WO2017142311A1

WO2017142311A1 - Multi-object tracking system and multi-object tracking method using same

Info

Publication number: WO2017142311A1
Application number: PCT/KR2017/001659
Authority: WO
Inventors: 곽정환; 전문구; 양이화
Original assignee: 광주과학기술원
Priority date: 2016-02-17
Filing date: 2017-02-15
Publication date: 2017-08-24
Also published as: KR20170097265A

Abstract

The present invention relates to a multi-object tracking system and a multi-object tracking method using the same. The multi-object tracking system comprises: a tracking camera for photographing at intervals of a predetermined unit time, moving objects to be tracked; a grouping module for receiving the photographed images from the tracking camera and grouping, into unit groups, the moving objects, located within a predetermined setting distance, from among the moving objects displayed in the images; and a tracking module for tracking a unit group including the moving object selected from among the unit groups on the basis of the images.

Description

Multiple Object Tracking System and Multiple Object Tracking Method Using the Same

The present invention relates to a multi-object tracking system and a multi-object tracking method using the same, and to a multi-object tracking system and a multi-object tracking method using the same by tracking a group of moving objects displayed on images captured by a tracking camera into unit groups. It is about.

Object tracking refers to accurately extracting the moving path of a moving object. In the past, many studies on multi-object tracking have been conducted. Most multi-object tracking systems consist mainly of data association framework, and various methods have been applied to solve data association problems.

In general, tracking people in complex environments such as shopping malls and campus streets is one of the difficult problems. This is because, in a complex environment, the accuracy of tracking a person is reduced due to a phenomenon such as occlusion. Conventionally, in the case of humans, a multi-view mulit-part model, that is, a model of a human body divided into parts such as a head-shoulder, a head-waist, a whole body, and the like is used for object tracking. come.

In the conventional data association technique, a comparison is performed in a pairwise manner with all objects existing in the screen of the previous image frame and all objects existing in the screen of the next frame. However, in general video environment, the movement of objects is continuous and has local proximity, which makes the comparison between objects far apart from each other quite inefficient.

In addition, the conventional tracking method is markedly inferior in tracking performance when the camera position is lowered or when the head is covered and when the human shape is half vertically covered.

The present invention was devised to improve the above-mentioned problems, and set moving objects adjacent to each other in images taken by a tracking camera into one unit group, track the movement of the unit group, and position the tracking camera. The purpose of the present invention is to propose a multi-object tracking system and a multi-object tracking method using the same, which can minimize interference with the obstruction phenomenon of moving objects.

The multi-object tracking system according to the present invention for achieving the above object is a tracking camera for photographing the tracking target moving objects at predetermined unit time intervals, and receiving the images taken from the tracking camera, the movement displayed on the images And a grouping module for grouping the moving objects located within a predetermined set distance among objects into a unit group, and a tracking module for tracking a unit group including the selected moving object among the unit groups based on the images.

The grouping module may include moving objects in the unit group spaced apart from each other further than the set distance with time, or at least one of the moving objects included in the unit group in another adjacent unit group. When the set moving object is adjacent to the set distance, the unit group grouped in the image is released, and the moving objects located within the set distance are regrouped into the unit group.

Meanwhile, the multi-object tracking system according to the present invention preferably further includes an object modeling module for modeling the moving objects of the unit group grouped by the grouping module according to a predetermined identification criterion.

The identification criteria include color histogram, histogram of gradient, and color self similarity.

When the moving object is a human, the object modeling module classifies the moving object into a plurality of unit parts according to body parts, and models the unit by the unit parts.

When the moving object is a human, the object modeling module includes the first unit part from the head to the shoulder, the second unit part from the head to the body, and the center line of the human body extending in the vertical direction. The third unit part, which is the left part, is divided into the fourth unit part, which is the right part of the whole body, based on the center line of the person extending in the vertical direction, and models the first to fourth unit parts.

When the tracking module determines that the unit groups selected from each of the images are similar according to predetermined similar information, it is preferable to connect the positions of the unit groups in each of the images to display the tracking path.

