CN105404856A - Public traffic vehicle seat occupied state detection method - Google Patents

Abstract

Disclosed is a public traffic vehicle seat occupied state detection method. A monitored seat region is selected firstly; according to bus GPS data, video image frames of a vehicle operating between stations are acquired; on the basis of a gray image, an abnormal illumination region of the seat region is removed; blue filtering is performed in an HSV color space, and a chroma filtering coefficient is calculated; for an initial frame, if the chroma filtering coefficient is greater than a high threshold, a vacant seat is judged to exist, otherwise the vacant seat is judged to not exist; for a non-initial frame, if the chroma filtering coefficient is greater than the high threshold, a vacant seat is judged to exist, and if the chroma filtering coefficient is less than a low threshold, the vacant seat is judged to not exist; if the chroma filtering coefficient is between the low threshold and the high threshold, a difference superposed region between a current frame and a previous frame is extracted, and a difference superposed coefficient is calculated; if the difference superposed coefficient is greater than a difference threshold, a vacant seat is judged to exist, otherwise, a similar superposed region is extracted, and a similar region superposed coefficient is calculated; and, if the similar region superposed coefficient is greater than a similar threshold, a vacant seat is judged to exist, otherwise the vacant seat is judged to not exist in the monitored region. The method has the advantage of high detection precision.

Description

A kind of public transit vehicle seat occupancy states detection method

Technical field:

The present invention relates to a kind of expanded application of bus camera monitoring system, particularly relate to a kind of public transit vehicle seat occupancy states detection method based on image processing techniques.

Background technology:

Urban public transport because carrying capacity is large, security and operational efficiency high, be the important component part of urban transportation, be considered to the approach of the best solving urban traffic congestion.The punctual, comfortable of public transit vehicle can increase the attractive force of public transport to passenger.Whether have vacant position in bus is one of the important evaluation index of public transport comfort level.

At present, in bus, room detection method is mainly obtained by seating capacity in contrast passenger inside the vehicle's number and car.The method of public transit vehicle passenger inside the vehicle programming count mainly contains: to swipe the card statistic law etc. based on the passenger flow statistics method of pressure pedal, infrared radiation detection apparatus statistic law, bill.Pressure pedal passenger flow statistics method and infrared radiation detection apparatus statistic law are passed through at upper and lower car door setting pressure pedal sensor or infrared radiation detection apparatus, the number that the real time measure is got on the bus and got off, thus the ridership in statistics car.When passenger crowded by time, the number of counting easily produces error, and sensor is expensive and easily damage.Bill swipe the card statistic law be passenger swipe the card at entrance door get on the bus, the exit door mode of getting off of swiping the card obtains passenger flow data, the method is simple, feasible, but cannot add up to get on the bus without card needs to pay existing part passenger.The passenger flow data that goes out of three kinds of method statistics above, needs concrete seating capacity in conjunction with public transport to judge the room situation of car door, and the result judged can produce the result of mistake along with the passenger flow error of statistics.In addition, direct surveillance's method is also had, mainly through human eye real-time monitored public transport video data, to judge ridership in car and Seats Occupied Information.It is lower to there is efficiency in direct surveillance's method, and cost of labor is higher, is not suitable for extensive detection.

Summary of the invention:

The object of the present invention is to provide a kind of public transit vehicle seat occupancy states detection method.

For achieving the above object, the present invention adopts following technical scheme to realize:

A kind of public transit vehicle seat occupancy states detection method, comprises the following steps:

Step 0: the seating area Ba choosing supervision; Wherein, selected seating area Ba is binary map, the region representation seating area of all pixel compositions of Ba (x, y)=1, the non-seating district of region representation of all pixel compositions of Ba (x, y)=0;

Step 1: vehicle arrives at a station detection

Read the GPS location data of public transit vehicle, if the position data of the GPS location data of this moment public transit vehicle and bus parking website overlaps, then reset by frame number, i.e. k=0, proceeds to step 1; If position data does not overlap, then proceed to step 2;

Step 2: read the video frame image in the bus also between storage sites

Sample frequency is 5S/ frame, the picture frame symbol E of collection _krepresent, k refers to the frame number of current image frame, each reading one frame, and by frame number cumulative 1, i.e. k=k+1, image size is M × N, wherein, M and N separated image frame E _khead office and total row;

Step 3: gray processing process

The picture frame E that step 2 is obtained _kcarry out gray processing process, obtain gray level image f _k;

Step 4: remove illumination abnormal area

In the seating area Ba that step 0 obtains, remove gray level image f _kabnormal area of illumination, namely strong or too low region is crossed in illumination, obtains seating area Se to be measured _k;

Step 5:RGB color space goes to hsv color space

By the RGB image E collected in step 2 _kbe converted to hsv color space, obtain HSV image Be _k;

