CN105608422A - Intelligent monitoring detection method for overloading of passenger car - Google Patents

Abstract

The invention provides an intelligent monitoring detection method for the overloading of a passenger car, and the method comprises the steps: loading a human head detection classifier; enabling a central aisle of the passenger car to be divided into an AOI (area of interest) of human head detection, and obtaining a video image sequence of the AOI in real time; carrying out human head detection of a real-time obtained video image of the AOI through employing the human head detection classifier, recording the detected human head information, carrying out the matching of the human head information detected in a current frame with the recorded human head information, and updating the recorded human head information; carrying out the counting of the number of times that the same human head is detected in a time period: judging that the passenger car is in a state of overloading if the number exceeds a preset threshold value, and sending alarm prompt. The method employs the technology of image processing, can detect whether the passenger car is in the state of overloading or not under the condition of no parking, is high in monitoring accuracy, has little leaked and wrong detection, is high in speed, and is low in cost.

Description

A kind of passenger carriage overloading intelligent monitoring detection method

Technical field

The present invention relates to intelligent monitoring technology field, specifically a kind of passenger carriage overloading intelligent monitoring detectsMethod.

Background technology

Along with improving constantly of human living standard and developing rapidly of transportation industry, go out to beatWork, the number of being on home leave, travelling increase day by day, especially festivals or holidays, and highway visitor related to thisFortune (coach, sightseeing bus etc.) quantity is but relatively less, and therefore passenger traffic overload is very serious.

Passenger traffic overload not only causes harmful effect to highway facilities and passenger transport market, also to vehicle itselfImpact with driving, the control of vehicle, make it easily to occur traffic accident. Data card according to statisticsReal, overload is the first cause that causes major traffic accidents at present, has great traffic over halfAccident is caused by overload.

To passenger stock overload behavior in violation of rules and regulations, hand over rule to formulate corresponding punishment regulations, in recent yearsAlso strengthened management intensity. But current detection method mainly still adopts naked eyes to detect, not onlySpeed is slower, also there will be law-executor to supervise not tight phenomenon.

Summary of the invention

The object of the present invention is to provide a kind of passenger carriage overloading intelligent monitoring detection method, utilize videoImage processing techniques, to whether existing passenger to detect in passenger vehicle passageway, judges that with this passenger vehicle isNo existence overload.

Technical scheme of the present invention is:

A kind of passenger carriage overloading intelligent monitoring detection method, comprises the following steps:

(1) load the number of people and detect grader;

(2) area-of-interest detecting for the number of people, Real-time Obtaining sense delimited in passenger vehicle central gangwayThe sequence of video images of region-of-interest;

(3) adopt the number of people to detect grader, the video image of the area-of-interest to Real-time ObtainingCarry out number of people detection, and employing records with drag the people's header detecting:

Head_i＝{x_i，y_i，score_i，frame_i}

Wherein, Head_iRepresent the information of the number of people i of record, (x_i，y_i) represent the center of mass point of number of people i,score_iRepresent the number of times that number of people i is detected, frame_iRepresent the moment that number of people i is detected;

(4) people's header present frame being detected mates with the people's header recording,Upgrade the people's header having recorded;

(5) number of times that in statistics a period of time, the same number of people is detected, if exceed predetermined threshold value,Judge that passenger vehicle exists overload, carries out alarm.

Described passenger carriage overloading intelligent monitoring detection method, in step (1), the described number of people detectsGrader is mainly trained and is obtained by following steps:

A, collect some positive and negative training sample image, and to training sample image normalized;

The HOG feature of b, extraction training sample image,

C, by the HOG feature input Cascade cascade classifier of training sample image, it is right to completeThe number of people detects the training of grader.

Described passenger carriage overloading intelligent monitoring detection method, in step (4), described by present frameThe people's header detecting mates with the people's header recording, and upgrades the number of people letter having recordedBreath, comprises the following steps:

A, adopt following formula, calculate successively the number of people that present frame detects and the number of people having recorded itBetween match index:

$d_{ji}=\sqrt{{(x_{jnow}-x_i)}^2+{(y_{jnow}-y_i)}^2}$

Wherein, d_jiRepresent that mating between the present frame number of people j detecting and the number of people i recording refers toNumber, (x_jnow，y_jnow) represent the center of mass point of number of people j, (x_i，y_i) represent the center of mass point of number of people i;