The similar information includes motion information of a moving object included in the selected unit group, appearance modeling information of moving objects included in the unit group modeled by the object modeling module, and information on a difference in photographing time between the images. Included.

Meanwhile, in the multi-object tracking method according to the present invention, a step of capturing images including a plurality of moving objects at intervals of a predetermined unit time through a tracking camera, and displayed on each of the images based on the images captured by the tracking camera. A grouping step of grouping the moving objects located within a predetermined set distance among the moving objects into a unit group, and after the grouping step is completed, a unit group including the moving object selected from the unit groups in each of the images; It includes a tracking step of tracking the movement path over time.

The grouping step may include an initial grouping step of grouping the moving objects located within a predetermined set distance among the moving objects in a unit group from one of the images photographed by the tracking camera, and after the initial grouping step is completed. And at least one of the moving objects included in the unit group of the initial image in the remaining ones of the images, spaced apart from each other further than the set distance, or included in the unit group of the initial image. When the moving object of adjacent to the moving object included in another adjacent unit group by the set distance, the group group is released, and regrouping the mobile objects located within the set distance into the unit group Regrouping step.

Meanwhile, the multi-object tracking method according to the present invention further includes an object modeling step of modeling the moving objects included in the unit group grouped in the grouping step between the grouping step and the tracking step according to a predetermined identification criterion. It is preferable.

In the object modeling step, when the moving object is a human, the moving object is divided into a plurality of unit parts according to body parts, and modeled for each unit part.

In the object modeling step, when the moving object is a human, the whole body is based on the first unit part from the head to the shoulder, the second unit part from the head to the body, and the center line of the person extending in the vertical direction. The third unit part, which is the left part of, is divided into the fourth unit part, which is the right part of the whole body, based on the center line of the person extending in the vertical direction, and modeled by the first to fourth unit parts.

In the tracking step, if it is determined that the unit group selected from each of the images is similar to each other according to predetermined similar information, the location of the unit group selected to each of the images is connected to each other and displayed as a tracking path.

The similar information includes motion information of a moving object included in the unit group selected from the images, appearance modeling information of the moving objects included in the unit group generated through the object modeling step, and elapsed time between the images. Include information about

The multi-object tracking system according to the present invention and the multi-object tracking method using the same set the moving objects adjacent to each other in the image photographed by the tracking camera as one unit group, and track the movement of the unit group from the selected moving object. Since the analysis of distant moving objects is not required, tracking time is saved and accuracy of tracking result value is improved.

In addition, in the multi-object tracking system and the multi-object tracking method using the same according to the present invention, when the moving object is a human model, the body is divided into a plurality of parts when modeling the moving object, thereby preventing interference with the obstruction phenomenon of the moving object. Minimized to improve the accuracy of the analysis.

1 is a block diagram of a multi-object tracking system according to the present invention;

2 is a conceptual diagram illustrating a grouping operation of a grouping module of the multi-object tracking system of FIG. 1,

3A and 3B are conceptual views illustrating a regrouping operation of a grouping module of a multi-object tracking system.

4 is a diagram illustrating a grouping operation of a grouping module by using an actual captured image.

5 is a conceptual diagram illustrating a plurality of unit part areas separated by an object modeling module of a multi-object tracking system according to the present invention;

6 is a conceptual diagram illustrating a motion trajectory of a tracklet displayed by a tracking module of a multi-object tracking system according to the present invention;

7 is a flowchart illustrating a multiple object tracking method according to the present invention.

The multi-object tracking system according to the present invention includes a tracking camera for capturing tracking target objects at predetermined unit time intervals, and a preset setting among the moving objects displayed on the images by receiving images captured by the tracking camera. And a grouping module for grouping the moving objects located within a distance into a unit group, and a tracking module for tracking a unit group including the selected moving object among the unit groups based on the images.