Step 6: to HSV image Be _kcarry out color filter

The seating area Se to be measured obtained in step 4 _kin region, to the HSV image Be obtained in step 5 _kcarry out color filter, obtain room district Ca _k;

Step 7: calculate colourity filter factor a

Step 8: initial frame judges

Judge current detection frame whether as initial frame, namely judge whether k equals 1; If k=1, proceed to step 9; If k ≠ 1, then proceed to step 10;

Step 9: calculate initial frame seat occupancy mark, and proceed to step 21

Use fg _krepresent the image seat occupancy mark of kth (k=1) frame, computing formula is as follows:

In formula, Ta ₁for high threshold, value is 0-1;

Step 10: the seat occupancy mark calculating non-initial frame

Calculate the seat occupancy mark of non-initial frame, computing formula is as follows:

Wherein, fg _k=2 represent that kth frame seat occupancy states is unknown, need to carry out subsequent analysis; Ta ₂for Low threshold, value is 0-1 and meets Ta ₂≤ Ta ₁;

Step 11: seat occupancy states judges

If fg _k≠ 2, i.e. fg _k=0 or fg _k=1, proceed to step 21; If fg _k=2, proceed to step 12;

Step 12: the difference value calculating present frame and front frame

The colour difference, the saturation degree that calculate present frame and front frame are respectively poor, luminance difference and fusion poor;

Step 13: detect the difference section of present frame and front frame and similar district

For the arbitrary pixel in kth frame picture, if when the difference value of present frame and front frame is greater than the threshold value of restriction, think that this pixel is discrepancy, the region be made up of discrepancy is difference section; If when the difference value of present frame and front frame is less than the threshold value of restriction, think that this pixel is similitude, the region be made up of similitude is similar district;

At seating area Se to be measured _kin, segmentation threshold is selected to difference value, obtains segmentation figure Mf _k; Segmentation figure Mf _kmiddle pixel value is the region of the pixel composition of 1 is difference section, uses symbol M D _krepresent, pixel value is the region of the pixel composition of 0 is similar district, uses symbol M I _krepresent;

Step 14: extract difference overlapping region

Extract difference section MD _kwith room district Ca _koverlapping region, obtain room disparity map HD _k, computing formula is as follows:

Step 15: calculated difference district overlap coefficient af

The computing formula of difference section overlap coefficient af is as follows:

$af=\frac{S_{H_k}}{S_{M_k}}---\left(16\right)$

Wherein, for the room disparity map HD that step 14 obtains _karea, namely this pixel values in regions is the number of pixels of 1; for the difference section MD that step 13 obtains _karea, namely this pixel values in regions is the number of pixels of 1;

Step 16: calculate seat occupancy mark

Calculate seat occupancy mark fg _k, computing formula is as follows:

Wherein, Ta ₃for discrepancy threshold, its value is 0-1;

Step 17: seat occupancy states judges

If fg _k=0, proceed to step 21; If fg _k=2, proceed to step 18;

Step 18: extract similar overlapping region

Extract similar district MI _kwith room district Ca _k-1overlapping region, obtain room similar diagram Ls _k, computing formula is as follows:

Step 19: calculate similar district overlap coefficient as

The computing formula of similar district overlap coefficient as is as follows:

$as=\frac{S_{L_k}}{S_{C_k}}---\left(19\right)$

Wherein, for the room similar diagram Ls that step 18 obtains _karea, namely this pixel values in regions is the number of pixels of 1; Sc _krepresent the room district Ca that step 6 obtains _karea, namely this pixel values in regions is the sum of all pixels of 1;

Step 20: calculate seat occupancy mark fg _k

Calculate seat occupancy mark fg _k, computing formula is as follows:

Wherein, Ta ₄for similar threshold value, its value is 0-1;

Step 21: seat occupancy states judges

If fg _k=0, judging that this moment seat is unoccupied, is vacant seat state; If fg _k=1, judge that this moment loses one's seat use, for without seat state.

The present invention further improves and is, in step 3, and the picture frame E that step 2 is obtained _kcarry out gray processing process computing formula as follows:

f _k(x,y)＝0.3×R _k(x,y)+0.59×G _k(x,y)+0.11×B _k(x,y)(1)

Wherein, f _k(x, y) represents gray level image f _kthe gray-scale value of pixel (x, y), R _k(x, y), G _k(x, y) and B _k(x, y) represents picture frame E respectively _kthe red color component value of location of pixels (x, y), green component values and blue color component value.

The present invention further improves and is, in step 4, in the seating area Ba that step 0 obtains, removes gray level image f _kabnormal area of illumination computing formula as follows:

In formula, δ ₁for low clear zone threshold value, value is 0-50; δ ₂for highlight bar threshold value, value is 230-256.

