I_t(z)=I_t(z)-mean(I_t(z))

Wherein z represents the pixel coordinate in Target_Region,Representing matrix multiplying, x^*For tracking target's center Point coordinates, i.e. centerPoint.A is normalized parameter, and its value is：

(6) tracking objective degrees of confidence distribution map is obtained

Wherein b is normalized parameter, and its value is：

α takes 2.25, β to take 1 in the present embodiment.

(7) tracking object space context model is set up

Wherein F () represents leaf transformation computing, F in quick Fourier^-1() represents fast fourier inverse transformation computing.

(8) according to t frame target following results, t+1 frame video image data are obtained, and calculate t+1 frame video figures As data carry out target following, the position coordinates of tracking target in the current frame is obtainedWherein t=1,2,3 ..., M, M be Track frame of video sum.Specific method is as follows：

1) tracking target context prior probability model is obtained

The same step 3 of specific method (5).

2) the space-time context model that t+1 frames track target is set up

Wherein ρ is Studying factors, and ρ takes 0.075 in the present embodiment.

3) the confidence level distribution map c that t+1 frames track target is calculated_t+1(z), computational methods are

Wherein represent dot product operation.

4) calculating tracking position of object point coordinates in t+1 two field pictures is

5) size factor σ is updated_t, computational methods are：

Wherein,It is the confidence level distribution map that t frames track target, s '_tIt is to estimate according to continuous two frames tracing figure picture The target scale of meter, is represented with the ratio of t frames and t-1 frame objective degrees of confidence result of calculations；It is according to continuous n recently The average value of the target scale of frame tracking Image estimation, s_t+1It is the target scale that Image estimation is tracked according to a new frame.In order to anti- Only produce over adaptation problem and reduce the noise pollution as caused by evaluated error, mesh in a new frame is estimated using filtering method Target size, λ ＞ 0 are the fixed value filtering parameters of setting.

6) the spatial context model that t+1 frames track target is updated

Step 4: three sub-district tracking results are judged, if two or more upward vertical movement is more than N number of Pixel, then judge that counter vehicle carriage 4 is lifted.Otherwise, it is determined that counter vehicle carriage 4 is not lifted.Using T seconds time as interval, I.e. every T seconds, each sub-district is reset according to step 2 automatically.N takes 10, T to take 5 in the present embodiment.

Step 5: if it is determined that counter vehicle carriage 4 is lifted, then sending pre-warning signal to Central Control Room control device 7, prompting is hung Action is sling in car driver pause, and links up field condition with truck drivers.Otherwise, hoistman completes normal rise according to program Hang.

When delimiting tracking sub-district in tracking area-of-interest, sub-district number is the odd number not less than 3, in order to carry out The judgement of tracking result, is prevented effectively from container container car and slings phenomenon.By area-of-interest to be tracked in the present embodiment 3 are divided into, at this time running efficiency of system highest.

Independently tracking slings early warning system to a kind of container container car tracked based on machine vision with anti-, is taken the photograph by first As first 1, second camera 2 carries out real-time tracking to the running region of container car 3, when finding that container car 3 is moved upwards, illustrate goods Cabinet car 3 is lifted together with container 11, then the first video alarm machine 4, the second video alarm machine 5 send alarm signal, is led to Crossing the control device 7 of Central Control Room 6 notifies hoistman to stop lifting in time, so as to prevent accident.Wherein, the first camera 1, First camera 2 is installed on the left end 9-1 and right-hand member 9-2 of the bottom bracket 9 of tyre crane 8, highly concordant with counter vehicle carriage 10, depending on Field is perpendicular to the travel direction of container car 3, and the first camera 1, second camera 2 are respectively to the first video alarm machine 4, the second video The video image that the input tracking of alarming machine 5 is obtained.The first video alarm machine 4, the second video alarm machine 5 are using based on quick The track algorithm of Fourier's series on-line study is handled the video image of acquisition, when judging that container car is lifted, to Central Control Room control device 7 sends alarm signal, reminds hoistman pause to sling action, and link up live feelings with truck drivers Condition.Otherwise, hoistman completes normal lifting according to program.

Independently tracking slings method for early warning to container container car disclosed by the invention with anti-, becomes using based on fast Flourier Change the track algorithm of on-line study, storage yard operation area can be carried out in real time, comprehensively to monitor, while accurately believed Breath processing and judgement, realize the significant improvement that " people's air defense " arrives " technical precaution ", can be prevented effectively from packaging caused by artificial maloperation Case container car is lifted the generation of accident.

Obviously, described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.It is based on Embodiments of the invention, the every other implementation that those skilled in the art are obtained under the premise of creative work is not made Example, belongs to the scope of protection of present invention.

