CN102930242B - Bus type identifying method - Google Patents

Abstract

The invention discloses a bus type identifying method, belonging to the technical field of processing a computer video. The bus type identifying method comprises the following steps of: carrying out mixed Gaussian modeling on a monitoring video, and carrying out bus type initial identification on a vehicle; secondly, establishing a corresponding 3D (three-dimensional) model according to the position information of the vehicle and the characteristics of a bus; meanwhile, extracting a characteristic segment by using an LSD (Large Screen Display) segment extracting algorithm; and finally, matching the 3D model of the vehicle and the segment characteristics by using a comprehensive algorithm of combining template matching and shortest distance matching. According to the bus type identifying method provided by the invention, the 3D model is established by using the vehicle position information and the bus type characteristics, so that the preparation work of a car type database can be avoided; meanwhile, the vehicle to be judged is initially judged by using the bus type characteristics, so that the 3D modeling and matching process can be carried out on all vehicles in the scene, and the calculation amount is reduced. Finally, the calculation accuracy is improved, and the calculation amount can be reduced by using a combined matching algorithm.

Description

A kind of bus model recognizing method

Technical field

The invention belongs to computer video processing technology field, be specially a kind of bus model recognizing method, especially design a kind of bus model recognizing method being suitable for public security bayonet socket monitoring application.

Background technology

At present, vehicle cab recognition technology plays more and more important effect in public security monitoring, and bus is as a kind of important urban transportation tool, the key monitoring object especially in monitoring.

Feature Points Matching method and 3D model matching method are two kinds of common vehicle detection methods.But Feature Points Matching method needs whole 2D features of extraction usually (as edge line segment, edge pixel point etc.) to carry out with the 2D feature of model mating (see: Grimson, W., " The combinatorics of heuristic searchtermination for object recognition in cluttered environment; " IEEE Trans.PAMI., vol.13, no.9, pp.920-935,1991.) thus its calculated amount is larger, real-time is relatively poor, can not directly apply to bayonet socket real-time monitoring system.The method of traditional 3D coupling (see: (and see: Tan, T.N., Sullivan, G.D., Baker, K.D., " Model-based localization and recognition of road vehicles, " Int.J.Comput.Vis., vol.27, no.1, pp.5-25,1998.) its recognition accuracy depends on the complete degree of 3D model bank, but major applications equipment is difficult to have complete 3D auto model database, and method complexity linearly can increase with the increase of model quantity, is therefore also difficult to directly apply on bayonet socket watch-dog.

Find through retrieval; publication number is the Chinese invention patent of 101783076A; method for quick vehicle type recognition under a kind of video monitoring mode of this disclosure of the invention; implement according to following steps: road monitoring apparatus is set; and vehicle is divided into car, with the taxi of particular color mark, minibus, in-between car, bus and high capacity waggon; step 1, initialization, carry out training study to video monitoring apparatus; Step 2, extract the area in vehicle target region and the length of boundary rectangle thereof and wide, structure individual features, and be compact car, in-between car, large car by vehicle target rough sort; Step 3, the dominant hue feature multiple target of compact car being extracted respectively to vehicle body identify taxi, then extract compact car vehicle window relative seat feature parameter, determine minibus or car further; Step 4, extraction roof brightness parameter and roof textural characteristics parameter, determine whether large car is bus.This patent when whether judgement is bus adopts roof brightness and roof texture as principal character, is thus subject to the impact of illumination variation; And the present invention is using the shape of bus and LSD Eigenvector as principal character, the impact of change on method of thus illumination is less, makes the robustness of method better, is conducive to improving the accuracy identified.

Summary of the invention

The object of the invention is to overcome above-mentioned the deficiencies in the prior art part, a kind of new bus model recognizing method is proposed, the method is based on 3D Model Matching and LSD Eigenvector extraction algorithm, avoid the preliminary work in model data storehouse and 3D modeling and matching process are carried out to vehicles all in scene, decrease calculated amount, improve computing accuracy rate and reduce calculated amount simultaneously.

