CN106530825B - Electric bicycle and mechanical transport collision detection method based on ST-MRF model - Google Patents
Electric bicycle and mechanical transport collision detection method based on ST-MRF model Download PDFInfo
- Publication number
- CN106530825B CN106530825B CN201611030277.4A CN201611030277A CN106530825B CN 106530825 B CN106530825 B CN 106530825B CN 201611030277 A CN201611030277 A CN 201611030277A CN 106530825 B CN106530825 B CN 106530825B
- Authority
- CN
- China
- Prior art keywords
- electric bicycle
- motion vector
- block
- conflict
- pixels
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/166—Anti-collision systems for active traffic, e.g. moving vehicles, pedestrians, bikes
Abstract
The invention discloses electric bicycles and mechanical transport collision detection method based on ST-MRF model, it is characterized in that, include the following steps: step 1, traffic camera shooting 3 d video images, single electric bicycle and automobile are tracked simultaneously using ST-MRF model, obtain the traffic information of single electric bicycle and automobile；The 3 d video images that traffic camera is shot are converted into two-dimensional coordinate data by coordinates transformation method by step 2；The stop of step 3, prediction electric bicycle and automobile in next frame image；Step 4 establishes driver's conflict judgment criterion according to the distance between stop.By vehicle movement track following, the traffic conflict detection method between a kind of electric bicycle to make a dash across the red light and the car of normally travel is provided, real-time, accurate traffic information can be provided for traffic management department and identified for accident.
Description
Technical field
The present invention relates to a kind of electric bicycles based on ST-MRF model and mechanical transport collision detection method.
Background technique
As city video equipment quantity, coverage rate increase rapidly, the promotion of Computer Image Processing level, by video
Image processing techniques becomes a kind of trend for City ITS.Traffic transport system engineering and information, 2013 the 13rd
Rolled up for the 3rd phase: 65-70 pages of discloses a kind of adaptive vehicle track algorithm research (Zhou Jun, Cheng Lin) based on ST-MRF, base
In ST-MRF model adaptive vehicle track algorithm can and vehicle bigger in volume of traffic appearance mutually block in the case where,
Accurate acquisition vehicle tracking data, important data basis is provided for later traffic incidents detection.
In some small and medium-sized cities, since population is more intensive, economy is in developing stage, electric bicycle go on a journey quantity compared with
It is more, it goes on a journey commonplace, but electric bicycle driver causes the behavior of making a dash across the red light repeated due to lacking awareness of safety, causes
Traffic accident takes place frequently, and causes a large amount of personnel, property loss.By a large amount of accident statistics, it is found that accident area occurred frequently is exactly
City intersection, so outstanding to traffic conflict detection is carried out between vehicle and the electric bicycle to make a dash across the red light at intersection
It is important.
Traffic conflict detection research in the prior art about electric bicycle and car is very few, herein in current city
At intersection for electric bicycle traffic conflict detection research it is weaker on the basis of, study and propose it is a set of be suitble in
Traffic conflict detection method at the intersection of small city between vehicle and the electric bicycle to make a dash across the red light, and driver is conflicted
Judgment criterion is used in the judgement that conflict spectrum is carried out in video image processing.
Summary of the invention
Conflict with mechanical transport inspection in view of the above-mentioned problems, the present invention provides a kind of electric bicycle based on ST-MRF model
Survey method is provided between a kind of electric bicycle to make a dash across the red light and the car of normally travel by vehicle movement track following
Traffic conflict detection method, can be provided for traffic management department real-time, accurate traffic information for accident identify.
To realize above-mentioned technical purpose and the technique effect, the invention is realized by the following technical scheme:
Electric bicycle and mechanical transport collision detection method based on ST-MRF model, which is characterized in that including as follows
Step:
Step 1, traffic camera shoot 3 d video images, track single electrical salf-walking simultaneously using ST-MRF model
Vehicle and automobile obtain the traffic information of single electric bicycle and automobile, such as vehicle, the volume of traffic, speed and coordinate value；
The 3 d video images that traffic camera is shot are converted into two-dimensional coordinate number by coordinates transformation method by step 2
According to；
The stop of step 3, prediction electric bicycle and automobile in next frame image；
Step 4 establishes driver's conflict judgment criterion according to the distance between stop.
It is preferred that in step 1, the energy function U of ST-MRF modelstmrfIt is:
In formula: first part a (Nyk-μNy)2It indicates in target map, the neighbouring relations of label, second part b (Mxyk-
μMxy)2Indicate the hiding relation of the label in successive objective map, Part IIIIndicate company related with motion vector
Texture relationship in continuous image, Part IVIndicate the fortune in motion vector map
The neighbouring relations of dynamic vector；
Nyk: refer to that the adjacent pixel blocks of a block of pixels and the block of pixels have the number of identical label；
Nxk: indicate the number of the adjacent pixel blocks of a block of pixels；
Dxyk: the texture correlation between the image at t-1 moment and the image of t moment is represented, when blocking, respectively
Calculate the probability for belonging to each vehicle；
Mxyk: the number of pixels of shield portions in two block of pixels of partial occlusion；
μNy: neighborhood group, if using 8- neighborhood group, μ Ny=8 be maximum value；
Ck: current pixel block；
Bk: adjacent pixel blocks；
At (t-1) moment, the difference of the motion vector of current pixel block and adjacent pixel blocks；
A, b, c, f and μMxyFor the parameter of setting.