Hereinafter, a multi-mobile object tracking system and a multi-mobile object tracking method using the same will be described in detail with reference to the accompanying drawings. As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to a specific disclosed form, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the drawings, similar reference numerals are used for similar elements. In the accompanying drawings, the dimensions of the structures are shown in an enlarged scale than actual for clarity of the invention.

Terms such as first and second may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, action, component, part, or combination thereof described on the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

1 illustrates a multiple object tracking system 10 according to an embodiment of the present invention.

Referring to the drawing, the multi-object tracking system 10 receives the tracking camera 11 for capturing the tracking target moving objects at predetermined unit time intervals, and the images captured by the tracking camera 11 to receive the images. A grouping module 12 for grouping the moving objects located within a predetermined set distance among the moving objects displayed in the unit group, and the mobile objects of the unit group grouped by the grouping module 12 are identified. An object modeling module 13 for modeling according to a reference and a tracking module 14 for tracking a unit group including the moving object selected from the unit groups based on the images.

The tracking camera 11 is installed on an upper part of a building or a facility to photograph moving objects located in a predetermined area. The tracking camera 11 photographs the area at predetermined unit time intervals, and the photographed images are transmitted to the grouping module 12 through the communication module although not shown in the drawing.

Meanwhile, in the illustrated example, a structure in which one tracking camera 11 is provided is illustrated. However, the number of the tracking cameras 11 may not be limited to the illustrated example, but a plurality of tracking cameras 11 may be provided.

The grouping module 12 receives the images captured by the tracking camera 11 to group the moving objects. In this case, the grouping module 12 may be configured such that the moving objects in the unit group are spaced apart from each other farther than the set distance, or that at least one of the moving objects included in the unit group is adjacent to each other. When adjacent to the moving object included in the unit group by the set distance, the unit group grouped in the image is released, and the mobile objects located within the set distance are regrouped into the unit group.

FIG. 2 illustrates a grouping process of the grouping module 12. FIG. 3A illustrates a state in which the grouping module 12 resets one unit group to two unit groups according to movement of moving objects. 3B illustrates a state in which the grouping module 12 resets two unit groups to one unit group according to the movement of the moving objects.

In FIG. 2, images are arranged from left to right over time, and a moving object is displayed with dots having a predetermined color in each image. Referring to the image located on the leftmost side in FIG. 2, the moving objects located within a predetermined set distance among the moving objects are set as one unit group. That is, one moving object located at the bottom of the image is set as the first unit group, and two moving objects located above the first unit group are set as the second unit group, and three moving objects located above the second unit group. Mobile objects are set to the third unit group.

At this time, in the center image of FIG. 2, which is an image taken after the unit time has elapsed, the moving object of the first unit group was moved to the second unit group side as time passed, and the moving objects of the third unit group were moved. One of them moves to the second unit group side.

Next, referring to the rightmost image of FIG. 2, the moving objects of the first unit group of the leftmost image are adjacent by a set distance from the moving objects of the second unit group, and the grouping module 12 is illustrated in FIG. 3B. Similarly, the existing first unit group and the second unit group are released, and the moving objects included in the existing first unit group and the moving objects included in the existing second unit group are set as one new first unit group. . In addition, one of the moving objects included in the existing third unit group is spaced farther than the set distance from the other moving objects of the third unit group, the grouping module 12 as shown in Figure 3a The third unit group is released, and the new second unit group sets one moving object spaced farther than the set distance among the moving objects included in the existing third unit group, and the remaining one of the existing third unit groups Set the moving object to the new third unit group.

4 illustrates a state in which the grouping module 12 groups and regroups moving objects based on the actually photographed image. In FIG. 4, three images are arranged from left to right according to the photographed time sequence.

In the image (a) of FIG. 4, three moving objects on the left are set to one unit group A, two moving objects on the center are set to one unit group B, and one moving object on the right is It is set to one unit group (C).

Referring to images (b) and (c) of FIG. 4, as time elapses, one moving object located to the left of the moving objects included in the unit group A is farther than the set distance from the remaining moving objects. Spaced apart, the remaining moving objects were adjacent to the moving objects of the moving unit group (B). At this time, the grouping module 12 releases the unit group (A) and the unit group (B), sets one moving object located on the leftmost side of the image as a new unit group (A), and moves the four adjacent groups. Set the object to the new unit group (B).

