CN107274395A - A kind of bus gateway head of passenger detection method based on empirical mode decomposition - Google Patents

Abstract

Belong to public transport video passenger's statistics application field, the invention discloses a kind of bus gateway head of passenger detection method based on empirical mode decomposition；First to image f_j(i) jth row image carries out empirical mode decomposition in, obtains the intrinsic mode function of jth row imageRecycle the intrinsic mode function of jth rowObtain the object function F of jth row_j(i)；Finally utilize the object function F_j(i) the threshold value t of image is obtained_q, utilize the threshold value t_qImage is split, the extraction to head of passenger in video is completed, this invention removes the influence of complex background in bus running, improves head of passenger identification and the degree of accuracy extracted and reliability, effectively eliminates the flase drop to passenger's clothes.

Description

A kind of bus gateway head of passenger detection method based on empirical mode decomposition

Technical field

The present invention relates to the extraction of moving target in video analysis, particularly a kind of bus based on empirical mode decomposition Gateway head of passenger detection method, for the extraction to head of passenger in public transit system monitor video.

Background technology

The key that bus passenger number is accurately counted is to correctly identify out from video sequence and feels emerging in image The target of interest, the lasting observation of the motion state progress to the target afterwards, judges the number and the direction of motion of target, most whereby Pass through the statistics for being provided with number of counting rule afterwards.The mode of one good extraction target can greatly improve target with The efficiency of the subsequent treatment such as track and goal behavior understanding.

Due to image and video sequence be vulnerable to such as illumination, temperature, shade factor influence so that moving target Extracting turns into an extremely difficult job.Conventional Moving Object Recognition has time differencing method, optical flow method, based on study Method, background subtraction etc..

Time differencing method is exactly that moving target gray scale has notable change between two adjacent frames using in the video sequence Change, and the little principle of static target grey scale change, the two continuous frames pixel grey scale in video sequence is subjected to calculus of differences, led to The region of threshold decision target is crossed, a kind of method of the detection of moving target is thereby realized.This method amount of calculation is few, detection Speed is fast, is easily achieved and is not influenceed by illumination factor.But for the moving target of pure color, this method can only be detected The edge of target, the segmentation of image is extracted after this have impact on.Time differencing method can only also recognize the target of motion simultaneously, it is impossible to Static target is detected, therefore after target halts locomotion, time differencing method will be unable to follow the trail of target, and this can cause target to chase after The loss of track.

Moving object detection based on optical flow method is time domain change and the correlation for utilizing the pixel grey scale in image sequence To determine the direction of motion of each pixel, i.e., by studying the change of gradation of image in time, distinguish the prospect of target with Background.While the advantage of optical flow method is light stream independent of background, the movable information of moving target is also carried, and In the case that camera is moved, Moving Objects can be still detected, the stability of algorithm is very high.But moved with optical flow method Target detection calculates time-consuming, and real-time and practicality are all poor.Optical flow method is excessively poor to the ability of the disturbance of light simultaneously, and Static target can not be detected.

Recognition methods based on machine learning by carrying out feature extraction to foreground target to be detected, set up positive sample and Negative sample, passes through the training to sample so that grader can be recognized and sample characteristics as target class to be detected.To target When being identified with following the trail of, the feature of image in slidably search window, calculation window is directly set up on image, is then utilized Grader examines whether these windows are prospect class, and then the center of slip scan window is run again to next position, with This realizes the identification and tracking to foreground target.Recognition methods based on machine learning is highly dependent on training and the feature of sample The selection of point, while this method amount of calculation is very big, operational efficiency is not high.

Background subtraction is, by present image or image sequence and background image progress calculus of differences, then to be selected by threshold value Select a kind of moving target detecting method in the region for extracting moving target.Under appropriate conditions, this method results in motion The region of target and more complete profile, it is adaptable to the target dynamic detection under video camera static position, in video monitoring Field is a kind of conventional target detection technique.But background subtraction is easily influenceed by light weather, therefore is easily multiplied Objective body shade and light intensity influence.

Existing common methods all can not accurately and accurately extract the head of passenger in monitoring, hold in detection process It is vulnerable to illumination, passenger's clothes, the influence of passenger's body shade.This is caused above and below the bus monitoring using the number of people as digit Missing inspection, false drop rate are high in car statistical system, the situation that system is low to illumination robustness, running efficiency of system is low.

The content of the invention

Based on above technical problem, the invention provides a kind of bus gateway passenger's head based on empirical mode decomposition Portion's detection method, is solved because of missing inspection, mistake in passenger getting on/off statistical system caused by illumination, passenger's clothes, passenger's body shade Inspection rate is high, low to illumination robustness and the low technical problem of running efficiency of system.

