CN107153819A - A kind of queue length automatic testing method and queue length control method - Google Patents

Abstract

The invention provides an automatic queue length detection method and a queue length control method, which detect pedestrians based on the human body structure model algorithm; use the feature online selection and promotion algorithm to track pedestrians; according to the detected people, use the least squares polynomial curve fitting based on RANSAC The method fits the queue shape of people, thereby estimating the length of the queue; realizes the transformation from the image coordinate system to the world coordinate system through image calibration, and calculates the actual queue length; calculates the curve length according to the curve fitting, and then according to the queue length and the queue length The distance between adjacent people, counting the number of people in the queue. The queuing length automatic detection method and queuing length control method of the present invention can quickly and accurately detect the queuing length of pedestrians in the monitoring scene, avoid the problem of target loss, and reduce missed detection; through real-time monitoring of the queuing situation, the processing window can be adjusted in time, thereby Effectively solve the problem of waiting in line and better serve passengers.

Description

A queue length automatic detection method and queue length control method

technical field

The invention relates to the technical field of image processing, in particular to a queue length automatic detection method and a queue length control method.

Background technique

In public environments such as airports, passengers often need to wait in line when checking in for boarding passes and entering security checkpoints. Due to the concentration of people, when the number of people is large, the waiting time for passengers in line will be prolonged, which will cause passengers' anxiety; at the same time, Passengers need to arrive at the airport a long time in advance to avoid flight delays, which greatly reduces the passenger experience. There are above problems in public places such as hospitals equally.

The existing solution is that the staff of the airport use manual guidance to reduce the waiting time of passengers in line; since manual guidance is completely dependent on the staff's control of the situation on the spot, it is time-consuming and laborious, and the efficiency is also quite low.

Therefore, how to better serve passengers, reduce the waiting time of passengers in line, improve airport work efficiency and reduce human resource expenses has become one of the problems to be solved urgently by those skilled in the art.

Contents of the invention

In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide an automatic queue length detection method and a queue length control method for solving the time-consuming and labor-intensive problem of manual grooming in the prior art.

In order to achieve the above purpose and other related purposes, the present invention provides a method for automatic detection of queue length, which at least includes:

Step S1: Obtain the pedestrian queuing video, and record the pedestrian queuing video;

Step S2: Detect pedestrians in the pedestrian queuing video based on the human structure model algorithm;

Step S3: After the pedestrian is detected, the feature value of the pedestrian image is continuously updated based on the feature online selection and promotion algorithm, so as to track and locate the pedestrian;

Step S4: according to the detected pedestrians, fitting the queue shape curve of pedestrians;

Step S5: Image calibration, converting image coordinates into world coordinates;

Step S6: Calculate the actual length of the queue and the distance between adjacent pedestrians according to the queue shape curve, so as to calculate the number of people in the queue.

Preferably, step S2 specifically includes: constructing a two-dimensional model based on pedestrian parts according to the human body structure, and identifying pedestrians by matching the structural features of pedestrian images with the two-dimensional model.

Preferably, the online promotion algorithm further includes: N feature selectors as strong classifiers; M features form a feature pool as weak classifiers; when a new sample arrives, N feature selectors are generated sequentially, each Each generation updates the cumulative correct sample weights and cumulative incorrect sample weights of M features, each feature selector uses the feature with the current minimum cumulative error rate as its corresponding weak classifier, and the combination of N feature selectors A strong classifier is formed, and the new position of the target is determined by evaluating with a strong classifier in the vicinity of the target position in the previous frame.

More preferably, the method for updating the accumulative correct sample weights and accumulative misclassified sample weights of the M features specifically includes:

When h _m (x) = y,

When h _m (x)≠y,

Among them, h _m , (m=1,...,M) is a feature, x is a window image with the same size as the target area, y is used to represent a positive sample or a negative sample, y=0 is a negative sample, and y=1 is a positive sample, For accumulating classification correct sample weights, is the cumulative weight of misclassified samples, λ is the current weight of the sample, and the initial value is 1.

More preferably, the strong classifier is formed by combining N feature selectors according to weights:

The weight α _n satisfies the following relationship:

in, is the feature selector, and ε _n is the cumulative error rate corresponding to feature n.