The object modeling module 13 models moving objects included in a unit module grouped from the grouping module 12 and models the moving objects according to a predetermined identification criterion. The identification criteria include color histogram, histogram of gradient, and color self similarity.

In this case, when the moving object is a human, the object modeling module 13 divides the moving object into a plurality of unit parts according to body parts, and models the unit according to the unit parts. The unit parts are shown separately.

Referring to FIG. 5, the moving object is located on the left side of the whole body based on the first unit part (HS) from the head to the shoulder, the second unit part (HT) from the head to the body, and the center line of the person extending in the vertical direction. The third unit part LB, which is a part, is divided into a fourth unit part RB, which is a right part of the whole body, based on the center line of the person extending in the vertical direction, and models the first to fourth unit parts.

In this case, the moving object modeling of the object modeling module 13 will be described in more detail as follows. The moving object (r) displayed in the image

It can be expressed as Here, l is a label indicating a type selected from the first to fourth unit parts HS, HT, LB, and RB, a is an appearance model, p is a position of a moving object, t is the time in view of the image frame index.

In addition, the combined detection response (o) in the image

It is expressed as And the target hypothesis M is

The v denotes a visible part of the moving object. If v has no first to second unit parts (HS, HT, LB, RB), it means that the moving object is not visible. After acquiring the detection response, the visual part is processed, which calculates and updates the appearance model corresponding to each part of the moving object body.

In the appearance model, the above identification criteria of each moving object, namely, color histogram, histogram of gradients, and color self similarity are used.

The color histogram is for indicating the color appearance of the moving object, and a standard color histogram generally used in the art is applied. In the color histogram, an RGB color space is applied, but not limited thereto, and a Hue Saturation Value (HSV) and YCbCr may be applied.

Single channel histograms are single vectors with 8 bins for each channel.

Results in a 24-dimensional vector.

The 32-dimensional gradient histogram characteristic is associated with 8 orientations bins within a range 2x2 cell of a moving object.

) And 32-dimensional color self-similarity (

).

In this case, an appearance descriptor (a) of the selected tracklet tracklet T is expressed by Equation 1 below.

here

Is the characteristic vector of the color histogram,

Is a 32-dimensional gradient histogram characteristic vector,

Is a 32-dimensional color self-similarity characteristic vector based on the Hue Saturation Value (HSV). In addition, the subscript S is an index of the subregion of the moving object. The number of sub-regions is determined according to a label indicating a type selected from among the first to fourth unit parts HS, HT, LB, and RB. In the case of the whole body of a moving object including a third unit part (LB) and a fourth unit part (RB) for extracting a feature and configuring each corresponding feature pool for a human body partition, You can divide an object's area into nine subareas. The second to fourth unit parts HT, LB, and RB may be divided into four subregions, and the first unit part HS may have one subregion. In this case, three subtypes (featur type), namely, color histogram, histogram of gradients, and color self similarity are evaluated for each subregion. Thus, for example, a total of 27 feature vectors are generated for the whole body including the third and fourth unit parts constituting one feature pool. At this point, the similarity between the two image regions is calculated by the above described appearance descriptor (a).

To calculate the scores of RGB color histogram, gradient histogram and color self-similarity

And Bhattacharyya distance measures. The similarity measure of descriptors is

,

Is displayed.

Calculation of the similarity between the appearance descriptors in different areas for each label of the appearance model and the human body part, and then the similarity measurement vector shown in Equation 2 below by linking the similarity measurement.

Can be formed.

The feature vector provides a feature pool of the moving object for generating a tracklet. The characteristic vector is also used in a boosting algorithm for constructing a strong classifier.

Two cases

Wow

The strong model, which determines the affinity score of the appearance between, is designed as a linear combination of similar measures calculated by Equation 2. Therefore, the strong classifier is expressed by Equation 3 below.