It is preferred that experimental calibration parameter index: a=1/2, b=1/256, c=32/1000000, f=1/4, μMxy=0.
It is preferred that according to the target map, motion vector and present image of previous moment, while considering the fortune in adjacent block
The similitude of texture relationship in dynamic vector and consecutive image come minimize current time target map and motion vector most
Small energy, specifically comprises the following steps:
01) motion vector that all pieces are obtained by block matching method determines the original state V (t- of motion vector map
1；T)=V0, V (t-1；T) moment t-1 to t, each piece of motion vector are indicated；
02) according to the original state of motion vector map, the initial of target map is set by each piece of candidate label
State X (t)=y0；
03) estimate that target map and motion vector map are carrying out X (t)=yi and V (t-1；T) total after=Vi iteration
Energy, i are the number of iterations；
04) random target map X (t)=yi and motion vector map V (t-1 of the conversion under current state simultaneously；T)=
Target map X (t+1)=yi+1 and motion vector map V (t of the Vi to NextState；T+1)=Vi+1；
05) in step 03) and 04) between iterative cycles operation, until X (t) and V (t-1；T) energy function value is converged to
UstmrfTo minimum.
It is preferred that in step 3, the braking distance S of electric bicycle and automobile in next frame image are as follows:
In formula, ξ is speed；It is attachment coefficient；ψ is road longitudinal grade degree, %, and upward slope is positive, and descending is negative.
It is preferred that in step 4, driver's judgment criterion that conflicts is divided into 4 grades of conflicts: 1 grade, behavior of making a dash across the red light；It is 2 grades, slight to rush
It is prominent；3 grades, danger conflict；4 grades, Serious conflicts or accident occurs, criterion difference is as follows:
When intersection is amber light, and the two-dimensional coordinate of electric bicycle is fallen into the coordinate range of intersection, then is sentenced
It is set to 1 grade of behavior of making a dash across the red light；
When intersection is red light, firstly, by the electric bicycle and car tracing acquisition vehicle in image
Heart X, Y coordinate predict the braking distance of electric bicycle and car with formula 2；Then, estimation two cars are in next frame image
Stop, and the distance between find out two cars stop；Finally, marking the stop of two cars in two-dimensional coordinates
On, and be mapped on real road:
If the extended line of the stop of two cars is non-intersecting, it is determined as 2 grades of slight conflict；
If the distance between stop of two cars in safe range and does not work as two cars of the label on coordinate
When motion profile intersects, then it is determined as 3 grades of dangerous conflict；
If the distance between stop of two cars is less than or equal to 0, it is determined as 4 grades of Serious conflicts or thing occurs
Therefore.
The beneficial effects of the present invention are:
(1) it proposes a kind of ST-MRF model based on optimization while tracking the algorithm of electric bicycle and automobile.The party
Method can not only optimize the spatial distribution of image, moreover it is possible to optimize distribution of the image sequence along time shaft, improve pervious vehicle with
Track algorithm is in the case where the intersection vehicles serious shielding of traffic disturbance the shortcomings that poor robustness.
(2) traffic conflict detection method of the invention is suitable for the intersection of various geometries, is rotated by coordinate
3-D image is converted into 2-D data by algorithm, and (electric bicycle is small for the vehicle in different motion direction at calculating intersection
Automobile) motion profile and two-dimensional coordinate, and experimental data is compared with data measured by commercialization Autoscope software,
It obtains and uses the error of technology traffic data obtained smaller, precision is relatively high.
(3) conflicted judgment criterion according to driver, establish the levels of conflict of 4 grades: 1 grade, behavior of making a dash across the red light；2 grades, slightly
Conflict；3 grades, danger conflict；4 grades, Serious conflicts (accident), and level Four conflict is detected by Success in Experiment.
(4) when the electric bicycle to make a dash across the red light, which occurs 3 grades with car, to conflict, this method energy success prediction is potentially handed over
Interpreter's event, can provide real-time, accurate, reliable traffic information, avoid traffic accident generation for traffic management department.
Detailed description of the invention
Fig. 1 is label flow chart of the present invention；
Fig. 2 is coordinate rotation schematic diagram；
Fig. 3 is vehicle label analogous diagram of the present invention；
Fig. 4 is second level conflict analogous diagram in present invention test；
Fig. 5 is three-level conflict analogous diagram in present invention test；
Fig. 6 is level Four conflict analogous diagram in present invention test.
Specific embodiment
Technical solution of the present invention is described in further detail with specific embodiment with reference to the accompanying drawing, so that ability
The technical staff in domain can better understand the present invention and can be practiced, but illustrated embodiment is not as to limit of the invention
It is fixed.
Electric bicycle and mechanical transport collision detection method based on ST-MRF model, include the following steps:
Step 1, traffic camera shoot 3 d video images, track single electrical salf-walking simultaneously using ST-MRF model
Vehicle and automobile obtain the single traffic information, such as vehicle, the volume of traffic, speed and coordinate value of electric bicycle and automobile etc..