More preferably, the new position of the target satisfies the following relationship:

in, p _w (w=1,...,W) is the target candidate position within the search range,

is the feature selector, and α _n is the weight.

More preferably, the features in the online boosting algorithm include: Haar features, edge direction histogram features, and features of local binary patterns based on spatial block information.

More preferably, the calculation method of the Haar feature adopts integral histogram.

More preferably, the calculation method of the edge direction histogram feature is as follows: the direction θ of the gradient of the pixel point is quantized into multiple angle areas in the interval θ∈(-π, π], and each pixel point in the area The gradient is used for statistics, and the corresponding amplitude is added to the corresponding angle to obtain a histogram.

More preferably, the direction of the pixel gradient satisfies the following relationship:

Among them, the gradient magnitude is extracted in the horizontal direction the magnitude of the gradient in the vertical direction A is the sample image input, * is the two-dimensional convolution operation.

Preferably, the local binary pattern operators with different radii and sampling points are unioned to obtain the features of the local binary pattern based on spatial block information:

Among them, LBP is the characteristic value of the local binary pattern, P is the number of adjacent pixels, and R is the radius.

Preferably, the queue shape curve is fitted using a least squares polynomial curve fitting method based on RANSAC.

In order to achieve the above purpose and other related purposes, the present invention provides a queue length control method, the queue length control method at least includes:

Set up monitoring devices in occasions where queuing length control is required to collect videos of pedestrians queuing;

Adopt above-mentioned queuing length automatic detection method to detect queuing length;

When the queue length exceeds the preset number of people, an early warning message is sent to the client, and the client adjusts the queue to solve the problem of waiting in line.

Preferably, the method for the client to adjust the queuing queue includes: adjusting the service window, counting passenger flow, or guiding pedestrians to queue.

As mentioned above, the queue length automatic detection method and the queue length control method of the present invention have the following beneficial effects:

The queuing length automatic detection method and queuing length control method of the present invention can quickly and accurately detect the queuing length of pedestrians in the monitoring scene, avoid the problem of target loss, and reduce missed detection; through real-time monitoring of the queuing situation, the processing window can be adjusted in time, thereby Effectively solve the problem of waiting in line and better serve passengers.

Description of drawings

Fig. 1 is a schematic flow chart of the automatic queue length detection method of the present invention.

2 to 4 are schematic diagrams showing three local binary mode features of the present invention.

Fig. 5 is a schematic diagram of 8*8 patches in the local binary pattern algorithm based on spatial block information of the present invention.

FIG. 6 is a schematic diagram of a histogram of pixels in the local binary pattern algorithm based on spatial block information of the present invention.

Fig. 7 is a schematic diagram showing the calculation principle of the number of people in the queue of the present invention.

FIG. 8 is a schematic diagram of the queue length control method of the present invention.

Component designation description

r1～r3 sampling radius

S1～S6 steps

detailed description

Embodiments of the present invention are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation modes, and various modifications or changes can be made to the details in this specification based on different viewpoints and applications without departing from the spirit of the present invention.

Please refer to Figure 1 to Figure 7. It should be noted that the diagrams provided in this embodiment are only schematically illustrating the basic idea of the present invention, and only the components related to the present invention are shown in the diagrams rather than the number, shape and shape of the components in actual implementation. Dimensional drawing, the type, quantity and proportion of each component can be changed arbitrarily during actual implementation, and the component layout type may also be more complicated.

As shown in Figure 1, the present invention provides a kind of queue length automatic detection method, comprises the following steps:

Step S1: Acquire video of pedestrians queuing, and record the video of pedestrians queuing.

Specifically, the video of pedestrians queuing is obtained from the monitoring system, and the video of pedestrians queuing is recorded into the detection system of the present invention.

Step S2: Detecting pedestrians in the pedestrian queuing video based on a human structure model algorithm.

Specifically, a shape-based approach is used to construct a pedestrian detection algorithm based on a human body structure model. In this embodiment, pedestrians are identified by constructing a human body model. More specifically, construct a two-dimensional model based on pedestrian parts according to the structure of the human body, and match the two-dimensional model by extracting the structural features (human body structure) of the pedestrian image; If the corresponding human structure model matches, the pedestrian is detected; if some structural features of the extracted pedestrian image do not match the corresponding human structure model in the 2D model, the pedestrian is not detected. The method can effectively deal with the occlusion problem and can infer the pose of the human body.