A pair of cases is required to enter or return the actual value to distinguish positive pairs from negative pairs.

Is relatively large

Wow

It is easy to represent the same moving object. In the range [-1, + 1]

Is normalized. At this time,

The sign of the predicted label,

Interpret on scale.

The appearance model for each body part is calculated by offline processing. The boosting procedure is the same as Algorithm 1 described in Table 1 below.

At this time, the loss function of the boosting algorithm is shown in Equation 4 below.

Where w is the weight for each training sample to be updated while boosting. Goals are achieved by sequentially adding new weak classifiers.

Z is the least

To find.

optimal weak ranking classifier in round t

Can be found and weight

Equation 5 below may be calculated.

The detailed algorithm of the weak learner is the same as Algorithm 2 described in Table 2 below.

The tracking module 14 tracks the unit group including the selected moving object among the unit groups based on the images. At this time, if the analyst conducting the tracking selects one of the moving objects displayed on the image, the tracking module 14 selects a unit group including the selected moving object and connects the positions of the selected unit group in the images to track the path. Is displayed.

When the tracking module 14 determines that the unit groups selected from each of the images are similar according to predetermined similar information, it is preferable to connect the positions of the unit groups in each of the images and display them as a tracking path. . In this case, the similar information includes motion information of the moving object included in the selected unit group, appearance modeling information of the moving objects included in the unit group modeled by the object modeling module 13, and a photographing time between the images. It is desirable to include information about the differences.

The operation of the tracking module 14 will be described in more detail as follows.

First, tracklets included in the same unit group in different images are connected. In this case, pairwise linking confidence between a plurality of tracklets is determined by the motion information of the tracklets and the appearance modeling information of the moving objects included in the unit group modeled by the object modeling module 13. It is defined, and the appearance is shown in Equation 6 below.

Here, GM is a grouping module 12. The grouping module 12 returns a result value as true if a plurality of tracklets are connected based on the unit group information.

For the motion association measurement, first, the forward velocity and the backward velocity of the tracklet of the corresponding motion model are calculated as shown in FIG. 6.

The forward velocity is calculated from the refined position of the tail response of the tracklet while the backward velocity is calculated from the refined position of the head response of the tracklet.

In forward and backward directions, the motion model is Gaussian, respectively.

And Gaussian

Is represented by. And the association measurement of the motion is described as in Equation (7).

Where G is a zero-mean Gaussian function,

Is

Tail and

Is the time difference between the heads.

Wow

Is

The head or tail parts of the refined position and speed.

On the other hand, the correlation of time is expressed by a simple step function as shown in Equation 8 below.

Tail of

If it appears before the head of

Wow

It is possible to make connections between them.

The appearance model is obtained from a linear combination of several similarity measures of a set of local descriptors as described above. The appearance model is calculated for the part area of each object. The appearance model is then updated with detection response along the tracklet when the part is visible.

And the measurement of the association of the appearance model is described by the following equation (9).

After calculating the pairwise association of the tracklets using Equation 6, a cost matrix C is constructed, and a Hungarian algorithm is applied to obtain an optimal association. The cost matrix C is defined as in Equation 10 below.

here,

Is the pairwise association cost,

, Where

Is the terminating cost.

Is the cost to initialize to a new or fale track,

silver

Zero matrix.

Is the end of the tracklet, where N is the normalization factor, r is the missed detection rate, w is the size of the temporal sliding window, and s is the buffering of the object due to occlusion. The number of video frames.

Is the initialization period of the tracklet, where g is the video frame interval.

Is a false alarm period, where L is the total number of image frames.

Finally, an optimal assignment matrix obtained by applying the Hungarian algorithm to the cost matrix C

Indicates. And each

The object trajectory for is managed according to the assignment matrix using the rules detailed above. The rule will be described in detail as follows.

And

If,

Wow

Is connected,

And

If,

Is finished.

And

If,

when

Is the new track,

when,

Is a false track.

Algorithms for the operation of the tracking module 14 described above are described in Table 3 below.