The technical solution adopted by the present invention is as follows：

A kind of bus gateway head of passenger detection method based on empirical mode decomposition, comprises the following steps：

Step 1：To image f_j(i) jth row image carries out empirical mode decomposition in, obtains the natural mode of vibration of jth row image FunctionWherein j represents the row sequence number of image, and i represents the row sequence number of image, and p represents the rank number of intrinsic mode function；

Step 2：Utilize the intrinsic mode function of jth rowObtain the object function F of jth row_j(i)；

Step 3：Utilize the object function F_j(i) the threshold value t of image is obtained_q, utilize the threshold value t_qImage is divided Cut.

Further, step 1 is comprised the following steps that：

S201：Gray processing is carried out to image, the gray value of each pixel of image is obtained, the average gray value of image is calculated, The gray value of the pixel of each in image is subtracted into the average gray value, the image f after being handled_j(i)；

S202：Determine image f_j(i) maximum and minimum of jth row in, formula are as follows：

Maximum：f_j(i)-f_j(i)>0&&f_j(i+1)-f_j(i)<0 (1),

Minimum：f_j(i)-f_j(i)<0&&f_j(i+1)-f_j(i)>0 (2)；

S203：The coenvelope line of maximum formation and the lower bag of minimum formation in S202 are tried to achieve using cubic spline interpolation Winding thread, calculates the average of the coenvelope line and lower envelope lineWherein p represents the rank number of intrinsic mode function；

S204：Order

JudgeWhether satisfaction turns into the condition of intrinsic mode function, if meeting, makesIf no Meet, then makeRepeat step S202-S204；Wherein n isCycle-index；

S205：Order

And make

Repeat step S202-S205 untilFor monotonic sequence or only one of which extreme point, the K of jth row image is obtained Rank intrinsic mode functionWherein p ∈ [1, P], P are total exponent number of mode function.

Further, object function F_j(i) calculation formula is as follows：

Wherein m represents that the intrinsic mode function in the pth rank in P ranks, p ∈ [m, P] is low frequency function.

Further, threshold value t_qDetermination process it is as follows：

S401：Determine the object function F of jth row image_j(i) maximum and minimum, the formula of use are as follows：

Maximum：F_j(i)-F_j(i)>0&&F_j(i+1)-F_j(i)<0 (7),

Minimum：F_j(i)-F_j(i)<0&&F_j(i+1)-F_j(i)>0 (8),

The object function F_j(i) there is Q in adjacent extreme value, the adjacent extreme value includes a maximum and minimum； Wherein q is f1 to the gray value of maximum in adjacent extreme value, and the gray value of minimum is f2, the row serial number of the maximum , there is interval, q ∈ [1, Q] between wherein i1 and i2 in i1, the row serial number i2 of the minimum；

S402：The slope k between adjacent extreme value is obtained using the gray value and row sequence number of adjacent extreme value_q, formula is such as：

The slope k_qMaximum

S403：Utilize the slope k_qThreshold value t is set_q, formula is：

t_q=ε (f1-f2)+f2 (0<ε<1) (11),

Wherein, k is worked as_q<μk_max(0<μ<1) when, 0<ε≤0.5；k_q>μk_max(0<μ<1) when, 0.5<ε≤1；

Work as f1-f2<During T, t_q=t_q-1, wherein T is the fluctuation threshold of setting；

S404：Utilize the threshold value t_qRow threshold division is entered to the pixel between adjacent extreme value, works as f_j(i)>t_qWhen, order f_j(i)=255；Work as f_j(i)≤t_qWhen, make f_j(i)=0.

In summary, by adopting the above-described technical solution, the beneficial effects of the invention are as follows：

1st, the grey-level sequence of image is become gentle using empirical mode decomposition, and retains topmost feature in image, Retain the feature of head part in image, eliminate the irrelevant informations such as most shade in image, passenger's clothes.

2nd, the intrinsic mode function produced using empirical mode decomposition builds object function, and determines figure using object function As the threshold value of segmentation, that is, the larger place of grey scale change in image is determined, the Image Segmentation Methods Based on Features of head part in image is come out, enters one Step eliminates the irrelevant information in image.

3rd, the Threshold Segmentation Algorithm that the present invention is provided is Local threshold segmentation method, can effectively filter out less gray scale ripple Dynamic, operand is small, accuracy is high, real-time is good, analyzed in real time suitable for video.

4th, the present invention can effectively remove passenger's clothes, shade, illumination and extract the influence that brings to personage's head feature, subtract False drop rate and loss that small head is extracted, while having good robustness to illumination, running efficiency of system is high.