For achieving the above object, the technical solution used in the present invention is: first carry out Gaussian modeling to monitor video, obtains pending vehicle foreground image; Then carry out bus vehicle to vehicle tentatively to identify; Secondly according to the positional information of vehicle and the feature of bus vehicle, corresponding 3D model is set up; Use LSD line segments extraction algorithm to extract the Eigenvector of vehicle simultaneously; Finally adopt a kind of integration algorithm in conjunction with template matches and bee-line coupling, the 3D model of vehicle and line segment feature are mated.Confirm through experiment, the present invention has higher accuracy for bus vehicle cab recognition, and real-time meets bayonet socket watch-dog demand.

The inventive method specifically comprises the following steps:

The first step: carry out mixed Gaussian background modeling to monitor video, obtains pending vehicle foreground image.

Second step: under world coordinate system, carries out bus vehicle to obtained vehicle and tentatively identifies.

Concrete steps are:

1. contours extract is carried out to the vehicle foreground image obtained in the first step, obtain N number of connected region Ω _k, k=1,2 ..., N, to each connected region Ω _k, one can be obtained and comprise Ω _kand the rectangular area R that area is minimum _k, k=1,2 ..., N.

2. under world coordinate system, by rectangle R _k2 p in bottom ₁, p ₂structure judging point p _m, specifically:

$\left\{\begin{array}{c}{p}_m.x=(p_1.x+p_2.x)/2\\ p_m.y=(p_1.y+p_2.y)/2-l\\ p_m.z=p_1.z\end{array}\right.$

Wherein, l represents bus length of wagon, R _kfor comprising the minimum rectangle frame of the whole profile of vehicle to be identified, (x, y, z) represents the three-dimensional coordinate of each point in world coordinate system.

3. under image coordinate system, according to p _mand R _kposition relationship tentatively judge whether this vehicle is bus, if p _m∈ R _k, then tentatively judge that this vehicle is bus, carry out the 3rd step; If then judge that this vehicle is non-bus, judge to terminate.

3rd step: under world coordinate system, according to bus vehicle feature and R _kpositional information, builds the 3D model of Current vehicle.

Concrete steps are:

1. R _k2, bottom p ₁, p ₂as 3D model front bottom end two point.

2. by p ₁, p ₂build bus 3D model front top 2 p ₃, p ₄, specifically:

$\left\{\begin{array}{c}{p}_3.x=p_1.x,p_3.y=p_1.y,p_3.z=p_1.z+h\\ p_4.x=p_2.x,p_4.y=p_2.y,p_4.z=p_2.z+h\end{array}\right.$

Wherein, h represents bus height, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system; .

3. according to the perspective view of bus on xOy, 2 p bottom 3D model tail end are built ₅, p ₆, specifically:

$\left\{\begin{array}{c}\mathrm{\θ}=\arctan \left(\frac{p_1.y-p_2.y}{p_2.x-p_1.x}\right)\\ p_j.x=p_i.x-l\sin \mathrm{\θ}\\ p_j.y=p_i.x-l\cos \mathrm{\θ}\end{array}\right.$

Wherein, (i, j) ∈ { (1,5), (2,6) }, θ represents in perspective view, line segment (p ₁, p ₂) with the angle of y-axis, l represents bus length of wagon, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system.

4. by p ₅, p ₆build 3D model tail end 2, top p ₇, p ₈, be specially:

$\left\{\begin{array}{c}{p}_7.x=p_5.x,p_7.y=p_5.y,p_7.z=p_5.z+h\\ p_8.x=p_6.x,p_8.y=p_6.y,p_8.z=p_6.z+h\end{array}\right.$

Wherein, h represents bus height, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system.。

5. obtain 12 line segment aggregates by 3D model 8 end points, wherein can judge according to the visual angle of bayonet socket camera and the position of end points: line segment (p ₁, p ₂) (p ₁, p ₃) (p ₂, p ₄) (p ₃, p ₄) (p ₃, p ₇) (p ₄, p ₈) be camera visible segment, line segment (p ₅, p ₆) be the invisible line segment of camera, line segment (p ₂, p ₆) (p ₆, p ₈) then need according to p ₆position judge, specifically: under image coordinate system, if then (p ₂, p ₆) (p ₆, p ₈) be camera visible segment; If p ₆∈ R (p ₁₂₃₄), then it is invisible line segment.R (p ₁₂₃₄) represent by a p ₁, p ₂, p ₃, p ₄the rectangle formed.In like manner, to line segment (p ₁, p ₅) (p ₅, p ₇) judge.Select camera visible segment as the line segment aggregate of 3D model.So far the 3D model of this vehicle can be obtained.