The present invention further improves and is, in step 5, by the picture frame E collected in step 2 _kbe converted to hsv color space computing formula as follows:

$H_k(x,y)=arccos\frac{\frac{1}{2}\[\left(R_k\right(x,y)-G_k(x,y))+\left(R_k\right(x,y)-B_k(x,y))\]}{\sqrt{{\left(R_k\right(x,y)-G_k(x,y\left)\right)}^2+\left(R_k\right(x,y)-B_k(x,y))\left(G_k\right(x,y)-B_k(x,y))}}---\left(3\right)$

$S_k(x,y)=\frac{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y))-min\left(R_k\right(x,y),G_k(x,y),B_k(x,y))}{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y))}---\left(4\right)$

$V_k(x,y)=\frac{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y))}{255}---\left(5\right)$

Wherein, H _k(x, y), S _k(x, y) and V _k(x, y) is respectively the chromatic value of pixel (x, y), intensity value and brightness value.

The present invention further improves and is, in step 6, and the seating area Se to be measured obtained in step 4 _kin region, to the HSV image Be obtained in step 5 _kcarry out color filter computing formula as follows:

In formula, h _lfor low Chroma threshold, span is 0-1; h _hfor high chroma threshold value, span is 0-1, and meets h _l< h _h; s _lfor low saturation threshold value, span is 0-1; s _hfor high saturation threshold value, value is 0-1, and meets s _l< s _h; v _lfor low brightness threshold, value is 0-1; v _hfor high brightness threshold value, value is 0-1, and meets v _l< v _h.

The present invention further improves and is, in step 7, the computing formula of colourity filter factor a is as follows:

$a=\frac{S_{C_k}}{S_{S_k}}---\left(7\right)$

Wherein, represent the room district Ca that step 6 obtains _karea, namely in this region in pixel value be the sum of all pixels of 1; represent that step 4 obtains seating area Se to be checked _karea, namely this pixel values in regions is the sum of all pixels of 1.

The present invention further improves and is, in step 12, the colour difference, the saturation degree that calculate present frame and front frame are poor, luminance difference and to merge the computing formula differed from as follows:

Ch _k(x,y)＝|H _k(x,y)-H _k-1(x,y)|(10)

Cs _k(x,y)＝|S _k(x,y)-S _k-1(x,y)|(11)

Cv _k(x,y)＝|V _k(x,y)-V _k-1(x,y)|(12)

Cz _k(x,y)＝Ch _k(x,y)+Cs _k(x,y)+Cv _k(x,y)(13)

Wherein, Ch _kfor the colour difference of kth frame picture and front frame; Cs _kfor the saturation degree of kth frame picture and front frame is poor; Cv _kfor the luminance difference of kth frame picture and front frame; Cz _kfor the fusion of kth frame picture and front frame is poor.

The present invention further improves and is, in step 13, computing formula is as follows:

Wherein, C ₁, C ₂, C ₃and C ₄be segmentation threshold, C ₁, C ₂and C ₃value be 0.15-0.3, C ₄value be 0.5-0.7.

Relative to prior art, the video acquisition device that the present invention is based in bus obtains the vedio data in car, excavates video image information, analyzes the Seats Occupied Information of public transit vehicle.The present invention has following realistic meaning: one is the service level improving public transport, improves the trip attraction power of public transit vehicle.In bus, Seats Occupied Information is that passenger rides one of the important evaluation index of comfort level or crowding evaluation, after seat in car is occupied, crowding in further analysis car can be bus dispatching department and provides corresponding early warning signal, for vehicle scheduling provides Data support.Two is for public transit vehicle Truck type choice provides foundation.Because the volume of passenger traffic of different public bus network there are differences, by detecting Seats Occupied Information, choose reasonable the vehicle of this circuit can be applicable to.The automatic testing method of Seats Occupied Information in the bus that the present invention proposes, the advantages such as the method has simply, economy, accuracy of detection are high, travelling speed is fast, robustness is good.

Accompanying drawing illustrates:

Fig. 1 is seating area Ba schematic diagram.

Fig. 2 is picture frame E _kschematic diagram.

Fig. 3 is gray level image f _kschematic diagram.

Fig. 4 is seating area Se to be measured _kschematic diagram.

Fig. 5 is room district Ca _kschematic diagram.

Fig. 6 is segmentation figure Mf _kschematic diagram.

Fig. 7 is room disparity map HD _kschematic diagram.

Fig. 8 is room similar diagram Ls _kschematic diagram.

Embodiment:

Below in conjunction with drawings and Examples, the present invention is described in further detail.

For monitoring zones of different in bus, current bus is generally laid with 3 and above camera, to gather the view data of zones of different at diverse locations such as front portion, middle part and rear portions.In a particular embodiment, select the vedio data of bus front channels, that uses method of the present invention supervision dress circle takies situation.The invention provides a kind of public transit vehicle seat occupancy states detection method, concrete implementation step is expressed as follows:

Step S0: the seating area Ba choosing supervision.