Step S3: After the pedestrian is detected, the image features of the pedestrian are continuously updated based on the feature online selection and promotion algorithm, so as to track and locate the pedestrian. Through the online selection and promotion algorithm, the problem that the target was detected in the previous frame and the target was lost in the current frame is effectively solved, and the missed detection is reduced.

The strong classifier in the online boosting algorithm includes N feature selectors Each feature selector shares a feature pool, the feature pool includes M features h _m , (m=1, . . . , M), and the M features represent weak classifiers. The online boosting algorithm maintains and updates the cumulative classification correct sample weights of each feature all the time and misclassified sample weights In this way, the tracking of pedestrians is realized.

Specifically, after the target is located, the target area within the tracking window is used as a positive sample, and several background areas outside the tracking window and the size of the target are used as negative samples. When a new sample (x, y), y ∈ {0, 1} arrives, N feature selectors Generated sequentially, each generation updates M features h _m , (m=1,...,M), where the new sample (x, y) is the target area in the tracking window in the current frame image (positive sample) and some background areas (negative samples) of the same size as the target area outside the tracking window, x is a window image of the same size as the target area, y is used to represent a positive sample or a negative sample, y=0 is a negative sample, y= 1 is a positive sample:

When h _m (x) = y,

When h _m (x)≠y,

Among them, λ is the current weight value of the sample, and the initial value is 1.

After the update is complete, each feature selector will select the feature with the current minimum cumulative error rate as its corresponding weak classifier: (m ⁺ is the feature of the current minimum cumulative error rate), where ε _m is the cumulative error rate of each feature, and the weight corresponding to the cumulative error rate of feature n is ε _n is the cumulative error rate corresponding to feature n, The current minimum cumulative error rate.

nth feature selector The weight λ of the generated sample (x, y) is also increased or decreased according to whether it is misclassified. After N times of selection, the strong classifier is composed of the selected features according to their respective weights:

The new position of the target is determined by evaluating the range around the target position in the previous frame with a strong classifier. Suppose the search range is S, the target candidate position within the search range S is p _w (w=1,...,W), and the new target position is x _new , then Among them, w ⁺ is the candidate position with the largest total evaluation after passing through N strong classifiers, which satisfies the following formula:

in, That is, the feature selector that passes through the feature n at the candidate position p _w From the above formula, it can be seen that N feature selectors are required to evaluate W candidate positions during target positioning.

Specifically, in this embodiment, the features of the online selection and promotion algorithm include: Haar features, edge direction histogram features, and features of local binary patterns based on spatial block information, which are generated in the overall area of the target, through three The calculation of these features recognizes images with obvious rectangular characteristics, edge images and global textures, reducing errors and errors caused by a single feature.

More specifically, Haar features are classified into four categories: edge features, linear features, center features, and diagonal features, which are combined into feature templates. There are two kinds of rectangles, white and black, in the feature template, and the feature value of the template is defined as the sum of the pixels of the white rectangle minus the sum of the pixels of the black rectangle. The Haar eigenvalue reflects the grayscale variation of the image. For example, some features of the face can be simply described by rectangular features, such as: the eyes are darker than the cheeks, the sides of the bridge of the nose are darker than the bridge of the nose, and the mouth is darker than the surroundings. In this embodiment, the integral histogram is used to calculate the Haar feature, and the sum of pixels in all regions in the image can be obtained by traversing the image only once, which can greatly improve the efficiency of image feature value calculation. The main idea of the integral histogram is to save the sum of pixels in the rectangular area formed by the image from the starting point to each point as an array element in the memory. When calculating the pixel sum of a certain area, you can directly index the elements of the array. There is no need to recalculate the sum of pixels in this area, thus speeding up the calculation. The integral histogram can use the same time (constant time) to calculate different features at multiple scales, thus greatly improving the detection speed. Through a large number of object images with obvious Haar features (rectangles), the classifier is trained by pattern recognition. The classifier is a cascade, and each level retains the object features that enter the next level at approximately the same recognition rate. Candidate objects, and the sub-classifiers at each level are composed of many Haar features (calculated from the integral histogram, and save the position), there are horizontal, vertical, and inclined, and each feature has a threshold and two branch values, with an overall threshold for each level of sub-classifiers. When recognizing an object, the integral histogram is also calculated to prepare for the later calculation of the Haar feature, and then the window with the same size as the window with the object during training is used to traverse the entire image, and then the window is gradually enlarged, and the same traversal is performed to search for the object; Move the window to a position, that is, calculate the Haar feature in the window, compare it with the threshold of the Haar feature in the classifier after weighting to select the left or right branch value, and compare the accumulated branch value of one level with the threshold of the corresponding level, which is greater than the threshold Only then can they pass to the next round of screening.