4th step: adopt LSD line segments extraction algorithm to carry out Eigenvector extraction to the vehicle obtained in second step.

5th step: the method adopting stencil matching method and knearest neighbour method to combine carries out matching primitives to the Eigenvector obtained in the 3D model obtained in the 3rd step and the 4th step, obtains recognition result.

Concrete steps are:

1. adopt template matching method, calculate matching factor η ₁, specifically:

Wherein, represent the overlapping region between 3D model gray-scale map and Eigenvector gray-scale map; ψ is 3D model gray-scale map, and its pixel value is 0 or 1; for LSD Eigenvector gray-scale map; represent and morphological dilations process is carried out to image; ∑ I represents and carries out pixel summation to gray level image I; represent and threshold process is done to gray-scale map I, specifically:

${\mathrm{Threshold}}_t^v\left(I_{x,y}\right)=\left\{\begin{array}{cc}v,& I_{x,y}\≥v\\ 0,& \mathrm{otherwise}\end{array}\right.$

2. according to η ₁judge whether Current vehicle is bus, specifically: if η ₁>TH _l, TH _lfor setting threshold value, representation feature line segment overlaps with 3D model large percentage, thinks that this vehicle may be bus, enters next step, and employing knearest neighbour method judges further; If η ₁>TH _h, TH _hfor setting threshold value and TH _h>TH _l, then think that this vehicle is bus, judge to terminate; If η ₁≤ TH _l, then think that this vehicle is non-bus, judge to terminate.

3. bee-line algorithm is implemented to the vehicle needing in previous step to judge further, order represent removing overlapping region 3D illustraton of model; represent removing overlapping region eigenvector figure; I (α) represents the gray-scale value of pixel α; P (α) represents the position of pixel α under image coordinate system, initialization counter sum=0.Then minimum distance method is specifically implemented as follows:

(1) to pixel α ∈ ψ _hand I (α) ≠ 0, centered by p (α), step is that the length of side sets up square aearch window.

(2) calculate pixel α to arrive bee-line, be specially

(3) if d (α)≤d _tH, then sum=sum+1 and I (β is made ₀)=0, d _tHfor setting threshold value.

(4) step (1) is repeated until ψ _hin pixel travel through completely.

4. matching factor η is calculated ₂, be specially:

${\mathrm{\η}}_2=\frac{\mathrm{sum}}{\mathrm{\Σ}{\mathrm{\ψ}}_h}$

According to η ₂judge whether Current vehicle is bus, specifically: if η ₂>TH _d, TH _dfor setting threshold value, representation feature line segment overlaps with 3D model large percentage, judges that this vehicle is bus; Otherwise, then judge that this vehicle is non-bus, judge to terminate.

Compared with prior art, main contributions of the present invention and feature are: 1) utilize vehicle position information and bus vehicle feature to build 3D model, thus do not need complete model data storehouse; 2) utilize bus vehicle feature to treat and judge that vehicle tentatively judges, to avoid carrying out 3D modeling and matching process to vehicles all in scene, decrease calculated amount; 3) a kind of matching algorithm of combination is adopted, template matching method is first adopted to mate for the first time 3D model and Eigenvector, knearest neighbour method only need carry out matching operation to residue 3D model pixel point and Eigenvector pixel after first coupling, improves computing accuracy rate and reduces calculated amount simultaneously.The present invention is particularly useful for the bus vehicle cab recognition analysis of public security bayonet socket application.

Accompanying drawing explanation

By reading the detailed description done non-limiting example with reference to the following drawings, other features, objects and advantages of the present invention will become more obvious:

Fig. 1 is bus model recognizing method overall block flow diagram of the present invention.

Fig. 2 is schematic diagram in the embodiment of the present invention, and wherein (a) tentatively identifies signal for bus, and (b) (c) is bus perspective view on xOy face.