Due to the difference of bus model and camera installation site, in the video image that different buses collects, also can there is certain difference in the size of seating area and position, need carry out the delimitation of seating area according to different buses.Camera in embodiment mainly monitors front-seat seating area, first the selected seating area needing to monitor, conventional seating area method for selecting has chooses method and artificial selected method automatically.Automatically method of choosing is according to the positional information residing for seat and colouring information, automatically the selected seating area needing to monitor.Artificial selected method be staff by visual determination, the manually selected seating area needing to monitor, and selected seating area is stored for subsequent use.Compare and automatically choose method, the precision of artificial selected method is higher, and what adopt in the present embodiment is manually select method.The seating area delimited represents with symbol Ba, and artificial selected seating area Ba is binary map, Ba (x, the region representation seating area of all pixels composition of y)=1, the non-seating district of region representation of all pixel compositions of Ba (x, y)=0, as shown in Figure 1.

Proceed to step S1.

Step S1: vehicle arrives at a station detection.

Read the GPS location data of public transit vehicle, if the position data of the GPS location data of this moment public transit vehicle and bus parking website overlaps, then reset by frame number, i.e. k=0, proceeds to step S1; If position data does not overlap, then proceed to step S2.

Step S2: read the video frame image in the bus also between storage sites.

Sample frequency is 5S/ frame, the picture frame symbol E of collection _krepresent, k refers to the frame number of current image frame, each reading one frame, by frame number cumulative 1, i.e. k=k+1.The image gathered is RGB color format, and image size is that M × N, M and N represent picture frame E respectively _khead office and total row, i.e. M=576, N=704.What choose in the present embodiment is that the 12nd two field picture carries out seat detection, picture frame E _kas shown in Figure 2, k=12.

Proceed to step S3.

Step S3: gray processing process.

By the picture frame E that step S2 obtains _kcarry out gray processing process, obtain gray level image f _k, as shown in Figure 3.Computing formula is as shown in formula (1):

f _k(x,y)＝0.3×R _k(x,y)+0.59×G _k(x,y)+0.11×B _k(x,y)(1)

Wherein, f _k(x, y) represents gray level image f _kthe gray-scale value of pixel (x, y), R _k(x, y), G _k(x, y) and B _k(x, y) represents picture frame E respectively _kthe red color component value of location of pixels (x, y), green component values and blue color component value.

Proceed to step S4.

Step S4: remove illumination abnormal area.

Bus internal cause uneven illumination is even, makes the brightness in some region too high or too low, needs to remove these regions when detecting.In the seating area Ba that step S0 obtains, remove abnormal area of illumination, namely strong or too low region is crossed in illumination, obtains seating area Se to be measured _k, as shown in Figure 4.Computing formula is as shown in formula (2):

In formula, δ ₁for low clear zone threshold value, value is 0-50; δ ₂for highlight bar threshold value, value is 230-256.In the present embodiment, δ ₁get 30, δ ₂get 250.

Proceed to step S5.

Step S5:RGB color space goes to hsv color space.

Hsv color space is convenient to perform color filter, therefore the image E will collected in step S2 _kbe converted to hsv color space, obtain HSV image Be _k, computing formula is as follows:

$H_k(x,y)=arccos\frac{\frac{1}{2}\&lsqb;\left(R_k\right(x,y)-G_k(x,y))+\left(R_k\right(x,y)-B_k(x,y))\&rsqb;}{\sqrt{{\left(R_k\right(x,y)-G_k(x,y\left)\right)}^2+\left(R_k\right(x,y)-B_k(x,y))\left(G_k\right(x,y)-B_k(x,y))}}---\left(3\right)$

$S_k(x,y)=\frac{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y))-min\left(R_k\right(x,y),G_k(x,y),B_k(x,y))}{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y))}---\left(4\right)$

$V_k(x,y)=\frac{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y))}{255}---\left(5\right)$

Wherein, H _k(x, y), S _k(x, y) and V _k(x, y) is respectively the chromatic value of pixel (x, y), intensity value and brightness value.

Proceed to step S6.

Step S6: to HSV image Be _kcarry out color filter.

At seating area Se to be measured _kin (obtaining in step S4) region, to the HSV image Be obtained in step S5 _kcarry out color filter, obtain room district Ca _k, as shown in Figure 5.In the present embodiment, public transport dress circle is blue, at seating area Se to be measured _kin region, to HSV image Be _kcarry out blue color filtered, computing formula is as shown in formula (6):

In the present embodiment, h _lfor low Chroma threshold, value is 0.5-0.6; h _hfor high chroma threshold value, value is 0.65-0.75; s _lfor low saturation threshold value, value is 0.15-0.25; s _hfor high saturation threshold value, value is 0.8-0.9; v _lfor low brightness threshold, value is 0.2-0.35; v _hfor high brightness threshold value, value is 0.9-1.0.In the present embodiment, h _lvalue is 0.55, h _hvalue is 0.7, s _lvalue is 0.2, s _hvalue is 0.8, v _lvalue is 0.3, v _hvalue is 0.9.