In vehicle tracking algorithm, block of pixels and ST-MRF (Spatial Temporal Markov Random Field)
In scene be it is one-to-one, i.e., each block of pixels and a vehicle communication are got up, that is, assigned to each block of pixels
The label of one corresponding vehicle.When using vehicle tracking algorithm, an initial label distribution is determined first, is then led to
The energy function distribution label crossed in ST-MRF carries out refinement and optimization processing.In extractive process, algorithm considers block in the time
The connection in domain and space (two-dimentional x-y coordinate), that is, the block between consecutive image connection and the connection between adjacent block
Label, by ST-MRF model, is assigned to block of pixels by system.Block corresponds to a scene in ST-MR model, and block is numbered as mesh
The texture of a part and background image for marking region is different, these blocks flock together referred to as target map, on each piece
There is the motion vector for representing block, ST-MRF model is exactly the target that next moment is estimated according to the target map of previous moment
Map, symbolically are as follows:
G (t-1)=g, G (t)=h: gray value of the image G in moment t-1 is g, in moment t, gray value h.It is right
It in each pixel, can indicate are as follows: G (t-1；α, β)=g (α, β), G (t；α, β)=h (α, β), in formula, α, β are that pixel is sat
Mark, G (t-1；α, β)=g (α, β) is meant that: coordinate is α in image G, and gray value of the pixel of β in moment t-1 is equal to figure
As coordinate is α, the corresponding gray value of the pixel of β, G (t in g；α, β)=h (α, β) is meant that: coordinate is α in image G, β's
It is α, the corresponding value of the pixel of β that value of the pixel in moment t, which is equal to coordinate in image h,.
X (t-1)=x, X (t)=y: target map X is detected label in moment t-1 and is distributed as x, the quilt in moment t
The label detected is distributed as y.For each block of pixels, can indicate are as follows: Xk(t-1)=xk,Xk(t)=yk, in formula, k is
The number of block, Xk(t-1)=xkBe meant that: label distribution of k-th of block of pixels in moment t-1 is x in target map X
In k-th of block of pixels label；Xk(t)=ykIt is meant that: label of k-th of block of pixels in moment t point in target map X
Cloth is the label of k-th of block of pixels in y.
V(t-1；T)=v: the motion vector of moment t-1 to t, all pixels block are estimated as v.In this case, often
The motion vector of a block can indicate are as follows: Vk=(t-1；T)=vk: moment t-1 to t, the motion vector of k-th of block of pixels are vk。
By ST-MRF model, previous moment image G (t-1)=g, present image G (t)=h and previous moment are given
Target map X (t-1)=x can determine target map X (t)=y and motion vector map V (t- with maximum a posteriori probability simultaneously
1；T)=V.Therefore, optimization aim map and the problem of motion vector map, have reformed into the energy function of minimum formula (1):
UstmrfIt is the energy function of ST-MRF model, can be minimized by relaxed algorithm, in formula: first part a (Nyk-
μNy)2It indicates in target map, the neighbouring relations of label, second part b (Mxyk-μMxy)2It indicates in successive objective map
The hiding relation of label, Part IIIIndicate the texture relationship in consecutive image related with motion vector, Part IVIndicate the neighbouring relations of the motion vector in motion vector map.
Nyk: refer to that the adjacent pixel blocks of a block of pixels and the block of pixels have the number of identical label；
Nxk: indicate the number of the adjacent pixel blocks of a block of pixels；
Dxyk: the texture correlation between the image at t-1 moment and the image of t moment is represented, when blocking, respectively
It calculates and belongs to the probability of each vehicle, enable block of pixels the coordinate value of time t is [x (t), y (t)], motion vector is [u (t), v
(t)], in the coordinate value of time t-1 are as follows: [x (t-1), y (t-1)]=[x (t)-u (t), y (t)-v (t)], when the figure of former frame
The vehicle label O as inMMotion vector beWhen:
G(t；α,β)；Pixel (α, β) is in the gray-scale intensity of time t, and each block of pixels is made of 8 × 8 pixels, so 0
≤dα≤8,0≤dβ≤8；
Mxyk: the number of pixels of shield portions in two block of pixels of partial occlusion；
μNy: neighborhood group, if using 8- neighborhood group, μ Ny=8 be maximum value；
Ck: current pixel block；
Bk: adjacent pixel blocks；
At (t-1) moment, the difference of the motion vector of current pixel block and adjacent pixel blocks；
A, b, c, f and μMxyFor the parameter of setting.It is preferred that experimental calibration parameter index: a=1/2, b=1/256, c=32/
1000000, f=1/4, μMxy=0.
ST-MRF model energy while segmentation object boundary and motion vector.Its process optimized is exactly according to previous moment
Target map, motion vector and present image, while considering that the texture in the motion vector and consecutive image in adjacent block closes
The similitude of system minimizes the target map at current time and the least energy of motion vector, specifically comprises the following steps:
01) motion vector that all pixels block is obtained by simple block matching method, then determines motion vector map
Original state V (t-1；T)=V0, V (t-1；T) moment t-1 to t, the motion vector of all pixels block are indicated；
02) according to the original state of motion vector map, target map is set by the candidate label of each block of pixels
Original state X (t)=y0.In this process, in single target area, single label of being selected is named to block of pixels, so
And more than one is selected label and gives block of pixels name around Ouluding boundary, elsewhere without candidate label to block of pixels
Name.