If b is d_jiBe less than predetermined threshold value, judge the present frame number of people j detecting and the people who recordsI matches, and the two is the same number of people, adopts following formula to upgrade the information of the number of people i having recorded:

$\left\{\begin{array}{c}{x}_i=x_{jnow}\\ y_i=y_{jnow}\\ {\mathrm{score}}_i={\mathrm{score}}_i+1\\ {\mathrm{frame}}_i={\mathrm{frame}}_{now}\end{array}\right.$

Wherein, frame_nowRepresent the present frame moment;

If the number of people j that present frame detects does not all mate with all numbers of people that record, judge the number of peopleJ is a new number of people, adopts following formula to carry out record to it:

Head_new＝{x_new，y_new，score_new，frame_new}

$\left\{\begin{array}{c}{x}_{new}=x_{jnow}\\ y_{new}=y_{jnow}\\ {\mathrm{score}}_{new}=1\\ {\mathrm{fram}}_{new}={\mathrm{frame}}_{new}\end{array}\right.$

Wherein, Head_newRepresent the information of the new number of people of present frame record, (x_new，y_new) tableShow the center of mass point of the new number of people, score_newRepresent the number of times that the new number of people is detected, initial valueBe 1, frame_newRepresent the moment that the new number of people is detected;

If all numbers of people that certain number of people that c has recorded detects with present frame all do not mate, meterTime between the moment that this number of people of calculating the present frame moment and recorded is detected for the last timePoor, if reach predetermined threshold value, the information of this number of people having recorded is carried out to zero clearing processing.

Beneficial effect of the present invention is:

As shown from the above technical solution, the present invention utilizes image processing techniques, can not stopIn situation, whether passenger vehicle is overloaded and detected, there is high, the undetected flase drop of the monitoring degree of accuracy few, fastThe features such as degree is fast, cost is low.

Brief description of the drawings

Fig. 1 is method flow diagram of the present invention;

Fig. 2 is HOG feature extraction flow chart;

Fig. 3 is Cascade cascade classifier flow chart;

Fig. 4 is number of people coupling and renewal flow chart.

Detailed description of the invention

Further illustrate the present invention below in conjunction with the drawings and specific embodiments.

As shown in Figure 1, a kind of passenger carriage overloading intelligent monitoring detection method, comprises the following steps:

S1, the loading number of people detect grader:

It is the grader of having trained that the number of people loading detects grader, and the training number of people detects and dividesFirst class device needs to prepare training sample, then extracts the HOG feature of training sample, finally utilizesThe Cascade cascade classifier training number of people detects grader, and concrete steps are as follows:

S11, the collection training number of people detect the required some positive and negative training sample image of grader, justThe ratio of negative training sample image is roughly 1:2, and training sample image is normalized;Positive training sample be by be arranged on the vertical or oblique upper of the number of people to the number of people image taken of camera,Negative training sample is the arbitrary image that does not comprise number of people image.

The HOG feature of S12, extraction training sample image, flow chart as shown in Figure 2, specifically walksRapid as follows:

Gamma space and the color space of S121, standardized training sample image, gamma compressionFormula is as shown in formula (1):

I(x,y)＝I_r(x,y)^gamma(1)

Wherein, gamma is compression parameters, I_r(x, y) represents the front coordinate of processing in training sample imageThe grey scale pixel value that point (x, y) is located, I (x, y) represents to process the grey scale pixel value of rear corresponding points;

The gradient of S122, the each pixel of calculation training sample image, adopts following formula (2), (3)Ask gradient magnitude R (x, y) and gradient direction θ (x, y):

$R(x,y)=\sqrt{{\left(I\right(x+1,y)-I(x-1,y\left)\right)}^2+{\left(I\right(x,y-1)-I(x,y+1\left)\right)}^2}---\left(2\right)$

θ(x,y)＝arccos(I(x+1,y)-I(x-1,y)/R(x,y))(3)

S123, training sample image is divided into several cells, to the gradient of each cellHistogram carries out the projection of regulation weight;

S124, cell is combined into large piece (block), normalized gradient histogram in piece,Normalization formula is as shown in formula (4):

$v\←v/\sqrt{\left|\right|v|{|}_2^2+{\mathrm{\ϵ}}^2}---\left(4\right)$

Wherein, ε is a very little constant, and avoiding denominator is 0;

S125, the histogram vectors in all is combined into a large HOG characteristic vector.