Fig. 3 is 3D model 8 end points process of establishing schematic diagram in the embodiment of the present invention.

Fig. 4 is visible segment schematic diagram in the embodiment of the present invention, wherein: bus visible segment under (a) camera angles; B () is the invisible line segment of bus and possibility visible segment schematic diagram.

Fig. 5 (a) (b) is LSD Eigenvector extraction schematic diagram.

Fig. 6 is identifying schematic diagram in the embodiment of the present invention: (a) judges vehicle for waiting, b () is Eigenvector gray-scale map, c () is 3D model gray-scale map, d () represents in stencil matching method, 3D model gray-scale map and Eigenvector gray-scale map stack result, e () represents 3D illustraton of model and Eigenvector figure overlapping region, (f) is 3D illustraton of model removing overlapping region schematic diagram.

Fig. 7 is embodiment of the present invention bayonet socket monitoring scene schematic diagram, and camera and horizontal plane angle are less than 45 °.

Embodiment

Below in conjunction with specific embodiment, the present invention is described in detail.Following examples will contribute to those skilled in the art and understand the present invention further, but not limit the present invention in any form.It should be pointed out that to those skilled in the art, without departing from the inventive concept of the premise, some distortion and improvement can also be made.These all belong to protection scope of the present invention.

Embodiment

The video sequence that the present embodiment adopts is public security bayonet socket monitoring scene sequence.

The bus model recognizing method that the present embodiment relates to, comprises following concrete steps:

The first step: carry out mixed Gaussian background modeling to video sequence, obtains pending vehicle foreground image.

Second step: under world coordinate system, carries out bus vehicle to obtained vehicle and tentatively identifies.

Concrete steps are:

1. pair vehicle foreground image carries out contours extract, obtains N number of connected region Ω _k, k=1,2 ..., N, to each connected region Ω _k, one can be obtained and comprise Ω _kand the rectangular area R that area is minimum _k, k=1,2 ..., N.

2. under world coordinate system, by rectangle R _k2 p in bottom ₁, p ₂structure judging point p _m, specifically:

$\left\{\begin{array}{c}{p}_m.x=(p_1.x+p_2.x)/2\\ p_m.y=(p_1.y+p_2.y)/2-l\\ p_m.z=p_1.z\end{array}\right.$

Wherein, l represents bus length of wagon, l=10 in the present embodiment, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system.

3. under image coordinate system, according to p _mand R _kposition relationship tentatively judge vehicle vehicle, if p _m∈ R _k, then tentatively judge that this vehicle is bus, carry out the 3rd step; If then judge that this vehicle is non-bus, judge to terminate.As shown in Figure 2 (a) shows, in figure, rectangle frame is R _k, by 2 p bottom it ₁, p ₂judging point p can be constructed _m, because bus vehicle body is longer, after under coordinate conversion to image coordinate system, judging point p _mmeet p _m∈ R _k.

3rd step: under world coordinate system, is approximately rectangular parallelepiped and R according to bus vehicle _kpositional information, builds the 3D rectangular parallelepiped model of Current vehicle.

Concrete steps are:

1.R _k2, bottom p ₁, p ₂as 3D model front bottom end two point.

2. be approximately square by bus headstock, can according to p ₁, p ₂build bus 3D model front top 2 p ₃, p ₄, specifically:

$\left\{\begin{array}{c}{p}_3.x=p_1.x,p_3.y=p_1.y,p_3.z=p_1.z+h\\ p_4.x=p_2.x,p_4.y=p_2.y,p_4.z=p_2.z+h\end{array}\right.$

Wherein, h represents bus height, h=2.5 in the present embodiment, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system.

3., according to the perspective view of bus on xOy, as shown in Fig. 2 (b) (c), 2 p bottom 3D model tail end can be built by geometric relationship ₅, p ₆, specifically:

$\left\{\begin{array}{c}\mathrm{\θ}=\arctan \left(\frac{p_1.y-p_2.y}{p_2.x-p_1.x}\right)\\ p_j.x=p_i.x-l\sin \mathrm{\θ}\\ p_j.y=p_i.x-l\cos \mathrm{\θ}\end{array}\right.$

Wherein, (i, j) ∈ { (1,5), (2,6) }, θ represents in perspective view, line segment (p ₁, p ₂) with the angle of y-axis, as shown in Figure 2 (c).L represents bus length of wagon, l=10 in the present embodiment, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system.