03) estimate that target map and motion vector map are carrying out X (t)=yi and V (t-1；T) total after=Vi iteration
Energy, i are the number of iterations；
04) in order to search the optimal result of target map and motion vector map, random conversion simultaneously is under current state
Target map X (t)=yi and motion vector map V (t-1；T) target map X (the t+1)=yi+1 of=Vi to NextState
With motion vector map V (t；T+1)=Vi+1；
05) in step 03) and 04) between iterative cycles operation, until X (t) and V (t-1；T) energy function value is converged to
UstmrfTo minimum.
Finally, the belongingness of the minimum decision block by energy function, energy function is smaller, CkBlock is more likely to belong to
Corresponding vehicle, according to the differentiation electric bicycle and car of different sizes of image block.
In vehicle label procedure, compare the average gray and threshold size of block on detection rectangle frame, if the gray scale of block of pixels
Greater than threshold value, illustrate with the presence of vehicle, and give vehicle label, in order to verify vehicle tracking algorithm, vehicle label tires out since 1
Meter.After vehicle enters detection rectangle frame, need to track vehicle.Track algorithm based on ST-MRF model is along time sequence
Column automatically update the shape of vehicle.For this update, algorithm estimates motion vector in the block of vehicle region.In each picture
On plain block, estimate each target pixel block in the motion vector of consecutive frame with the matching process of block.
The block of pixels on rectangle frame is detected it can be seen from Fig. 1 label flow chart to the detection calculating of vehicle and picture frame
Between the calculating of motion vector run over time.That is, detecting vehicle presence on rectangle frame when the t time is detected
Afterwards, the initial label of vehicle is obtained, then tracking module begins to calculate block of pixels corresponding to the consecutive frame with block matching method
Motion vector, and to detection rectangle frame on initial vehicle label shift, obtain by detection rectangle frame after sequence chart
The vehicle label of picture, as shown in Figure 3.
Image block-matching technique will test initial vehicle label obtained in module and come into the vehicle of detection zone
Label is moved, to convert vehicle label sequence for moving vehicle image sequence.The label sequence is entire video
Important Floor layer Technology in event detection technology provides technology platform to block the solution of problem, while label sequence is traffic
The extraction of parameter provides data source.
The 3 d video images that traffic camera is shot are converted into two-dimensional coordinate number by coordinates transformation method by step 2
According to.
Since the intersection obtained from video camera is indirect angle, it is difficult to obtain speed other than the volume of traffic, vehicle
Two-dimensional coordinate etc. information, thus it is more difficult calculate between the electric bicycle and car to make a dash across the red light reacting braking away from
From correct judgement cannot be provided to the severity of traffic conflict.In order to obtain actual running speed, coordinate value, need by
3-D image is converted into 2-D data coordinate, as shown in Fig. 2, specifically comprising the following steps:
01) global coordinates system (X is establishedG,YG), derived from video camera and the reference axis (X based on imageM,YM)；
02) it calculates and passes through intersection polygon detecting region (x1,y1), (x2,y2), (x3,y3) and (x4,y4) extended line
(p1,q1) and (p2,q2)；
03) rotational alignment W is derived1, with (x3,y3) it is dot, from (p2,q2) angle rotation-θ obtains W1；
04) straight line (p1,q1)→(x1,y1) extended line hand over W1For (x5,y5), point (x5,y5) and point (x3,y3) between appoint look for
A bit (xn,yn), then calculate point (xn,yn) arrive point (p1,q1) linear equation；
05) straight line (p is calculated2,q2)→(x3,y3), (p2,q2)→(x4,y4) intersection point (p2,q2)；
06) rotational alignment W similarly 03), is derived2, with (x3,y3) it is dot, from (p1,q1) rotation the angle θ obtain W2；
07) in straight line W2Take up an official post and looks for a bit (xm,ym), then calculate point (xm,ym) arrive point (p2,q2) linear equation；
08) straight line (p is calculated1,q1)→(x1,y1), (p1,q1)→(x3,y3) intersection point (p1,q1), straight line (p2,q2)→
(x3,y3), (p2,q2)→(x4,y4) intersection point (p2,q2)。
The stop of step 3, prediction electric bicycle and automobile in next frame image.
The research of most of conventional traffic conflicts is by employing observer to carry out conflict set recognition, and observer verifies in person to be worked as
Front direction have electric bicycle make a dash across the red light phenomenon when, whether the car of other direction normally travel has avlidance behavior, then right
Conflict after verifying is counted.However, this method depends on the judgement of people, accident and conflict cannot be fully considered
Severity.In addition, pervious judgement method for confliction detection only counts the electric bicycle to make a dash across the red light, without consider it is other because
Reaction braking distance between element, such as vehicle (electric bicycle and car), to reduce event detection precision.Cause
This, conflict spectrum is divided into 4 etc. in order to establish more accurate driver's conflict judgment criterion, according to reaction braking distance by the present invention
Grade: 1 grade, behavior of making a dash across the red light；It is 2 grades, slight to conflict；3 grades, danger conflict；4 grades, Serious conflicts or accident occurs, this result of study
Real-time, accurate, reliable traffic information can be provided for traffic management department, convenient for handling in time event.It carries out below
It is discussed in detail.