More specifically, the extraction process of edge-oriented histogram features is as follows: Set the sample image input as A, use the edge operator to extract the gradient amplitude in the horizontal direction as G _x , and the gradient amplitude in the vertical direction as G _y , calculate The formula is as follows, where * is a two-dimensional convolution operation:

Define the magnitude GM _xy of the gradient at this pixel point to satisfy the following relationship:

Secondly, the direction θ of the gradient of the pixel is defined to satisfy the following relationship:

Finally, the direction θ of the pixel gradient is quantified into N _C angle areas in the interval θ∈(-π, π], and the gradient of each pixel point in the area is counted, and the corresponding amplitude is added to the corresponding angle And obtain the histogram. The eigenvalue of each statistical interval in the histogram is the edge feature we need, and it can be collected into the entire eigenvector.

More specifically, since texture features are less affected by illumination etc., local binary patterns are selected as local texture features in pedestrian areas. Local binary mode obtains feature values by analyzing the relationship between a pixel and its surrounding pixels. For adjacent pixels, the local binary mode of the central pixel is calculated by the following method: compare the value g _p of the adjacent pixel with the gray value g _c of the central pixel, if it is greater than the central pixel The gray value of the point is recorded as 1, and if it is smaller than the gray value of the center pixel, it is recorded as 0, and vice versa. Afterwards, starting from a certain fixed position, linking the binary data obtained in a clockwise direction to obtain a binary data string with a length of eight bits, converting it into decimal data, and then obtaining the local binary pattern characteristics of the pixel value. The calculation method is as follows:

Among them, LBP is the characteristic value of the local binary pattern, P is the number of adjacent pixels, R is the radius, g _p is the value of adjacent pixels, and g _c is the gray value of the central pixel.

The local binary pattern describes the local texture information of pedestrians well, but cannot describe the global information of pedestrians. For example, the pattern on pedestrian clothes will reduce the detection accuracy of pedestrians. Furthermore, in this embodiment, a local binary pattern algorithm based on spatial block information is proposed. In this embodiment, the patch size is set to 8*8. In actual use, the required value can be set according to the algorithm requirements. The size of the plaques is not limited to this embodiment.

As shown in Figures 2 to 4, in this embodiment, three local binary pattern operators with different radii and numbers of sampling points are connected to obtain the feature L of the local binary pattern based on spatial block information. As shown in Figure 2, the sampling radius of the first local binary pattern feature is r1, and the number of sampling points is 8; as shown in Figure 3, the sampling radius of the second local binary pattern feature is r2, and the number of sampling points is 12; As shown in FIG. 4 , the sampling radius of the third local binary mode feature is r3, and the number of sampling points is 16; where r1<r2<r3. So that for each point of the image, three different local binary pattern features LBP _{P, R} can be obtained, and the union of these features contains the most important texture information of the image:

As shown in Figure 5, the gradient of each pixel is calculated in the blocky area of the patch, and the magnitude and direction of the gradient are calculated, which is consistent with the method of calculating the magnitude and direction of the gradient in the edge direction histogram feature. This will not repeat them one by one. As shown in Figure 6, the histogram statistics of the pixels in 8 directions (0 to 360 degrees, one direction every 45 degrees) in the block area are counted, and the result of the histogram statistics is a feature vector (8-dimensional feature vector).

Step S4: According to the detected pedestrians, fit the queue shape curve of pedestrians.