Proceed to step S7.

Step S7: calculate colourity filter factor a.

Colourity filter factor is for characterizing the colourity ratio at seat, and colourity ratio is larger, and the probability lost one's seat is less.The computing formula of colourity filter factor a is as shown in formula (7):

$a=\frac{S_{C_k}}{S_{S_k}}---\left(7\right)$

Wherein, represent the room district Ca that step S6 obtains _karea, namely in this region in pixel value be the sum of all pixels of 1; represent that step S4 obtains seating area Se to be checked _karea, namely this pixel values in regions is the sum of all pixels of 1.In the present embodiment, Ca _karea be 5680, Se _karea be 10021, coefficient a is 0.567.

Proceed to step S8.

Step S8: initial frame judges.

Judge current detection frame whether as initial frame, namely judge whether k equals 1.If k=1, proceed to step S9; If k ≠ 1, then proceed to step S10.

In embodiment, k=12, now proceeds to step S10.

Step S9: calculate initial frame seat occupancy mark.

Use fg _krepresent the image seat occupancy mark of kth (k=1) frame, computing formula is as shown in formula (8):

In formula, Ta ₁for high threshold, value is 0-1.In the present embodiment, Ta ₁value be 0.85.

Proceed to step S21.

Step S10: the seat occupancy mark calculating non-initial frame.

Calculate the seat occupancy mark of non-initial frame, namely during k ≠ 1, computing formula is as shown in formula (9):

Wherein, fg _k=2 represent that kth frame seat occupancy states is unknown, need to carry out subsequent analysis; Ta ₂for Low threshold, value is 0-1 and meets Ta ₂≤ Ta ₁, in the present embodiment, Ta ₂value be 0.55.Obtain colourity filter factor a=0.567 by step S7, its value is less than 0.85 and is greater than 0.55, and therefore taking of this moment marks fg _kbe 2.

Proceed to step S11.

Step S11: seat occupancy states judges.

If fg _k≠ 2, i.e. fg _k=0 or fg _k=1, proceed to step S21; If fg _k=2, proceed to step S12.

Step S12: the difference value calculating present frame and front frame.

The colour difference, the saturation degree that calculate present frame and front frame are respectively poor, luminance difference and fusion poor.Computing formula is as follows:

Ch _k(x,y)＝|H _k(x,y)-H _k-1(x,y)|(10)

Cs _k(x,y)＝|S _k(x,y)-S _k-1(x,y)|(11)

Cv _k(x,y)＝|V _k(x,y)-V _k-1(x,y)|(12)

Cz _k(x,y)＝Ch _k(x,y)+Cs _k(x,y)+Cv _k(x,y)(13)

Wherein, Ch _kfor the colour difference of kth frame picture and front frame; Cs _kfor the saturation degree of kth frame picture and front frame is poor; Cv _kfor the luminance difference of kth frame picture and front frame; Cz _kfor the fusion of kth frame picture and front frame is poor.

Proceed to step S13.

Step S13: detect the difference section of present frame and front frame and similar district.

For the arbitrary pixel in kth frame picture, if when the difference value of present frame and front frame is excessive, think that this pixel is discrepancy, the region be made up of discrepancy is difference section; If when the difference value of present frame and front frame is less, think that this pixel is similitude, the region be made up of similitude is similar district.

At seating area Se to be measured _kin, suitable segmentation threshold is selected to above each difference value, obtains segmentation figure Mf _k, as shown in Figure 6.Segmentation figure Mf _kmiddle pixel value is the region of the pixel composition of 1 is difference section, uses symbol M D _krepresent, pixel value is the region of the pixel composition of 0 is similar district, uses symbol M I _krepresent, computing formula is as shown in formula (14):

Wherein, C ₁, C ₂, C ₃and C ₄be segmentation threshold, C ₁, C ₂and C ₃value be 0.15-0.3, C ₄value be 0.5-0.7.In the present embodiment, C ₁value is 0.25, C ₂value is 0.25, C ₃value is 0.25, C ₄value is 0.6.

Proceed to step S14.

Step S14: extract difference overlapping region.

Extract difference section MD _k(being obtained by step S13) and room district Ca _kthe overlapping region of (being obtained by step S6), obtains room disparity map HD _k, as shown in Figure 7, computing formula is as shown in formula (15):

Proceed to step S15.