4. by p ₅, p ₆build 3D model tail end 2, top p ₇, p ₈, be specially:

$\left\{\begin{array}{c}{p}_7.x=p_5.x,p_7.y=p_5.y,p_7.z=p_5.z+h\\ p_8.x=p_6.x,p_8.y=p_6.y,p_8.z=p_6.z+h\end{array}\right.$

Wherein, h represents bus height, h=2.5 in the present embodiment, and (x, y, z) represents each point three-dimensional coordinate in world coordinate system.

Fig. 3 is the building process schematic diagram of above-mentioned 8 end points.

5. can obtain 12 line segment aggregates by 3D model 8 end points, but in camera perspective, only have part line segment visible, thus need to select camera visible segment as model line segment aggregate.Selection mode is as follows: as shown in Figure 4 (a), line segment (p ₁, p ₂) (p ₁, p ₃) (p ₂, p ₄) (p ₃, p ₄) (p ₃, p ₇) (p ₄, p ₈) be camera visible segment; As shown in Figure 4 (b), line segment (p ₅, p ₆) be the invisible line segment of camera, line segment (p ₂, p ₆) (p ₆, p ₈) then need according to p ₆position judge, specifically: under image coordinate system, if then (p ₂, p ₆) (p ₆, p ₈) be camera visible segment; If p ₆∈ R (p ₁₂₃₄), then it is invisible line segment.R (p ₁₂₃₄) represent by a p ₁, p ₂, p ₃, p ₄the rectangle formed.In like manner, to line segment (p ₁, p ₅) (p ₅, p ₇) judge.Select camera visible segment as the line segment aggregate of 3D model.So far the 3D model of this vehicle can be obtained.

4th step: adopt LSD line segments extraction algorithm to carry out Eigenvector extraction to the vehicle obtained in second step, Fig. 5 is LSD extraction algorithm effect schematic diagram.

5th step: matching primitives is carried out to the Eigenvector obtained in the 3D model obtained in the 3rd step and the 4th step, specifically:

1. adopt template matching method, calculate matching factor η ₁, specifically:

Wherein, represent the overlapping region between 3D model gray-scale map and Eigenvector gray-scale map, as shown in Figure 6 (e); ψ is 3D model gray-scale map, and its pixel value is 0 or 1, as shown in Figure 6 (c); for LSD Eigenvector gray-scale map, as shown in Figure 6 (b); represent and morphological dilations process is carried out, as shown in Fig. 6 (d) to image; ∑ I represents and carries out pixel summation to gray level image I; represent and threshold process is done to gray-scale map I, specifically:

${\mathrm{Threshold}}_t^v\left(I_{x,y}\right)=\left\{\begin{array}{cc}v,& I_{x,y}\≥v\\ 0,& \mathrm{otherwise}\end{array}\right.$

2. according to η ₁judge whether Current vehicle is bus, specifically: if η ₁>TH _l, TH _lfor setting threshold value, then representation feature line segment overlaps with 3D model large percentage, thinks that this vehicle may be bus, enters next step, and employing knearest neighbour method judges further; If η ₁>TH _h, TH _hfor setting threshold value, then think that this vehicle is bus, judge to terminate; If η ₁≤ TH _l, then think that this vehicle is non-bus, judge to terminate.TH in the present embodiment _l=0.7, TH _h=0.8.

3. in pair previous step, need the vehicle judged further to implement bee-line algorithm, order represent removing overlapping region 3D illustraton of model, as shown in Fig. 6 (f); represent removing overlapping region eigenvector figure; I (α) represents the gray-scale value of pixel α; P (α) represents the position of pixel α under image coordinate system, initialization counter sum=0.Then minimum distance method is specifically implemented as follows:

(1) to pixel α ∈ ψ _hand I (α) ≠ 0, centered by p (α), set up the square aearch window that window size is 3 × 3.