Single electric bicycle and automobile are tracked simultaneously using ST-MRF model, obtain single electric bicycle and vapour
After the speed and coordinate value of vehicle, the braking distance of each car can be computed.The calculating parameter of braking distance includes: to calculate
The top rake of speed, attachment coefficient and road.Wherein, the top rake of road can be by extending former frame (t-1) and present frame
(t) slope obtains；Coefficient of road adhesion refers to " highway technical standard ", can estimate vehicle next frame (t+ according to formula (2)
1) braking distance S:
In formula, ξ is speed；It is attachment coefficient, good bituminous pavement value is 1, good cement concrete pavement value
It is 0.7, wet road surface value is 0.5；ψ is road longitudinal grade degree, %, and upward slope is positive, and descending is negative.
Step 4 establishes driver's conflict judgment criterion according to the distance between stop.
Driver's conflict judgment criterion is divided into 4 grades of conflicts according to the braking distance of vehicle: 1 grade, behavior of making a dash across the red light；2 grades,
Slight conflict；3 grades, danger conflict；4 grades, Serious conflicts or accident occurs, criterion difference is as follows:
1 grade: behavior of making a dash across the red light
The conflict that intersection is solved using traditional method, when intersection or amber light, electric bicycle is driven
Member, which does not observe traffic rules and regulations, harum-scarum enters intersection, and such behavior is considered as the behavior of making a dash across the red light and is counted.Namely work as crossroad
When mouth is amber light, and the two-dimensional coordinate of electric bicycle is fallen into the coordinate range of intersection, then is determined as 1 grade of the row that makes a dash across the red light
For.
The two-dimensional coordinate that intersection image tracking method acquisition each car can be used is (related with time and signal) next
The number to make a dash across the red light is counted, each car passes in and out the number of intersection compared with signal time, time that then statistics is made a dash across the red light
Number.
2 grades: slight conflict
For example the electric bicycle of east-west direction traveling makes a dash across the red light into after intersection, the car of North and South direction traveling
Into intersection (North and South direction is green light at this time, and east-west direction is red light), conflict at this time can be electronic certainly by estimation
The distance between driving and the stop of car are judged.Firstly, by being obtained in vehicle to the vehicle tracking in image
Heart X, Y coordinate predict the braking distance of electric bicycle and car with formula 2；Then, estimation two cars are in next frame image
Stop, and the distance between find out two cars stop；Finally, marking the stop of two cars in two-dimensional coordinates
On, and be reflected on real road, to identify stop of the two cars on two-dimensional coordinate.If the stop of two cars
Extended line is non-intersecting, so that it may be judged as 2 grades of slight conflict.That is, the stop of two cars mutually disjoints, it is such
Conflict is considered as slightly conflicting；If the extended line of the stop of two cars intersects, two cars must be clashed, punching
Prominent degree of danger can be given according to 3 grades of conflicts and 4 grades of conflict judgment criterias and evaluate.
3 grades: danger conflict
3 grades of conflicts be when the distance between stop of two cars is less than safe distance and estimates two cars with
The case where track intersection of locus.In order to which the conflict judgment criterion studied under 3 grades of conflicts need to provide the judgement standard of quantization to slight conflict
Then, if whether the distance between two vehicle stops are in safe range.If the distance between two vehicle stops not in safe range and
It is considered as 3 grades of dangerous conflict when the motion profile intersection of two cars on coordinate when marking.
4 grades: Serious conflicts
On the basis of tracking the coordinate of vehicle movement track acquisition two cars, when the rectangular area part for representing two cars
When overlapping, being considered this conflict is traffic accident.Specifically, if the distance between stop of two cars is less than or equal to 0,
Then it is determined as 4 grades of Serious conflicts or accident occurs.
Experimental result and analysis:
The video of this experiment shooting is that morning 9:30 on November 14th, 2015 is handed at Huai'an Huai-Hai South Road and Liberation Road
One section of video of prong northing mouth, vehicle tracking technology through the invention can obtain the traffic parameters such as the volume of traffic and speed,
The data of acquisition are compared with the data of commercial Autoscope software measurement, as shown in table 1.As can be seen from the table
Difference measured by the volume of traffic, speed and the Autoscope software obtained under video environment between the volume of traffic, speed is little,
Illustrate that the error for the traffic data measured by the vehicle tracking technology is smaller, precision is relatively high.Therefore, vehicle movement track skill
Art can be used for traffic conflict detection and accident judgement between electric bicycle and car that intersection is made a dash across the red light.