Meanwhile, a multi-object tracking method using the multi-object tracking system 10 according to the present invention configured as described above will be described in detail with reference to FIG. 7. The multi-object tracking method includes a photographing step (S101), a grouping step (S102) (S102), an object modeling step (S103), and a tracking step (S104).

The photographing step (S101) is a step of capturing images including a plurality of moving objects at predetermined unit time intervals through the tracking camera 11. At this time, the image taken by the tracking camera 11, that is, the image is transmitted to the grouping module 12 through a communication module (not shown).

Grouping step (S102) is a step of grouping the moving objects located within a predetermined set distance of the moving objects displayed on each image based on the images taken from the tracking camera 11 in a unit group, the initial group step And a regrouping step.

The initial grouping step is a grouping of the moving objects located within a predetermined set distance among the moving objects in one of the images photographed by the tracking camera 11 into a unit group. The grouping module 12 groups the moving objects in the selected one of the images as described above.

In the regrouping step, after the initial grouping step is completed, the moving objects included in the unit group of the initial image are respectively spaced apart from each other farther than the set distance from the remaining ones of the images, or the unit of the initial image. When at least one of the moving objects included in the group is adjacent to the moving object included in another adjacent unit group by the set distance, the grouped unit group is released and the Re-grouping the moving objects into the unit group.

In the object modeling step S103, after the grouping step S102, the moving objects included in the unit group grouped in the grouping step S102 are modeled according to a predetermined identification criterion. In this case, as described above, the object modeling module 13 models according to predetermined identification criteria, that is, a color histogram, a histogram of gradients, and a color self similarity.

On the other hand, in the object modeling step (S103), when the moving object is a human, the object modeling module 13 divides the moving object into a plurality of unit parts according to the body part, and models for each unit part. In this case, as described above, the object modeling module 13 may be configured to move the moving object based on the first unit part from the head to the shoulder, the second unit part from the head to the body, and the center line of the person extending in the vertical direction. The third unit part, which is the left part, is divided into the fourth unit part, which is the right part of the whole body, based on the center line of the person extending in the vertical direction, and models the first to fourth unit parts.

In the tracking step S104, after the grouping step S102 is completed, the movement path is tracked over time in the unit group including the moving object selected from the unit groups in each of the images. In this case, when the analyst who tracks selects one of the moving objects displayed on the image, the tracking module 14 selects a unit group including the selected moving object, and the unit group selected from each of the images as described above. If they are determined to be similar according to the predetermined similarity information, it is preferable to connect the positions of the unit groups in each of the images and display them as a tracking path. In this case, the similar information includes motion information of a moving object included in the selected unit group, appearance modeling information of the moving objects included in the unit group modeled by the object modeling module 13, and a photographing time between the images. It is desirable to include information about the differences.

The multi-object tracking system 10 and the multi-object tracking method using the same according to the present invention configured as described above set moving objects adjacent to each other in images captured by the tracking camera 11 as one unit group. As the movement of the unit group is tracked, analysis of moving objects far from the selected moving object is not required, thus saving time required for tracking and improving accuracy of the tracking result value.

In addition, in the multi-object tracking system 10 and the multi-object tracking method using the same according to the present invention, when the moving object is a human, the model is divided into a plurality of parts when the moving object is modeled, thereby preventing the obstruction of the moving object. Interference is minimized to improve the accuracy of the analysis.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the scope of the invention. Thus, the present invention should not be limited to the embodiments set forth herein but should be construed in the broadest scope consistent with the principles and novel features set forth herein.

The present invention sets the moving objects adjacent to each other in the images taken by the tracking camera as a unit group, tracks the movement of the unit group, and prevents the interference of the obstruction of the moving objects according to the position of the tracking camera. Provides a multi-object tracking system that can be minimized and a multi-object tracking method using the same.