(2) calculate pixel α to arrive bee-line, be specially

(3) if d (α)≤d _tH, then sum=sum+1 and I (β ₀)=0, the present embodiment threshold value d _tH=1.5.

(4) step (1) is repeated until ψ _hin pixel travel through completely.

4. calculate matching factor η ₂, be specially:

${\mathrm{\η}}_2=\frac{\mathrm{sum}}{\mathrm{\Σ}{\mathrm{\ψ}}_h}$

According to η ₂judge whether Current vehicle is bus, specifically: if η ₂>TH _d, TH _dfor setting threshold value, representation feature line segment overlaps with 3D model large percentage, judges that this vehicle is bus; Otherwise, then judge that this vehicle is non-bus, judge to terminate.TH in the present embodiment _d=0.9.

The present embodiment is tested the video sequence comprising 100 buses and 150 other types vehicles, and test result is as shown in table 1.

Table 1. bus vehicle cab recognition result

Above specific embodiments of the invention are described.It is to be appreciated that the present invention is not limited to above-mentioned particular implementation, those skilled in the art can make various distortion or amendment within the scope of the claims, and this does not affect flesh and blood of the present invention.

Claims (3)

1. a bus model recognizing method, is characterized in that, comprises the following steps:

The first step: carry out mixed Gaussian background modeling to monitor video, obtains pending vehicle foreground image;

Second step: under world coordinate system, carries out bus vehicle to obtained vehicle and tentatively identifies; Concrete steps are:

1. contours extract is carried out to the vehicle foreground image obtained in the first step, obtain N number of connected region Ω _k, k=1,2 ..., N, to each connected region Ω _k, obtain one and comprise Ω _kand the rectangular area R that area is minimum _k, k=1,2 ..., N;

2. under world coordinate system, by rectangle R _k2 p in bottom ₁, p ₂structure judging point p _m:

$\left\{\begin{array}{c}{p}_m.x=(p_1.x+p_2.x)/2\\ p_m.y=(p_1.y+p_2.y)/2-l\\ p_m.z=p_1.z\end{array}\right.$

Wherein, l represents bus length of wagon; (x, y, z) represents the three-dimensional coordinate of each point in world coordinate system;

3. under image coordinate system, according to p _mand R _kposition relationship tentatively judge whether this vehicle is bus, if p _m∈ R _k, then tentatively judge that this vehicle is bus, carry out the 3rd step; If then judge that this vehicle is non-bus, judge to terminate;

3rd step: under world coordinate system, according to bus vehicle feature and R _kpositional information, builds the 3D model of Current vehicle;

4th step, adopts LSD line segments extraction algorithm to carry out Eigenvector extraction to the vehicle obtained in second step;

5th step: the method adopting stencil matching method and knearest neighbour method to combine carries out matching primitives to the Eigenvector obtained in the 3D model obtained in the 3rd step and the 4th step, obtains recognition result;

3D model described in 3rd step, the method for building up of its end points is: R _k2, bottom p ₁, p ₂as 3D model front bottom end two point; Again by p ₁, p ₂build bus 3D model front top 2 p ₃, p ₄:

$\left\{\begin{array}{c}{p}_3.x=p_1.x,p_3.y=p_1.y,p_3.z=p_1.z+h\\ p_4.x=p_2.x,p_4.y=p_2.y,p_4.z=p_2.z+h\end{array}\right.$

Wherein, h represents bus height, and (x, y, z) represents the three-dimensional coordinate of each point in world coordinate system;

Then according to the perspective view of bus on xOy, 2 p bottom 3D model tail end are built ₅, p ₆:

$\left\{\begin{array}{c}\mathrm{\θ}=\arctan \left(\frac{p_1.y-p_2.y}{p_2.x-p_1.x}\right)\\ p_j.x=p_i.x-l\sin \mathrm{\θ}\\ p_j.y=p_i.x-l\cos \mathrm{\θ}\end{array}\right.$

Wherein, (i, j) ∈ { (1,5), (2,6) }, θ represents in perspective view, line segment (p ₁, p ₂) with the angle of y-axis,