Specifically, a RANSAC-based least squares polynomial curve fitting method is used to fit the queue shape curve. The RANSAC algorithm (Random Sample Consensus) is an algorithm that calculates the mathematical model parameters of the data based on a set of sample data sets containing abnormal data, and obtains effective sample data. It was first proposed by Fischler and Bolles in 1981. The basic assumption of the RANSAC algorithm is that the sample contains correct data (inliers, data that can be described by the model) and abnormal data (outliers, data that deviates far from the normal range and cannot adapt to the mathematical model), that is, the data set contains noise. These abnormal data may be generated due to wrong measurements, wrong assumptions, wrong calculations, etc. At the same time, RANSAC also assumes that, given a correct set of data, there is a method that can calculate the model parameters that conform to these data. The basic idea of RANSAC is described as follows: ①Consider a model with the potential of the minimum sampling set b (b is the minimum number of samples required to initialize the model parameters) and a sample set Q, the number of samples of the set Q#(Q)>b, from Randomly select a subset T of Q containing b samples from Q to initialize the model D; ②The residual set TC=the sample set in Q\T whose error with model D is less than a certain threshold t and T constitutes T*. T* is regarded as a set of interior points, which constitute the consensus set of T; ③ If #(T*)≥B, it is considered that the correct model parameters are obtained, and the set T* (inliers) is used to adopt least squares and other methods to recalculate the new model D*; re-randomly select a new T, and repeat the above process. ④ After completing a certain number of sampling times, if no consistent set is found, the algorithm fails, otherwise, the largest consistent set obtained after sampling is selected to judge the inner and outer points, and the algorithm ends.

Step S5: Image calibration, transforming image coordinates into world coordinates.

Step S6: Calculate the actual length of the queue and the distance between adjacent pedestrians according to the queue shape curve, so as to calculate the number of people in the queue.

As shown in Figure 7, the length of the curve is calculated according to the curve fitting, that is, the length L of the queue, and the number of people in the queue is calculated according to the length L of the queue and the distance d between adjacent people in the queue. The calculation formula is as follows:

Number of people = L/d.

The invention can perform real-time, effective, fast and accurate pedestrian queue length detection under different application scenarios, different climates and illumination conditions.

As shown in Figure 8, the present invention also provides a queue length control method, the queue length control method at least includes:

Set up monitoring devices in occasions that require queuing length control to collect video of pedestrians queuing.

Specifically, in places where queue length control is required, such as public environments such as airports and hospitals, monitoring devices such as monitoring probes are installed.

The above-mentioned queue length automatic detection method is adopted to detect the queue length.

Specifically, pedestrians are detected based on the human body structure model algorithm; pedestrians are tracked using the feature online selection and promotion algorithm, which mainly solves the problem that the target is detected in the previous frame but the target is lost in the current frame, and reduces missed detection. The feature of the feature online selection and promotion algorithm is Haar features, edge direction histogram features, and local binary pattern features based on spatial block information. The online feature selection method automatically selects expressive features according to the scene where the target is located to improve the tracking effect; according to the detected person, Use the least squares polynomial curve fitting method based on RANSAC to fit the shape of the queue to estimate the length of the queue; realize the transformation from the image coordinate system to the world coordinate system through image calibration, so as to calculate the actual length of the queue; according to the curve fitting Calculate the length of the curve, that is, the length of the queue, and then calculate the number of people in the queue based on the length of the queue and the distance between adjacent people in the queue. For specific methods, refer to the above, and details will not be repeated here.

When the queue length exceeds the preset number of people, an early warning message is sent to the client, and the client adjusts the queue to solve the problem of waiting in line.

Specifically, the method for the client to adjust the queuing queue includes: after the operation management center receives the warning information, it will increase the service window in time to ease the waiting time in the queue; Guide the pedestrians in the queue with many people to the queue with few people; thus effectively solve the problem of waiting in line and serve passengers better.

The invention can monitor the queuing situation in the boarding area and the security check area in real time, adjust the queuing queue in time, and effectively solve the problem of queuing and waiting.