Claims (15)

1. a kind of prevent slinging method based on the container container car that machine vision is tracked, comprise the following steps：

Step 1: camera gathers current time t video image, container container car corresponding region video image is obtained, it is determined that The width and height of the video image；

Step 2: image the latter half is tracking area-of-interest in selected camera visual field, it will be drawn in tracking area-of-interest Several fixed tracking sub-districts, tracking sub-district number is the odd number not less than 3；Each tracking sub-district is rectangle, and its position is met The topological relation of setting；

Step 3: camera collection subsequent time t+1 video image, video alarm machine is to each sub-district in the step 2 Using the track algorithm based on fast Fourier transform on-line study, the position of each sub-district target shape frame is estimated；

Step 4: judging the tracking result of each sub-district, if the upward vertical movement for having more than half sub-district exceedes N number of pixel； If so, then judging that counter vehicle carriage is lifted；If it is not, then judging that counter vehicle carriage is not lifted, continue using T seconds time between Every each sub-district is reset according to step 2 automatically, and repeats step 3 to step 4；

Step 5: if it is determined that counter vehicle carriage is lifted, then video alarm machine sends pre-warning signal to Central Control Room control device, carries Action is sling in hoistman pause of waking up；Otherwise, hoistman completes normal lifting according to program；

Wherein, the camera visual field is perpendicular to container car travel direction, and the camera is obtained to the input tracking of video alarm machine The video image obtained；T=1,2,3 ..., M, M be tracking frame of video sum, N is number of pixels, and T adopts for the video image of setting The time interval of collection.

2. container container car according to claim 1 is anti-to sling method, it is characterised in that delimited in the step 2 It is 3 to track sub-district quantity.

3. container container car according to claim 2 is anti-to sling method, it is characterised in that selected in the step 2 with The method of track area-of-interest is：

1) sequence of video images t two field pictures are obtained, represent to track area-of-interest, wherein T_ with rectangle frame T_Region Region four element representations, respectively upper left angle point abscissa T_Region.x, upper left angle point ordinate T_Region.y, Rectangle width of frame T_Region.width, rectangle frame height T_Region.height；

2) target area of container carrier frame section is included in tracking rectangle frame T_Region；The selection of the target area Rule is as follows：

Upper left angle point abscissa T_Region.x=0,

Upper left angle point ordinate

Rectangle width of frame T_Region.width=image.width,

Rectangle frame height

3) tracking sub-district box_1 central point is (x1, y1), and tracking sub-district box_2 central point is (x2, y2), tracks sub-district Box_3 central point is (x3, y3), and the width or height of each tracking sub-district take For tracking sub-district Width account for the ratio of video image width or track the height of sub-district and account for the ratio of video image height；

4) three sub- zone position topological relation selection rules are：

X2=0.5*image.width,

Y2=T_Region.y+0.5*T_Region.height,

$x1=0.5*image.width-\frac{1}{P}*image.height+q,$

$y1=T\_\mathrm{Re}gion.y+0.5*T\_\mathrm{Re}gion.height-\frac{1}{P}*image.height+q,$

$x3=0.5*image.width+\frac{1}{P}*image.height-q,$

$y3=T\_\mathrm{Re}gion.y+0.5*T\_\mathrm{Re}gion.height+\frac{1}{P}*image.height-q;$

To track the ratio that the height of ratio or tracking sub-district that the width of sub-district accounts for video image width accounts for video image height Example；Q is the number of pixels of three tracking sub-district overlapping regions.

4. sling method according to any described container container car of Claims 2 or 3 is anti-, it is characterised in that the step 3 In the method for tracking target based on fast Fourier transform on-line study be：

(1) tracking target area is initialized；

(2) according to t frame target following results, the context prior probability model P (c (z) | o) of t frames tracking target is built；

(3) according to t frame target following results, the confidence level distribution map c (z) that t frames track target is built；

(4) the spatial context model that t frames track target is built

(5) according to t frame target following results, t+1 frame video images data is calculated and carry out target following, obtain tracking target Position coordinates in the current frameWherein t=1,2,3 ..., M-1, M for tracking video image frame number.

5. container container car according to claim 4 is anti-to sling method, it is characterised in that described pair of tracking target area The method initialized is：

(1) sequence of video images t two field pictures, the position of initialization tracking target area are obtained；

(2) tracking target context relevant range Context_Region is determined according to initialization result；

(3) Hanning window mouthful matrix M is defined_hmwindow；

(4) initialization size factor σ_tWith change of scale parameter s_t。

6. container container car according to claim 4 is anti-to sling method, it is characterised in that the acquisition target is current The method of position coordinates in frame is：

(1) tracking target context prior probability model P (c (z) | o) is built according to t+1 two field pictures；

(2) the space-time context model that t+1 frames track target is built

(3) the confidence level distribution map c that t+1 frames track target is built_t+1(z)；

(4) the position point coordinates that t+1 frames track target is calculated

(5) size factor σ is updated_t；

(6) the spatial context model that t+1 frames track target is updated

7. sling method according to any described container container car of claim 4 or 6 is anti-, it is characterised in that the tracking mesh The computational methods for putting on hereafter prior model P (c (z) | o) are：

$P\left(c\right(z\left)\right|o)=I\left(z\right)\⊗w_{\mathrm{\σ}}(z-x^{*})$

Wherein I (z) represents tracking target area T_Region grey scale pixel value through past average value processing, and is multiplied by Hanning window square Battle array is obtained, and z represents the pixel coordinate in T_Region,Representing matrix multiplying,