Claims

A tracking camera for capturing the tracking target objects at predetermined unit time intervals;

A grouping module for receiving images captured by the tracking camera and grouping the moving objects located within a predetermined set distance among the moving objects displayed on the images into a unit group;

And a tracking module for tracking a unit group including the selected moving object among the unit groups based on the images.
The method of claim 1,

The grouping module may be configured such that the moving objects in the unit group displayed in each image are spaced apart from each other further than the set distance or that at least one of the moving objects included in the unit group is adjacent to each other over time. When the moving object included in the unit group is adjacent to the set distance by the set distance, the multi-object, characterized in that for releasing the unit group grouped in the image, and regrouping the moving objects located within the set distance into a unit group Tracking system.
The method of claim 1,

And an object modeling module for modeling the moving objects of the unit group grouped by the grouping module according to a predetermined identification criterion.
The method of claim 3,

The identification criterion includes a color histogram, a histogram of gradients, and a color self similarity.
The method according to claim 3 or 4,

The object modeling module is a multi-object tracking system, characterized in that when the moving object is a human, the moving object is divided into a plurality of unit parts according to the body part, and modeled for each unit part.
The method of claim 5,

When the moving object is a human, the object modeling module includes the first unit part from the head to the shoulder, the second unit part from the head to the body, and the center line of the human body extending in the vertical direction. The third unit part, which is the left part, is divided into the fourth unit part, which is the right part of the whole body, based on the center line of the person extending in the vertical direction, and the multi-object tracking, characterized by modeling the first to fourth unit parts. system.
The method of claim 3,

If it is determined that the unit groups selected from each of the images are similar according to predetermined similarity information, the tracking module connects the positions of the unit groups in each of the images and displays them as a tracking path. Tracking system.
The method of claim 7, wherein

The similar information includes motion information of a moving object included in the selected unit group, appearance modeling information of moving objects included in the unit group modeled by the object modeling module, and information on a difference in photographing time between the images. Multi-object tracking system, characterized in that included.
Photographing images including a plurality of moving objects at predetermined unit time intervals through a tracking camera;

Grouping the moving objects located within a predetermined set distance among the moving objects displayed on the images based on the images photographed from the tracking camera into a unit group;

And after the grouping step is completed, a tracking step of tracking a moving path of the unit group including the moving object selected from the unit groups in each of the images as time passes. Way.
The method of claim 9,

The grouping step is

An initial grouping step of grouping the moving objects located within a predetermined set distance among the moving objects in a unit group in one of the images photographed by the tracking camera;

After the initial grouping step is completed, the moving objects included in the unit group of the initial image in the remaining ones of the images are spaced apart from each other further than the set distance or included in the unit group of the initial image. When at least one of the moving objects is adjacent to the moving object included in another adjacent unit group by the set distance, the unit group grouped in the image is released, and the And regrouping the mobile objects into the unit group.
The method of claim 9 or 10,

And an object modeling step of modeling the moving objects included in the unit group grouped in the grouping step between the grouping step and the tracking step according to a predetermined identification criterion.
The method of claim 11,

The identification criterion includes a color histogram, a histogram of gradients, and a color self similarity.
The method of claim 11,

In the object modeling step, when the moving object is a human, the moving object is divided into a plurality of unit parts according to the body part, multi-object tracking method characterized in that for modeling for each unit part.
The method of claim 11,

In the object modeling step, when the moving object is a human, the whole body is based on the first unit part from the head to the shoulder, the second unit part from the head to the body, and the center line of the person extending in the vertical direction. The third unit part, which is the left part of, is divided into the fourth unit part, which is the right part of the whole body, based on the center line of the person extending in the vertical direction, and is modeled for each of the first to fourth unit parts. Tracking method.
The method of claim 11,

In the tracking step, if it is determined that the unit group selected from each of the images is similar according to predetermined similar information, displaying the tracking path by connecting positions of the unit group selected to each of the images according to the time course. Characterized by multiple object tracking method.
The method of claim 15,

The similar information includes motion information of a moving object included in the unit group selected from the images, appearance modeling information of the moving objects included in the unit group generated through the object modeling step, and elapsed time between the images. Moving object tracking method comprising the information about.