In summary, the present invention provides an automatic queue length detection method and a queue length control method, which detect pedestrians based on the human body structure model algorithm; use the feature online selection and promotion algorithm to track pedestrians, and mainly solve the problem that the target detected in the previous frame is not detected in the current frame. The problem of loss, to reduce missed detection, feature online selection and promotion algorithm uses Haar feature, edge direction histogram feature and local binary mode feature based on spatial block information. The method of online feature selection depends on the scene where the target is located. Automatic selection of expressive features can improve the tracking effect; according to the detected people, the least squares polynomial curve fitting method based on RANSAC is used to fit the shape of the queue to estimate the length of the queue; the image coordinate system is realized through image calibration Convert to the world coordinate system to calculate the actual length of the queue; calculate the length of the curve based on curve fitting, that is, the length of the queue, and then calculate the number of people in the queue based on the length of the queue and the distance between adjacent people in the queue. The queuing length automatic detection method and queuing length control method of the present invention can quickly and accurately detect the queuing length of pedestrians in the monitoring scene, avoid the problem of target loss, and reduce missed detection; through real-time monitoring of the queuing situation, the processing window can be adjusted in time, thereby Effectively solve the problem of waiting in line and better serve passengers. Therefore, the present invention effectively overcomes various shortcomings in the prior art and has high industrial application value.

The above-mentioned embodiments only illustrate the principles and effects of the present invention, but are not intended to limit the present invention. Anyone skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical ideas disclosed in the present invention should still be covered by the claims of the present invention.

Claims (14)