1 video data of table and Autoscope data comparison
Traffic conflict test experience: on the basis of tracking vehicle movement track acquisition coordinate, with Huai-Hai South Road and liberation
For the intersection of road, traffic conflict detection is carried out using to captured video, when the electric bicycle and normal row to make a dash across the red light
For the car sailed when intersection undergoes traffic conflict, system passes through the braking distance of prediction electric bicycle and car,
It determines that two cars in the stop of next frame, and the distance between find out stop, driver's conflict is established according to this distance and is sentenced
Disconnected criterion determines conflict grade, and the results are shown in Table 2, and levels of conflict represents no electric bicycle for space and makes a dash across the red light.
2 traffic conflict testing result of table
| Time | Vehicle ID number | Vehicle center X-coordinate | Vehicle center Y coordinate | Levels of conflict | Speed (km/h) |
| 10:15 | 012 | 167.0 | 282.4 | 1 | 38.3 |
| 10:16 | 013 | 155.8 | 59.7 | 43.5 | |
| 10:17 | 014 | 227.2 | 97.0 | 3 | 32.5 |
| 10:18 | 016 | 163.3 | 288.4 | 1 | 38.3 |
| 10:19 | 017 | 145.6 | 60.4 | 43.4 | |
| 10:20 | 020 | 220.2 | 96.6 | 2 | 32.5 |
| 10:21 | 024 | 159.6 | 294.6 | 38.4 | |
| 10:22 | 028 | 134.5 | 60.4 | 3 | 43.4 |
| 10:23 | 030 | 211.3 | 95.0 | 1 | 32.6 |
| 10:24 | 031 | 155.3 | 301.4 | 2 | 38.5 |
| 10:25 | 032 | 122.8 | 60.9 | 43.5 | |
| 10:26 | 036 | 201.3 | 95.0 | 1 | 32.5 |
| 10:27 | 038 | 152.2 | 307.0 | 2 | 38.3 |
| 10:28 | 040 | 110.7 | 61.3 | 4 | 43.4 |
| 10:29 | 044 | 190.9 | 93.9 | 32.5 |
Conclude that record there are 10 electric bicycles to make a dash across the red light in level-one consequences of hostilities by experiment, in 10 vehicles,
6 enter second level conflict, and 3 are recorded in three-level conflict, finally have 1 vehicle to reach level Four levels of conflict, i.e., traffic will occur
Accident.Fig. 4,5,6 show the traffic conflict situation that tracking system detects, wherein Fig. 4 shows that 2 grades of conflict situations, ID number are
27 electric bicycle violates traffic signals but without other vehicle；Fig. 5 shows 3 grades of conflict situations, this conflict occurs
Between the car that the electric bicycle and ID number that ID number is 27 are 28, such situation, the electric bicycle that ID number is 27 is violated
Traffic signals mark the stop of the two cars on coordinate to intersect, but the rectangular area for representing two cars do not send out
Raw overlapping；Finally, Fig. 6 shows level Four conflict situations, this conflict occur electric bicycle and the ID number that ID number is 11 be 9 it is small
Between automobile, the rectangular area for representing two cars at this time has partly overlapped, it is believed that traffic accident will occur for this conflict.
The beneficial effects of the present invention are:
(1) it proposes a kind of ST-MRF model based on optimization while tracking the algorithm of electric bicycle and automobile.The party
Method can not only optimize the spatial distribution of image, moreover it is possible to optimize distribution of the image sequence along time shaft, improve pervious vehicle with
Track algorithm is in the case where the intersection vehicles serious shielding of traffic disturbance the shortcomings that poor robustness.
(2) traffic conflict detection method of the invention is suitable for the intersection of various geometries, is rotated by coordinate
3-D image is converted into 2-D data by algorithm, and (electric bicycle is small for the vehicle in different motion direction at calculating intersection
Automobile) motion profile and two-dimensional coordinate, and experimental data is compared with data measured by commercialization Autoscope software,
It obtains and uses the error of technology traffic data obtained smaller, precision is relatively high.
(3) conflicted judgment criterion according to driver, establish the levels of conflict of 4 grades: 1 grade, behavior of making a dash across the red light；2 grades, slightly
Conflict；3 grades, danger conflict；4 grades, Serious conflicts (accident), and level Four conflict is detected by Success in Experiment.
(4) when the electric bicycle to make a dash across the red light, which occurs 3 grades with car, to conflict, this method energy success prediction is potentially handed over
Interpreter's event, can provide real-time, accurate, reliable traffic information, avoid traffic accident generation for traffic management department.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure made by bright specification and accompanying drawing content perhaps equivalent process transformation or be directly or indirectly used in other correlation
Technical field, be included within the scope of the present invention.