Step S15: calculated difference district overlap coefficient af.

The computing formula of difference section overlap coefficient af is as follows:

$af=\frac{S_{H_k}}{S_{M_k}}---\left(16\right)$

Wherein, for the room disparity map HD that step S14 obtains _karea, namely this pixel values in regions is the number of pixels of 1; for the segmentation figure Mf that step S13 obtains _kmiddle difference section MD _karea, i.e. this Mf _kmiddle pixel value is the number of pixels of 1.In the present embodiment, room disparity map HD _karea be 3; Difference section MD _karea be 232, difference section overlap coefficient af is 0.012.

Proceed to step S16.

Step S16: calculate seat occupancy mark.

Calculate seat occupancy mark fg _k, computing formula is as shown in formula (17):

Wherein, Ta ₃for discrepancy threshold, its value is 0-1, in the present embodiment Ta ₃value is 0.7.Obtain difference section overlap coefficient af=0.012 by step S16, its value is less than 0.7, and therefore taking of this moment marks fg _kbe 2.

Proceed to step S17.

Step S17: seat occupancy states judges.

If fg _k=0, proceed to step 21; If fg _k=2, proceed to step S18.

Step S18: extract similar overlapping region.Extract segmentation figure Mf _kin similar district MI _k(being obtained by step S13) and room district Ca _k-1the overlapping region of (being obtained by step S5).Obtain room similar diagram Ls _k, as shown in Figure 8, computing formula is as shown in formula (18):

Proceed to step S19.

Step S19: calculate similar district overlap coefficient as.

The computing formula of similar district overlap coefficient as is as shown in formula (19):

$as=\frac{S_{L_k}}{S_{C_k}}---\left(19\right)$

Wherein, for the room similar diagram Ls that step S18 obtains _karea, namely this pixel values in regions is the number of pixels of 1; Sc _krepresent the room district Ca that step S6 obtains _karea, namely this pixel values in regions is the sum of all pixels of 1.In the present embodiment, room similar diagram Ls _karea be 4740; Room district Ca _karea be 4989, similar district overlap coefficient as is 0.95.

Proceed to step S20.

Step S20: calculate seat occupancy mark fg _k.

Calculate seat occupancy mark fg _k, computing formula is as shown in formula (20):

Wherein, Ta ₄for similar threshold value, its value is 0-1, Ta in the present embodiment ₄value is 0.56.Obtain difference section overlap coefficient as=0.95 by step S19, its value is greater than 0.56, and therefore taking of embodiment marks fg _kbe 0.

Proceed to step S21.

Step S21: seat occupancy states judges.

If fg _k=0, judging that this moment seat is unoccupied, is vacant seat state; If fg _k=1, judge that this moment loses one's seat use, for without seat state.

Proceed to step S1.

According to the technical scheme of above invention, gather the vedio data in bus, utilize the detection method that the present invention adopts, the automatic detection of Seats Occupied Information in car can be realized, in bus, Seats Occupied Information data can be used for the service level improving public transport, improve the trip attraction power of public transit vehicle, also can be public transit vehicle Truck type choice and Data support is provided.

From working time and accuracy of detection two aspect, the relative merits of the present invention program are described.

(1) working time.Between certain station, duration 2 points is run 3 seconds for certain bus in embodiment, gather 24 two field pictures altogether, at IntelI3M350 processor, the computing machine of 4GB internal memory, utilizes MATLAB software to emulate, measures the picture frame collected, average every frame runs 0.19s consuming time, compared with image frame acquisitions frequency 5s/ frame, the inventive method processing speed is fast, can be applicable in real time processing system.

(2) accuracy of detection.Use W _rrepresent the frame number be correctly detected, W represents the totalframes of detection, and η represents accuracy, and selected accuracy is as the evaluation index of accuracy of detection of the present invention, and strive for that rate is higher, accuracy of detection of the present invention is higher.Accuracy calculating formula is as follows:

$\mathrm{\&eta;}=\frac{W_r}{W}\&times;100\%---\left(21\right)$

Run to terminal to certain bus in the present embodiment from starting point, run duration: when 1 23 points, gather the image run between 684 frame stations altogether, the frame number be correctly detected is 636 frames, calculates for embodiment, and the accuracy of algorithm is 93%.In addition, the 8 sections of video images gathered are tested, finds that accuracy of the present invention is all more than 90%, has higher robustness.