1. a queue length automatic detection method is characterized in that, the queue length automatic detection method at least comprises: Step S1: Obtain the pedestrian queuing video, and record the pedestrian queuing video; Step S2: Detect pedestrians in the pedestrian queuing video based on the human structure model algorithm; Step S3: After the pedestrian is detected, the feature value of the pedestrian image is continuously updated based on the feature online selection and promotion algorithm, so as to track and locate the pedestrian; Step S4: according to the detected pedestrians, fitting the queue shape curve of pedestrians; Step S5: Image calibration, converting image coordinates into world coordinates; Step S6: Calculate the actual length of the queue and the distance between adjacent pedestrians according to the queue shape curve, so as to calculate the number of people in the queue. 2. The method for automatic detection of queue length according to claim 1, characterized in that: step S2 specifically comprises: constructing a two-dimensional model based on pedestrian components according to the human body structure, and matching the structural features of pedestrian images with the two-dimensional model , so as to identify pedestrians. 3. the queue length automatic detection method according to claim 1, is characterized in that: described online promotion algorithm further comprises: N feature selectors, as strong classifier; M features constitute a feature pool, as weak classifier ; When a new sample arrives, N feature selectors are generated sequentially, and each generation updates the cumulative classification correct sample weights and cumulative classification error sample weights of M features, and each feature selector uses the current minimum cumulative error rate The feature of is used as its corresponding weak classifier, and N feature selectors are combined to form a strong classifier. The new position of the target is determined by evaluating with the strong classifier in the vicinity of the target position in the previous frame. 4. the queue length automatic detection method according to claim 3, is characterized in that: the method for updating the accumulative classification correct sample weight and the accumulative classification error sample weight of M features specifically comprises: When h _m (x) = y, <mrow> <msubsup> <mi>&amp;lambda;</mi> <mi>m</mi> <mrow> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>r</mi> </mrow> </msubsup> <mo>=</mo> <msubsup> <mi>&amp;lambda;</mi> <mi>m</mi> <mrow> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>r</mi> </mrow> </msubsup> <mo>+</mo> <mi>&amp;lambda;</mi> </mrow> When h _m (x)≠y, <mrow> <msubsup> <mi>&amp;lambda;</mi> <mi>m</mi> <mrow> <mi>w</mi> <mi>r</mi> <mi>o</mi> <mi>n</mi> <mi>g</mi> </mrow> </msubsup> <mo>=</mo> <msubsup> <mi>&amp;lambda;</mi> <mi>m</mi> <mrow> <mi>w</mi> <mi>r</mi> <mi>o</mi> <mi>n</mi> <mi>g</mi> </mrow> </msubsup> <mo>+</mo> <mi>&amp;lambda;</mi> </mrow> Among them, h _m , (m=1,...,M) is a feature, x is a window image with the same size as the target area, y is used to represent a positive sample or a negative sample, y=0 is a negative sample, and y=1 is a positive sample, For accumulating classification correct sample weights, is the cumulative weight of misclassified samples, λ is the current weight of the sample, and the initial value is 1. 5. the queue length automatic detection method according to claim 3, is characterized in that: described strong classifier is formed by the combination of N feature selectors by weight: <mrow> <mi>H</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>s</mi> <mi>i</mi> <mi>g</mi> <mi>n</mi> <mrow> <mo>(</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>n</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msub> <mi>&amp;alpha;</mi> <mi>n</mi> </msub> <msubsup> <mi>h</mi> <mi>n</mi> <mrow> <mi>s</mi> <mi>e</mi> <mi>l</mi> </mrow> </msubsup> <mo>(</mo> <mi>x</mi> <mo>)</mo> <mo>)</mo> </mrow> </mrow> The weight α _n satisfies the following relationship: <mrow> <msub> <mi>&amp;alpha;</mi> <mi>n</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <mi>l</mi> <mi>n</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <mn>1</mn> <mo>-</mo> <msub> <mi>&amp;epsiv;</mi> <mi>n</mi> </msub> </mrow> <msub> <mi>&amp;epsiv;</mi> <mi>n</mi> </msub> </mfrac> <mo>)</mo> </mrow> </mrow> in, is the feature selector, and ε _n is the cumulative error rate corresponding to feature n. 6. the queue length automatic detection method according to claim 5, is characterized in that: target new position satisfies following relation: <mrow> <msub> <mi>x</mi> <mrow> <mi>n</mi> <mi>e</mi> <mi>w</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>p</mi> <msup> <mi>w</mi> <mo>+</mo> </msup> </msub> <mo>,</mo> </mrow> 1 in, p _w (w=1,...,W) is the target candidate position within the search range, is the feature selector, and α _n is the weight. 7. according to claim 1 or 3 described queuing length automatic detection methods, it is characterized in that: the feature in the described online promotion algorithm comprises: Haar feature, edge direction histogram feature and local binary pattern based on spatial block information Characteristics. 8. the queue length automatic detection method according to claim 7, is characterized in that: the calculating method of described Haar characteristic adopts integral histogram. 9. The queue length automatic detection method according to claim 7, characterized in that: the calculation method of the edge direction histogram feature is as follows: the direction θ of the pixel point gradient is quantified in the interval θ∈(-π, π] It is a multi-angle area, and the gradient of each pixel in the area is counted, and the corresponding amplitude is added to the corresponding angle to obtain a histogram. 10. the queue length automatic detection method according to claim 9, is characterized in that: the direction of described pixel point gradient satisfies following relation: <mrow> <mi>&amp;theta;</mi> <mo>=</mo> <mi>arctan</mi> <mrow> <mo>(</mo> <mfrac> <msub> <mi>G</mi> <mi>y</mi> </msub> <msub> <mi>G</mi> <mi>x</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> </mrow> Among them, the gradient magnitude is extracted in the horizontal direction the magnitude of the gradient in the vertical direction A is the sample image input, * is the two-dimensional convolution operation. 11. the queue length automatic detection method according to claim 1, is characterized in that: the local binary pattern operator with different radius and sampling point number is got union and obtains the feature based on the local binary pattern of spatial block information: <mrow> <mi>L</mi> <mo>=</mo> <munder> <mrow> <mi></mi> <mo>&amp;cup;</mo> </mrow> <mrow> <mi>P</mi> <mo>,</mo> <mi>R</mi> </mrow> </munder> <msub> <mi>LBP</mi> <mrow> <mi>P</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> </mrow> Among them, LBP is the characteristic value of the local binary pattern, P is the number of adjacent pixels, and R is the radius. 12. The queue length automatic detection method according to claim 1, characterized in that: the least squares polynomial curve fitting method based on RANSAC is used to fit the queue shape curve. 13. A queue length control method, characterized in that, the queue length control method at least comprises: Set up monitoring devices in occasions where queuing length control is required to collect videos of pedestrians queuing; Adopting the queue length automatic detection method as described in any one of claims 1 to 12 to detect the queue length; When the queue length exceeds the preset number of people, an early warning message is sent to the client, and the client adjusts the queue to solve the problem of waiting in line. 14. The queue length control method according to claim 13, characterized in that: the method for the client to adjust the queue comprises: adjusting the service window, counting passenger flow, or guiding pedestrians to queue.