Claims (4)
1. electric bicycle and mechanical transport collision detection method based on ST-MRF model, which is characterized in that including walking as follows
It is rapid:
Step 1, traffic camera shoot 3 d video images, tracked simultaneously using ST-MRF model single electric bicycle with
Automobile obtains the traffic information of single electric bicycle and automobile；
The 3 d video images that traffic camera is shot are converted into two-dimensional coordinate data by coordinates transformation method by step 2；
The stop of step 3, prediction electric bicycle and automobile in next frame image；
Step 4 establishes driver's conflict judgment criterion according to the distance between stop；
In the step 1, the energy function U of ST-MRF modelstmrfIt is:
In formula: first part a (Nyk-μNy)2It indicates in target map, the neighbouring relations of label, second part b (Mxyk-μMxy)2
Indicate the hiding relation of the label in successive objective map, Part IIIIndicate sequential chart related with motion vector
Texture relationship as in, Part IVIndicate the movement arrow in motion vector map
The neighbouring relations of amount；
Nyk: refer to that the adjacent pixel blocks of a block of pixels and the block of pixels have the number of identical label；
Nxk: indicate the number of the adjacent pixel blocks of a block of pixels；
Dxyk: the texture correlation between the image at t-1 moment and the image of t moment is represented, when blocking, is calculated separately
Belong to the probability of each vehicle；
Mxyk: the number of pixels of shield portions in two block of pixels of partial occlusion；
μNy: neighborhood group, if using 8- neighborhood group, μ Ny=8 be maximum value；
Ck: current pixel block；
Bk: adjacent pixel blocks；
At (t-1) moment, the difference of the motion vector of current pixel block and adjacent pixel blocks；
A, b, c, f and μMxyFor the parameter of setting；
According to the target map, motion vector and present image of previous moment, while considering the movement arrow in adjacent pixel blocks
It measures with the similitude of the texture relationship in consecutive image and minimizes the target map at current time and the minimum energy of motion vector
Amount, specifically comprises the following steps:
01) motion vector that all pixels block is obtained by block matching method, determines the original state V (t- of motion vector map
1；T)=V0, V (t-1；T) moment t-1 to t, the motion vector of each block of pixels are indicated；
02) according to the original state of motion vector map, the initial of target map is set by the candidate label of each block of pixels
State X (t)=y0；
03) estimate that target map and motion vector map are carrying out X (t)=yi and V (t-1；T) total energy after=Vi iteration
Amount, i is the number of iterations；
04) random target map X (t)=yi and motion vector map V (t-1 of the conversion under current state simultaneously；T)=Vi is arrived
Target map X (t+1)=yi+1 and motion vector map the V (t of NextState；T+1)=Vi+1；
05) in step 03) and 04) between iterative cycles operation, until X (t) and V (t-1；T) energy function value U is converged tostmrf
To minimum.
2. the electric bicycle according to claim 1 based on ST-MRF model and mechanical transport collision detection method,
It is characterized in that, in step 3, the braking distance S of electric bicycle and automobile in next frame image are as follows:
In formula, ξ is speed；It is attachment coefficient；ψ is road longitudinal grade degree, %, and upward slope is positive, and descending is negative.
3. the electric bicycle according to claim 2 based on ST-MRF model and mechanical transport collision detection method,
It is characterized in that, in step 4, driver's judgment criterion that conflicts is divided into 4 grades of conflicts: 1 grade, behavior of making a dash across the red light；It is 2 grades, slight to conflict；3
Grade, danger conflict；4 grades, Serious conflicts or accident occurs, criterion difference is as follows:
When intersection is amber light, and the two-dimensional coordinate of electric bicycle is fallen into the coordinate range of intersection, then is determined as 1
The behavior of making a dash across the red light of grade；
When intersection is red light, firstly, by the electric bicycle and car tracing acquisition vehicle center X, Y in image
Coordinate, with the braking distance of formula (2) prediction electric bicycle and car；Then, estimation two cars are in next frame image
The distance between stop, and find out two cars stop；Finally, on a two-dimensional coordinate axis by the stop label of two cars,
And it is mapped on real road:
If the extended line of the stop of two cars is non-intersecting, it is determined as 2 grades of slight conflict；
If the distance between stop of two cars is not in safe range and when the movement for marking the two cars on coordinate
When intersection of locus, then it is determined as 3 grades of dangerous conflict；
If the distance between stop of two cars is less than or equal to 0, it is determined as 4 grades of Serious conflicts or accident occurs.