Brief description of the drawings

Fig. 1 is the flow chart of invention；

Fig. 2 is the intensity profile figure before and after jth row image progress empirical mode decomposition；

Fig. 3 is the intensity profile figure before and after entire image progress empirical mode decomposition；

Fig. 4 is the artwork for image；

Fig. 5 is the detection figure of passenger's head portrait；

Fig. 6 is the image after being handled in video sequence using general Threshold Segmentation Algorithm；

Fig. 7 is the comparison diagram after being handled using background subtraction with the inventive method same two field picture in video sequence；

Fig. 8 is to handle obtained image to video sequence progress using the present invention.

Embodiment

All features disclosed in this specification, can be with any in addition to mutually exclusive feature and/or step Mode is combined.

The present invention is elaborated below in conjunction with the accompanying drawings.

A kind of bus gateway head of passenger detection method based on empirical mode decomposition, comprises the following steps：

Step 1：Step 1 is comprised the following steps that：

S201：Gray processing is carried out to image, the gray value of each pixel of image is obtained, the average gray value of image is calculated, The gray value of the pixel of each in image is subtracted into the average gray value, the image f after being handled_j(i)；

S202：Determine image f_j(i) maximum and minimum of jth row in, formula are as follows：

Maximum：f_j(i)-f_j(i)>0&&f_j(i+1)-f_j(i)<0 (12),

Minimum：f_j(i)-f_j(i)<0&&f_j(i+1)-f_j(i)>0 (13)；

S203：The coenvelope line of maximum formation and the lower bag of minimum formation in S202 are tried to achieve using cubic spline interpolation Winding thread, calculates the average of the coenvelope line and lower envelope lineWherein p represents the rank number of intrinsic mode function；

S204：Order

JudgeWhether satisfaction turns into the condition of intrinsic mode function, and the condition is：

(1)Extreme point and zero crossing number it is equal or difference one；

(2) at any time,Difference with zero is less than 0.1；

If meeting the condition, makeIf being unsatisfactory for the condition, makeRepeat Step S202-S204；Wherein n isCycle-index；

S205：Order

And make

Repeat step S202-S205 untilFor monotonic sequence or only one of which extreme point, the K of jth row image is obtained Rank intrinsic mode functionWherein p ∈ [1, P], P are total exponent number of mode function.

Step 2：Utilize the intrinsic mode function of jth rowObtain the object function F of jth row_j(i)；

Object function F_j(i) calculation formula is as follows：

Wherein m represents that the intrinsic mode function in the pth rank in P ranks, p ∈ [m, P] is low frequency function.

Step 3：Utilize the object function F_j(i) the threshold value t of image is obtained_q, utilize the threshold value t_qImage is divided Cut, threshold value t_qDetermination process it is as follows：

S401：Determine the object function F of jth row image_j(i) maximum and minimum, the formula of use are as follows：

Maximum：F_j(i)-F_j(i)>0&&F_j(i+1)-F_j(i)<0 (18),

Minimum：F_j(i)-F_j(i)<0&&F_j(i+1)-F_j(i)>0 (19),

The object function F_j(i) there is Q in adjacent extreme value, the adjacent extreme value includes a maximum and minimum； Wherein q is f1 to the gray value of maximum in adjacent extreme value, and the gray value of minimum is f2, the row serial number of the maximum , there is interval, q ∈ [1, Q] between wherein i1 and i2 in i1, the row serial number i2 of the minimum；

S402：The slope k between adjacent extreme value is obtained using the gray value and row sequence number of adjacent extreme value_q, formula is such as：

The slope k_qMaximum

S403：Utilize the slope k_qThreshold value t is set_q, formula is：

t_q=ε (f1-f2)+f2 (0<ε<1) (22),

Wherein, k is worked as_q<μk_max(0<μ<1) when, 0<ε≤0.5；k_q>μk_max(0<μ<1) when, 0.5<ε≤1；

Work as f1-f2<During T, t_q=t_q-1, wherein T is the fluctuation threshold of setting；

S404：Utilize the threshold value t_qRow threshold division is entered to the pixel between adjacent extreme value, works as f_j(i)>t_qWhen, order f_j(i)=255；Work as f_j(i)≤t_qWhen, make f_j(i)=0.

It is embodiments of the invention as described above.The present invention is not limited to the above-described embodiments, anyone should learn that The structure change made under the enlightenment of the present invention, the technical schemes that are same or similar to the present invention each fall within this Within the protection domain of invention.