Claims (8)

1. a public transit vehicle seat occupancy states detection method, is characterized in that, comprises the following steps:

Step 0: the seating area Ba choosing supervision; Wherein, selected seating area Ba is binary map, the region representation seating area of all pixel compositions of Ba (x, y)=1, the non-seating district of region representation of all pixel compositions of Ba (x, y)=0;

Step 1: vehicle arrives at a station detection

Read the GPS location data of public transit vehicle, if the position data of the GPS location data of this moment public transit vehicle and bus parking website overlaps, then reset by frame number, i.e. k=0, proceeds to step 1; If position data does not overlap, then proceed to step 2;

Step 2: read the video frame image in the bus also between storage sites

Sample frequency is 5S/ frame, the picture frame symbol E of collection _krepresent, k refers to the frame number of current image frame, each reading one frame, and by frame number cumulative 1, i.e. k=k+1, image size is M × N, wherein, M and N separated image frame E _khead office and total row;

Step 3: gray processing process

The picture frame E that step 2 is obtained _kcarry out gray processing process, obtain gray level image f _k;

Step 4: remove illumination abnormal area

In the seating area Ba that step 0 obtains, remove gray level image f _kabnormal area of illumination, namely strong or too low region is crossed in illumination, obtains seating area Se to be measured _k;

Step 5:RGB color space goes to hsv color space

By the RGB image E collected in step 2 _kbe converted to hsv color space, obtain HSV image Be _k;

Step 6: to HSV image Be _kcarry out color filter

The seating area Se to be measured obtained in step 4 _kin region, to the HSV image Be obtained in step 5 _kcarry out color filter, obtain room district Ca _k;

Step 7: calculate colourity filter factor a

Step 8: initial frame judges

Judge current detection frame whether as initial frame, namely judge whether k equals 1; If k=1, proceed to step 9; If k ≠ 1, then proceed to step 10;

Step 9: calculate initial frame seat occupancy mark, and proceed to step 21

Use fg _krepresent the image seat occupancy mark of kth (k=1) frame, computing formula is as follows:

In formula, Ta ₁for high threshold, value is 0-1;

Step 10: the seat occupancy mark calculating non-initial frame

Calculate the seat occupancy mark of non-initial frame, computing formula is as follows:

Wherein, fg _k=2 represent that kth frame seat occupancy states is unknown, need to carry out subsequent analysis; Ta ₂for Low threshold, value is 0-1 and meets Ta ₂≤ Ta ₁;

Step 11: seat occupancy states judges

If fg _k≠ 2, i.e. fg _k=0 or fg _k=1, proceed to step 21; If fg _k=2, proceed to step 12;

Step 12: the difference value calculating present frame and front frame

The colour difference, the saturation degree that calculate present frame and front frame are respectively poor, luminance difference and fusion poor;

Step 13: detect the difference section of present frame and front frame and similar district

For the arbitrary pixel in kth frame picture, if when the difference value of present frame and front frame is greater than the threshold value of restriction, think that this pixel is discrepancy, the region be made up of discrepancy is difference section; If when the difference value of present frame and front frame is less than the threshold value of restriction, think that this pixel is similitude, the region be made up of similitude is similar district;

At seating area Se to be measured _kin, segmentation threshold is selected to difference value, obtains segmentation figure Mf _k; Segmentation figure Mf _kmiddle pixel value is the region of the pixel composition of 1 is difference section, uses symbol M D _krepresent, pixel value is the region of the pixel composition of 0 is similar district, uses symbol M I _krepresent;

Step 14: extract difference overlapping region

Extract difference section MD _kwith room district Ca _koverlapping region, obtain room disparity map HD _k, computing formula is as follows:

Step 15: calculated difference district overlap coefficient af

The computing formula of difference section overlap coefficient af is as follows:

$af=\frac{S_{H_k}}{S_{M_k}}---\left(16\right)$

Wherein, S _hkfor the room disparity map HD that step 14 obtains _karea, namely this pixel values in regions is the number of pixels of 1; S _mkfor the difference section MD that step 13 obtains _karea, namely this pixel values in regions is the number of pixels of 1;

Step 16: calculate seat occupancy mark

Calculate seat occupancy mark fg _k, computing formula is as follows:

Wherein, Ta ₃for discrepancy threshold, its value is 0-1;

Step 17: seat occupancy states judges

If fg _k=0, proceed to step 21; If fg _k=2, proceed to step 18;

Step 18: extract similar overlapping region

Extract similar district MI _kwith room district Ca _k-1overlapping region, obtain room similar diagram Ls _k, computing formula is as follows:

Step 19: calculate similar district overlap coefficient as

The computing formula of similar district overlap coefficient as is as follows:

$as=\frac{S_{L_k}}{S_{C_k}}---\left(19\right)$

Wherein, SL _kfor the room similar diagram Ls that step 18 obtains _karea, namely this pixel values in regions is the number of pixels of 1; Sc _krepresent the room district Ca that step 6 obtains _karea, namely this pixel values in regions is the sum of all pixels of 1;

Step 20: calculate seat occupancy mark fg _k

Calculate seat occupancy mark fg _k, computing formula is as follows:

Wherein, Ta ₄for similar threshold value, its value is 0-1;

Step 21: seat occupancy states judges

If fg _k=0, judging that this moment seat is unoccupied, is vacant seat state; If fg _k=1, judge that this moment loses one's seat use, for without seat state.