4. the electric bicycle according to claim 1 based on ST-MRF model and mechanical transport collision detection method,
It is characterized in that, a=1/2, b=1/256, c=32/1000000, f=1/4, μMxy=0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611030277.4A CN106530825B (en) | 2016-11-16 | 2016-11-16 | Electric bicycle and mechanical transport collision detection method based on ST-MRF model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611030277.4A CN106530825B (en) | 2016-11-16 | 2016-11-16 | Electric bicycle and mechanical transport collision detection method based on ST-MRF model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106530825A CN106530825A (en) | 2017-03-22 |
| CN106530825B true CN106530825B (en) | 2019-01-22 |
Family
ID=58356105
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611030277.4A Active CN106530825B (en) | 2016-11-16 | 2016-11-16 | Electric bicycle and mechanical transport collision detection method based on ST-MRF model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106530825B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6747389B2 (en) * | 2017-06-29 | 2020-08-26 | 株式会社デンソー | Collision estimating device and collision estimating method |
| CN107590462B (en) * | 2017-09-13 | 2020-11-20 | 淮阴工学院 | Traffic flow parameter extraction method based on combination of pattern recognition and ST-MRF |
| CN107808549B (en) * | 2017-11-08 | 2020-06-16 | 华为技术有限公司 | Method, device and system for mobile control |
| CN112368757B (en) * | 2018-07-04 | 2022-11-01 | 日产自动车株式会社 | Driving assistance method and driving assistance device |
| CN111243274A (en) * | 2020-01-20 | 2020-06-05 | 陈俊言 | Road collision early warning system and method for non-internet traffic individuals |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4311426B2 (en) * | 2006-09-12 | 2009-08-12 | 住友電気工業株式会社 | Display system, in-vehicle device, and display method for displaying moving object |
| US9058524B2 (en) * | 2013-06-28 | 2015-06-16 | Bendix Commercial Vehicle Systems Llc | Measuring the range to an object, in an image, using size categorization |
| CN103942359B (en) * | 2014-03-13 | 2017-01-04 | 华南理工大学 | The modeling method of unsignalized intersection bicycle conflict avoiding behavior simulation model |
| CN104050837A (en) * | 2014-06-10 | 2014-09-17 | 奇瑞汽车股份有限公司 | Driving safety prompting method and device |
| CN104376725B (en) * | 2014-11-28 | 2017-02-01 | 东南大学 | Signalized intersection non-motor vehicle lane channelizing control method under expansion effect |
| CN105761547A (en) * | 2016-03-28 | 2016-07-13 | 安徽云森物联网科技有限公司 | Traffic collision pre-warning technique and system based on images |
-
2016
- 2016-11-16 CN CN201611030277.4A patent/CN106530825B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106530825A (en) | 2017-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106530825B (en) | Electric bicycle and mechanical transport collision detection method based on ST-MRF model | |
| CN103593678B (en) | A kind of long-span bridge vehicle dynamic load distribution detection method | |
| US9292750B2 (en) | Method and apparatus for detecting traffic monitoring video | |
| WO2022206942A1 (en) | Laser radar point cloud dynamic segmentation and fusion method based on driving safety risk field | |
| CN104106105B (en) | Stereoscopic article detection device | |
| CN108596129A (en) | A kind of vehicle based on intelligent video analysis technology gets over line detecting method | |
| CN103522970B (en) | Based on vehicle driving safety method of inspection and the system of machine vision | |
| CN113156421A (en) | Obstacle detection method based on information fusion of millimeter wave radar and camera | |
| Zhangyu et al. | A camera and LiDAR data fusion method for railway object detection | |
| CN106951879A (en) | Multiple features fusion vehicle checking method based on camera and millimetre-wave radar | |
| CN103400157B (en) | Road pedestrian and non-motor vehicle detection method based on video analysis | |
| CN104115188B (en) | Stereoscopic article detection means | |
| Chiu et al. | Automatic Traffic Surveillance System for Vision-Based Vehicle Recognition and Tracking. | |
| CN103136537B (en) | Vehicle type identification method based on support vector machine | |
| Feng et al. | Mixed road user trajectory extraction from moving aerial videos based on convolution neural network detection | |
| CN109791607A (en) | It is detected from a series of images of video camera by homography matrix and identifying object | |
| CN110379168A (en) | A kind of vehicular traffic information acquisition method based on Mask R-CNN | |
| CN110298300A (en) | A method of detection vehicle violation crimping | |
| CN107025657A (en) | A kind of vehicle action trail detection method based on video image | |
| CN106156752A (en) | A kind of model recognizing method based on inverse projection three-view diagram | |
| CN104115204B (en) | Stereoscopic article detection device | |
| CN103794050A (en) | Real-time transport vehicle detecting and tracking method | |
| Cheng et al. | Modeling mixed traffic in shared space using lstm with probability density mapping | |
| Jwa et al. | Kalman filter based railway tracking from mobile Lidar data | |
| CN111429756A (en) | Highway tunnel rear-end collision prevention early warning method based on video technology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 223005 8th Floor, Andong Building, No. 10 Haian Road, Lianshui County, Huaian City, Jiangsu Province Applicant after: Huaijin Polytechnical College Address before: 223005 Jiangsu Huaian Higher Education Park 1 East Road Applicant before: Huaijin Polytechnical College |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200831 Address after: No.12, Keji Road, higher education park, Huaian Economic and Technological Development Zone, Jiangsu Province Patentee after: JIANGSU TS TRAFFIC DESIGN & RESEARCH INSTITUTE Co.,Ltd. Address before: 223005 8th Floor, Andong Building, No. 10 Haian Road, Lianshui County, Huaian City, Jiangsu Province Patentee before: HUAIYIN INSTITUTE OF TECHNOLOGY |
|
| TR01 | Transfer of patent right | ||
| CP02 | Change in the address of a patent holder |
Address after: 210019 No. 8 East Street, Fuchunjiang, Jianye District, Nanjing City, Jiangsu Province Patentee after: JIANGSU TS TRAFFIC DESIGN & RESEARCH INSTITUTE Co.,Ltd. Address before: 223243 No.12, Keji Road, higher education park, Huaian Economic and Technological Development Zone, Jiangsu Province Patentee before: JIANGSU TS TRAFFIC DESIGN & RESEARCH INSTITUTE Co.,Ltd. |
|
| CP02 | Change in the address of a patent holder |