I (z)=I (z)-mean (I (z))

$I\left(z\right)=I\left(z\right)\⊗M_{hmwindow}$

$w_{\mathrm{\σ}}(z-x^{*})={\mathrm{ae}}^{-\frac{|z-x^{*}|}{2{\mathrm{\σ}}^2}}$

Wherein x^*For tracking target's center point coordinates, mean () represents image average, and a is normalized parameter, and its value is

$a=\frac{1}{\underset{z\∈{\mathrm{\Ω}}_{T\_rect}}{\Σ}{e}^{-\frac{|z-x^{*}|}{2{\mathrm{\σ}}^2}}}.$

8. container container car according to claim 4 is anti-to sling method, it is characterised in that the confidence of the target area Degree distribution map c (z) computational methods be：

$c\left(z\right)={\mathrm{be}}^{-|\frac{z-x^{*}}{\mathrm{\α}}{|}^{\mathrm{\β}}}$

Wherein b is normalized parameter, and its value is

$b=\frac{1}{\underset{z\∈{\mathrm{\Ω}}_{T\_rect}}{\Σ}{e}^{-|\frac{z-x^{*}}{\mathrm{\α}}{|}^{\mathrm{\β}}}}.$

9. container container car according to claim 4 is anti-to sling method, it is characterised in that the spatial context modelComputational methods be：

${h_t}^{sc}\left(x\right)=F^{-1}\left(\frac{F\left({\mathrm{be}}^{-|\frac{x-x^{*}}{\mathrm{\α}}{|}^{\mathrm{\β}}}\right)}{F\left(\right(I\left(x\right)w_{\mathrm{\σ}}(z-x^{*}))}\right).$

10. container container car according to claim 6 is anti-to sling method, it is characterised in that the t+1 frames track target Space-time context modelComputational methods be：

$H_{t+1}^{stc}\left(z\right)=(1-\mathrm{\ρ})H_t^{stc}\left(z\right)+{\mathrm{\ρh}}_t^{sc}\left(z\right)$

Wherein ρ is Studying factors.

11. container container car according to claim 6 is anti-to sling method, it is characterised in that the t+1 frames track target Confidence level distribution map c_t+1(z) computational methods are：

$c_{t+1}\left(z\right)=F^{-1}\left(F\right(H_{t+1}^{stc}\left(z\right))\·F(I_{t+1}\left(z\right)\⊗w_{{\mathrm{\σ}}_t}\left(z-x_t^{*}\right)\left)\right).$

12. container container car according to claim 6 is anti-to sling method, it is characterised in that the t+1 frames track target Position point coordinatesComputational methods are：

$x_{t+1}^{*}=\underset{x\∈{\mathrm{\Ω}}_c\left(x_t^{*}\right)}{\mathrm{argmax}}{c}_{t+1}\left(x\right).$

13. container container car according to claim 6 is anti-to sling method, it is characterised in that the renewal size factor σ_t Computational methods be：

$\left\{\begin{array}{c}{s}_t^{\′}=\sqrt{\frac{c_t\left(x_t^{*}\right)}{c_{t-1}\left(x_{t-1}^{*}\right)}}\\ {\stackrel{\‾}{s}}_t=\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{s}_{t-i}^{\′}\\ s_{t+1}=(1-\mathrm{\λ})s_t+\mathrm{\λ}{\stackrel{\‾}{s}}_t\\ {\mathrm{\σ}}_{t+1}=s_t{\mathrm{\σ}}_t\end{array}\right.$

WhereinIt is the confidence level distribution map that t frames track target, s '_tIt is the mesh that Image estimation is tracked according to continuous two frame Scale, is represented with the ratio of t frames and t-1 frame objective degrees of confidence result of calculations；It is according to continuous n frames tracking recently The average value of the target scale of Image estimation, s_t+1It is the target scale that Image estimation is tracked according to a new frame, λ ＞ 0 are settings Fixed value filtering parameter.

14. container container car according to claim 6 is anti-to sling method, it is characterised in that update t+1 frames tracking mesh Target spatial context modelComputational methods are：

$h_{t+1}^{sc}\left(z\right)=F^{-1}\left(\frac{F\left({\mathrm{be}}^{-|\frac{z-{x_t}^{*}}{\mathrm{\α}}{|}^{\mathrm{\β}}}\right)}{F\left(I_{t+1}\right(z)\⊗w_{{\mathrm{\σ}}_t}(z-{x_t}^{*}\left)\right)}\right).$

15. a kind of container container car tracked based on machine vision is anti-to sling system, including tyre crane, camera, container car, Container, video alarm machine, Central Control Room control device, it is characterised in that：The camera is installed on tyre crane bottom branch Frame, the highly vehicle frame with the container car are concordant, and the visual field of the camera is perpendicular to container car travel direction；The camera The video image that tracking is obtained is inputted to the video alarm machine, the video alarm machine is used to be existed based on fast Fourier transform The track algorithm of line study is handled the video image of acquisition, when judging that container car is lifted, and is set to Central Control Room control Preparation goes out alarm signal.