2. a kind of public transit vehicle seat occupancy states detection method according to claim 1, is characterized in that, in step 3, and the picture frame E that step 2 is obtained _kcarry out gray processing process computing formula as follows:

f _k(x,y)＝0.3×R _k(x,y)+0.59×G _k(x,y)+0.11×B _k(x,y)(1)

Wherein, f _k(x, y) represents gray level image f _kthe gray-scale value of pixel (x, y), R _k(x, y), G _k(x, y) and B _k(x, y) represents picture frame E respectively _kthe red color component value of location of pixels (x, y), green component values and blue color component value.

3. a kind of public transit vehicle seat occupancy states detection method according to claim 1, is characterized in that, in step 4, in the seating area Ba that step 0 obtains, removes gray level image f _kabnormal area of illumination computing formula as follows:

In formula, δ ₁for low clear zone threshold value, value is 0-50; δ ₂for highlight bar threshold value, value is 230-256.

4. a kind of public transit vehicle seat occupancy states detection method according to claim 1, is characterized in that, in step 5, by the picture frame E collected in step 2 _kbe converted to hsv color space computing formula as follows:

$H_k(x,y)=\arccos \frac{\frac{1}{2}\&lsqb;\left(R_k\left(x,y\right)-G_k\left(x,y\right)\right)+\left(R_k\left(x,y\right)-B_k\left(x,y\right)\right)\&rsqb;}{\sqrt{{\left(R_k\left(x,y\right)-G_k\left(x,y\right)\right)}^2+\left(R_k\left(x,y\right)-B_k\left(x,y\right)\right)\left(G_k\left(x,y\right)-B_k\left(x,y\right)\right)}}---\left(3\right)$

$S_k(x,y)=\frac{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y\left)\right)-min\left(R_k\right(x,y),G_k(x,y),B_k(x,y\left)\right)}{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y\left)\right)}---\left(4\right)$

$V_k(x,y)=\frac{max\left(R_k\right(x,y),G_k(x,y),B_k(x,y\left)\right)}{255}---\left(5\right)$

Wherein, H _k(x, y), S _k(x, y) and V _k(x, y) is respectively the chromatic value of pixel (x, y), intensity value and brightness value.

5. a kind of public transit vehicle seat occupancy states detection method according to claim 1, is characterized in that, in step 6, and the seating area Se to be measured obtained in step 4 _kin region, to the HSV image Be obtained in step 5 _kcarry out color filter computing formula as follows:

In formula, h _lfor low Chroma threshold, span is 0-1; h _hfor high chroma threshold value, span is 0-1, and meets h _l< h _h; s _lfor low saturation threshold value, span is 0-1; s _hfor high saturation threshold value, value is 0-1, and meets s _l< s _h; v _lfor low brightness threshold, value is 0-1; v _hfor high brightness threshold value, value is 0-1, and meets v _l< v _h.

6. a kind of public transit vehicle seat occupancy states detection method according to claim 1, is characterized in that, in step 7, the computing formula of colourity filter factor a is as follows:

$a=\frac{S_{C_k}}{S_{S_k}}---\left(7\right)$

Wherein, S _ckrepresent the room district Ca that step 6 obtains _karea, namely in this region in pixel value be the sum of all pixels of 1; S _skrepresent that step 4 obtains seating area Se to be checked _karea, namely this pixel values in regions is the sum of all pixels of 1.

7. a kind of public transit vehicle seat occupancy states detection method according to claim 1, is characterized in that, in step 12, the colour difference, the saturation degree that calculate present frame and front frame are poor, luminance difference and to merge the computing formula differed from as follows:

Ch _k(x,y)＝|H _k(x,y)-H _k-1(x,y)|(10)

Cs _k(x,y)＝|S _k(x,y)-S _k-1(x,y)|(11)

Cv _k(x,y)＝|V _k(x,y)-V _k-1(x,y)|(12)

Cz _k(x,y)＝Ch _k(x,y)+Cs _k(x,y)+Cv _k(x,y)(13)

Wherein, Ch _kfor the colour difference of kth frame picture and front frame; Cs _kfor the saturation degree of kth frame picture and front frame is poor; Cv _kfor the luminance difference of kth frame picture and front frame; Cz _kfor the fusion of kth frame picture and front frame is poor.

8. a kind of public transit vehicle seat occupancy states detection method according to claim 1, it is characterized in that, in step 13, computing formula is as follows:

Wherein, C ₁, C ₂, C ₃and C ₄be segmentation threshold, C ₁, C ₂and C ₃value be 0.15-0.3, C ₄value be 0.5-0.7.