L represents bus length of wagon, and (x, y, z) represents the three-dimensional coordinate of each point in world coordinate system;

Last by p ₅, p ₆build 3D model tail end 2, top p ₇, p ₈:

$\left\{\begin{array}{c}{p}_7.x=p_5.x,p_7.y=p_5.y,p_7.z=p_5.z+h\\ p_8.x=p_6.x,p_8.y=p_6.y,p_8.z=p_6.z+h\end{array}\right.$

Wherein, h represents bus height, and (x, y, z) represents the three-dimensional coordinate of each point in world coordinate system;

5th step concrete steps are:

1. adopt template matching method, calculate matching factor η ₁,

Wherein, represent the overlapping region between 3D model gray-scale map and Eigenvector gray-scale map; ψ is 3D model gray-scale map, and its pixel value is 0 or 1; for LSD Eigenvector gray-scale map _; represent and morphological dilations process is carried out to image; Σ I represents and carries out pixel summation to gray level image I; represent and threshold process done to gray-scale map I,

${\mathrm{Threshold}}_t^v\left(I_{x,y}\right)=\left\{\begin{array}{c}v,I_{x,y}\≥v\\ 0,\mathrm{otherwise}\end{array}\right.$

2. according to η ₁judge whether Current vehicle is bus, if η ₁> TH _l, TH _lfor setting threshold value, representation feature line segment overlaps with 3D model large percentage, thinks that this vehicle may be bus, enters next step, and employing knearest neighbour method judges further; If η ₁> TH _h, TH _hfor setting threshold value and TH _h> TH _l, then think that this vehicle is bus, judge to terminate; If η ₁≤ TH _l, then think that this vehicle is non-bus, judge to terminate;

3. bee-line algorithm is implemented to the vehicle needing in previous step to judge further, order represent removing overlapping region 3D illustraton of model; represent removing overlapping region eigenvector figure; I (α) represents the gray-scale value of pixel α; P (α) represents the position of pixel α under image coordinate system, initialization counter sum=0, then minimum distance method is specifically implemented as follows:

(1) to pixel α ∈ ψ _hand I (α) ≠ 0, centered by p (α), step is that the length of side sets up square aearch window;

(2) calculate pixel α to arrive bee-line, be specially

(3) if d (α)≤d _tH, then sum=sum+1 and I (β is made ₀)=0, d _tHfor setting threshold value;

(4) step (1) is repeated until ψ _hin pixel travel through completely;

4. matching factor η is calculated ₂, ${\mathrm{\η}}_2=\frac{\mathrm{sum}}{\mathrm{\Σ}{\mathrm{\ψ}}_h};$

According to η ₂judge whether Current vehicle is bus, if η ₂> TH _d, TH _dfor setting threshold value, representation feature line segment overlaps with 3D model large percentage, judges that this vehicle is bus; Otherwise, then judge that this vehicle is non-bus, judge to terminate.

2. bus model recognizing method according to claim 1, is characterized in that: in the 3rd step, after completing the structure to 3D model end points, route selection section (p ₁, p ₂) (p ₁, p ₃) (p ₂, p ₄) (p ₃, p ₄) (p ₃, p ₇) (p ₄, p ₈) be camera visible segment, line segment (p ₅, p ₆) be the invisible line segment of camera, line segment (p ₂, p ₆) (p ₆, p ₈) then need according to p ₆position judge, under image coordinate system, if then (p ₂, p ₆) (p ₆, p ₈) be camera visible segment; If p ₆∈ R (p ₁₂₃₄), then it is invisible line segment; R (p ₁₂₃₄) represent by a p ₁, p ₂, p ₃, p ₄the rectangle formed; In like manner, to line segment (p ₁, p ₅) (p ₅, p ₇) judge; Select camera visible segment as the line segment aggregate of 3D model.

3. bus model recognizing method according to claim 1, it is characterized in that: in the 5th step: first adopt stencil matching method to identify to 3D model gray-scale map and Eigenvector gray-scale map, adopt knearest neighbour method to judge further to the vehicle meeting condition for identification, knearest neighbour method only need mate remaining 3D model pixel point and Eigenvector pixel after template coupling.