S13, by the HOG feature input Cascade cascade classifier of training sample image, completeThe number of people is detected to the training of grader, the flow chart of Cascade cascade classifier as shown in Figure 3.

S2, obtain real-time video sequence:

The video sequence that the number of people detects is by the camera Real-time Collection that is arranged on passenger vehicle middle front part, and rootGenerally be sitting in passenger vehicle aisle and there is trend from back to front according to passenger carriage overloading personnel, by passenger vehicleBetween passageway delimit as number of people surveyed area, i.e. area-of-interest.

S3, carry out number of people detection at area-of-interest:

Carry out at area-of-interest the headform that number of people detection record detect, headform retouchesState as formula (5):

Head_i＝{x_i，y_i，score_i，frame_i}(5)

Wherein, Head_iRepresent the information of the number of people i of record, (x_i，y_i) represent the center of mass point of number of people i,score_iRepresent the number of times that number of people i is detected, frame_iRepresent the moment that number of people i is detected.

(x_i，y_i) calculating as formula (6):

(x_i，y_i)＝(x+Rect_w/2，y+Rect_h/2)(6)

Wherein, (x, y) represents the top left corner apex of number of people i head rectangle, Rect_w、Rect_hTable respectivelyThe wide and high of an i head rectangle of leting others have a look at.

S4, number of people coupling and renewal, as shown in Figure 4:

Passenger is in passenger vehicle running, and run duration is shorter, and displacement is less, the present invention is based on thisThe number of people that feature detects present frame carries out associated, coupling with the number of people recording.

Center of mass point (the x of the number of people j that S41, present frame detect_jnow，y_jnow) with the number of people i having recordedCenter of mass point (x_i，y_i) match index d_jiCalculate as formula (7):

$d_{ji}=\sqrt{{(x_{jnow}-x_i)}^2+{(y_{jnow}-y_i)}^2}---\left(7\right)$

If S42 is d_jiBe less than predetermined threshold value Thr_d, think number of people j that present frame detects with rememberThe number of people i of record matches, and the two is the same number of people, upgrades the information of the number of people i having recorded, as public affairsShown in formula (8):

$\left\{\begin{array}{c}{x}_i=x_{jnow}\\ y_i=y_{jnow}\\ {\mathrm{score}}_i={\mathrm{score}}_i+1\\ {\mathrm{frame}}_i={\mathrm{frame}}_{now}\end{array}\right.---\left(8\right)$

Wherein, frame_nowRepresent the present frame moment, the moment that number of people j is detected.

If the number of people j that present frame detects does not all mate with all numbers of people that record, think the number of peopleJ is a new number of people, and it is carried out to record, as shown in formula (9), (10):

Head_new＝{x_new，y_new，score_new，frame_new}(9)

$\left\{\begin{array}{c}{x}_{new}=x_{jnow}\\ y_{new}=y_{jnow}\\ {\mathrm{score}}_{new}=1\\ {\mathrm{fram}}_{new}={\mathrm{frame}}_{new}\end{array}\right.---\left(10\right)$

Wherein, Head_newRepresent the information of the new number of people of present frame record, (x_new，y_new) tableShow the center of mass point of the new number of people, score_newRepresent the number of times that the new number of people is detected, initial valueBe 1, frame_newRepresent the moment that the new number of people is detected.

If all numbers of people that certain number of people that S43 has recorded detects with present frame all do not mate,The time in the moment that this number of people that calculates the present frame moment and recorded is detected for the last timePoor, if meet following formula (11), the information of this number of people having recorded is carried out to zero clearing processing:

frame_now-frame_i≥Thr_t(11)

Wherein, frame_nowRepresent the present frame moment, frame_iRepresent that certain number of people of having recordedAfter moment of being once detected, Thr_tRepresent predetermined threshold value.

S5, judge whether passenger vehicle overloads:

The number of times that in statistics a period of time, the same number of people is detected, judges whether overload, and judgement is publicFormula (12) is as follows:

$\left\{\begin{array}{c}{\mathrm{frame}}_i={\mathrm{frame}}_{now}\\ {\mathrm{score}}_i<Thr\end{array}\right.---\left(12\right)$

Wherein, frame_i＝frame_nowRepresent present frame statistics, score_iRepresent by present frameThe number of times that number of people i is detected, Thr represents predetermined threshold value.