Claims (4)

1. a kind of bus gateway head of passenger detection method based on empirical mode decomposition, it is characterised in that：Including following Step：

Step 1：To image f_j(i) jth row image carries out empirical mode decomposition in, obtains the intrinsic mode function of jth row imageWherein j represents the row sequence number of image, and i represents the row sequence number of image, and p represents the rank number of intrinsic mode function；

Step 2：Utilize the intrinsic mode function of jth rowObtain the object function F of jth row_j(i)；

Step 3：Utilize the object function F_j(i) the threshold value t of image is obtained_q, utilize the threshold value t_qImage is split.

2. a kind of bus gateway head of passenger detection method based on empirical mode decomposition according to claim 1, It is characterized in that:Step 1 is comprised the following steps that：

S201：Gray processing is carried out to image, the gray value of each pixel of image is obtained, the average gray value of image is calculated, will scheme The gray value of each pixel subtracts the average gray value as in, the image f after being handled_j(i)；

S202：Determine image f_j(i) maximum and minimum of jth row in, formula are as follows：

Maximum：f_j(i)-f_j(i)>0&&f_j(i+1)-f_j(i)<0,

Minimum：f_j(i)-f_j(i)<0&&f_j(i+1)-f_j(i)>0；

S203：The coenvelope line of maximum formation and the lower envelope of minimum formation in S202 are tried to achieve using cubic spline interpolation Line, calculates the average of the coenvelope line and lower envelope lineWherein p represents the rank number of intrinsic mode function；

S204：OrderJudgeWhether satisfaction turns into the condition of intrinsic mode function, if meeting, Then makeIf it is not satisfied, then makingRepeat step S202-S204；Wherein n isCirculation Number of times；

S205：OrderAnd makeRepeat step S202-S205 untilFor dullness Sequence or only one of which extreme point, obtain the K rank intrinsic mode functions of jth row imageWherein p ∈ [1, P], P is mode Total exponent number of function.

3. a kind of bus gateway head of passenger detection method based on empirical mode decomposition according to claim 1, It is characterized in that:Object function F_j(i) calculation formula is as follows：

<mrow> <msub> <mi>F</mi> <mi>j</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>=</mo> <msubsup> <mi>&amp;Sigma;</mi> <mrow> <mi>p</mi> <mo>=</mo> <mi>m</mi> </mrow> <mi>P</mi> </msubsup> <msubsup> <mi>g</mi> <mi>j</mi> <mi>p</mi> </msubsup> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow>

Wherein m represents that the intrinsic mode function in the pth rank in P ranks, p ∈ [m, P] is low frequency function.

4. a kind of bus gateway head of passenger detection method based on empirical mode decomposition according to claim 1, It is characterized in that:Threshold value t_qDetermination process it is as follows：

S401：Determine the object function F of jth row image_j(i) maximum and minimum, the formula of use are as follows：

Maximum：F_j(i)-F_j(i)>0&&F_j(i+1)-F_j(i)<0,

Minimum：F_j(i)-F_j(i)<0&&F_j(i+1)-F_j(i)>0,

The object function F_j(i) there is Q in adjacent extreme value, the adjacent extreme value includes a maximum and minimum；Wherein Q is f1 to the gray value of maximum in adjacent extreme value, and the gray value of minimum is f2, the row serial number i1 of the maximum, institute The row serial number i2 of minimum is stated, there is interval, q ∈ [1, Q] between wherein i1 and i2；

S402：The slope k between adjacent extreme value is obtained using the gray value and row sequence number of adjacent extreme value_q, formula is such as：

<mrow> <msub> <mi>k</mi> <mi>q</mi> </msub> <mo>=</mo> <mfrac> <mrow> <mi>f</mi> <mn>1</mn> <mo>-</mo> <mi>f</mi> <mn>2</mn> </mrow> <mrow> <mi>i</mi> <mn>1</mn> <mo>-</mo> <mi>i</mi> <mn>2</mn> </mrow> </mfrac> <mo>,</mo> </mrow> 1

The slope k_qMaximum

S403：Utilize the slope k_qThreshold value t is set_q, formula is：

t_q=ε (f1-f2)+f2 (0<ε<1)

Wherein, k is worked as_q<μk_max(0<μ<1) when, 0<ε≤0.5；k_q>μk_max(0<μ<1) when, 0.5<ε≤1；

Work as f1-f2<During T, t_q=t_q-1, wherein T is the fluctuation threshold of setting；

S404：Utilize the threshold value t_qRow threshold division is entered to the pixel between adjacent extreme value, works as f_j(i)>t_qWhen, make f_j(i) =255；Work as f_j(i)≤t_qWhen, make f_j(i)=0.