If S6 passenger vehicle exists overloading, report to the police.

In the situation that passenger carriage overloading being detected, send this passenger vehicle in overload shape to remote serverThe information of state also sends a frame video image as checking. Administrative department receives the report for police service can be according to transmissionReal-time video processing, order passenger vehicle to stop overload behavior.

The above embodiment is only that the preferred embodiment of the present invention is described, notScope of the present invention is limited, design under the prerequisite of spirit this area not departing from the present inventionVarious distortion and improvement that those of ordinary skill is made technical scheme of the present invention, all should fall into thisIn the definite protection domain of claims of invention.

Claims (3)

1. a passenger carriage overloading intelligent monitoring detection method, is characterized in that, comprises the following steps:

(1) load the number of people and detect grader;

(2) area-of-interest detecting for the number of people, Real-time Obtaining sense delimited in passenger vehicle central gangwayThe sequence of video images of region-of-interest;

(3) adopt the number of people to detect grader, the video image of the area-of-interest to Real-time ObtainingCarry out number of people detection, and employing records with drag the people's header detecting:

Head_i＝{x_i,y_i,score_i,frame_i}

Wherein, Head_iRepresent the information of the number of people i of record, (x_i,y_i) represent the center of mass point of number of people i,score_iRepresent the number of times that number of people i is detected, frame_iRepresent the moment that number of people i is detected;

(4) people's header present frame being detected mates with the people's header recording,Upgrade the people's header having recorded;

(5) number of times that in statistics a period of time, the same number of people is detected, if exceed predetermined threshold value,Judge that passenger vehicle exists overload, carries out alarm.

2. passenger carriage overloading intelligent monitoring detection method according to claim 1, its feature existsIn, in step (1), the described number of people detects grader and is mainly trained and obtained by following steps:

A, collect some positive and negative training sample image, and to training sample image normalized;

The HOG feature of b, extraction training sample image,

C, by the HOG feature input Cascade cascade classifier of training sample image, it is right to completeThe number of people detects the training of grader.

3. passenger carriage overloading intelligent monitoring detection method according to claim 1, its feature existsIn, in step (4), described people's header that present frame is detected and the number of people letter recordingBreath mates, and upgrades the people's header having recorded, and comprises the following steps:

A, adopt following formula, calculate successively the number of people that present frame detects and the number of people having recorded itBetween match index:

$d_{ji}=\sqrt{{\left(x_{jnow}-x_i\right)}^2+{\left(y_{jnow}-y_i\right)}^2}$

Wherein, d_jiRepresent that mating between the present frame number of people j detecting and the number of people i recording refers toNumber, (x_jnow，y_jnow) represent the center of mass point of number of people j, (x_i，y_i) represent the center of mass point of number of people i;

If b is d_jiBe less than predetermined threshold value, judge the present frame number of people j detecting and the people who recordsI matches, and the two is the same number of people, adopts following formula to upgrade the information of the number of people i having recorded:

$\left\{\begin{array}{c}{x}_i=x_{jnow}\\ y_i=y_{jnow}\\ {\mathrm{score}}_i={\mathrm{score}}_i+1\\ {\mathrm{frame}}_i={\mathrm{frame}}_{now}\end{array}\right.$

Wherein, frame_nowRepresent the present frame moment;

If the number of people j that present frame detects does not all mate with all numbers of people that record, judge the number of peopleJ is a new number of people, adopts following formula to carry out record to it:

Head_new＝{x_new，y_new，score_new，frame_new}

$\left\{\begin{array}{c}{x}_{new}=x_{jnow}\\ y_{new}=y_{jnow}\\ {\mathrm{score}}_{new}=1\\ {\mathrm{frame}}_{new}={\mathrm{frame}}_{now}\end{array}\right.$

Wherein, Head_newRepresent the information of the new number of people of present frame record, (x_new，y_new) tableShow the center of mass point of the new number of people, score_newRepresent the number of times that the new number of people is detected, initial valueBe 1, frame_newRepresent the moment that the new number of people is detected;

If all numbers of people that certain number of people that c has recorded detects with present frame all do not mate, meterTime between the moment that this number of people of calculating the present frame moment and recorded is detected for the last timePoor, if reach predetermined threshold value, the information of this number of people having recorded is carried out to zero clearing processing.