KR102538919B1 - A Hybrid Object Tracking System and Method therefor - Google Patents

Abstract

The present invention relates to a system and method for tracking an object, and relates to an object information receiver for receiving information about an object to be tracked from video data consisting of a plurality of frames, a position of the object to be tracked in a previous frame of the video data and a current frame. A non-recognition tracking unit that tracks the movement of the object to be tracked using a non-recognition tracker, an object recognition tracker that recognizes the object to be tracked for each frame of the video data and tracks the movement of the object, and tracking of the object recognition tracker An object tracking system including a comprehensive tracking unit generating a comprehensive tracking result based on a result and a tracking result of the non-recognized tracking unit, and a method for tracking the object.

Description

Hybrid Object Tracking System and Method {A Hybrid Object Tracking System and Method therefor}

The present invention relates to a system and method for tracking an object, and relates to a system and method for enabling efficient and accurate movement tracking of an object in the process of tracking an object selected in an image.

As technologies for analyzing images and providing services based thereon increase, demand for technologies for recognizing a specific object in an image and tracking movement of the recognized object in the image is increasing.

Among conventional object tracking technologies, an object detection based tracking technology is widely used, which continuously recognizes a region having characteristics of an object to be tracked in each of a plurality of frames constituting an image. By doing so, it is a technology that tracks how the object moves according to the flow of time series in the image.

Prior art Korean Patent Publication No. 10-2016-0123647, "Apparatus and Method for Providing Video Additional Information Using Object Tracking" is a technique for providing various services using such object recognition-based tracking technology. However, in the case of using such object recognition tracking technology, it is necessary to pre-learn the object in order to understand the characteristics of the object to be recognized, and when the same object changes direction on the screen, it may be difficult to recognize it. There is a problem with being

In order to solve this problem, a non-recognition tracking technology that does not use object recognition has recently been developed. This technology, also called Visual Object Tracking (VOT), compares images of the current frame and the previous frame. , This is a technology that checks how an object in the previous frame moved in the current frame based on artificial intelligence. It has excellent tracking performance for the movement of an object, but it can be detected if tracking is missed (e.g., when an object goes out of the screen and then comes back in). There are problems that cannot be restored.

Korean Patent Publication No. 10-2016-0123647

An object of the present invention is to effectively track the movement of an object by using non-recognition tracking technology, and to track the movement of an object again even if the tracking of the object is stopped, such as when the object goes out of the screen and then returns. do.

An object of the present invention is to enable effective object tracking by supplementarily using an object recognition tracking method when the reliability is low while tracking an object using a non-recognition tracking method.

An object of the present invention is to enable effective object tracking with minimal computation by enabling an object recognition tracking method to be used only when the reliability of the non-recognition tracking method is low based on a reliability value.

An object of the present invention is to maximize tracking performance by generating a comprehensive tracking result by assigning a weight to the tracking result of each method based on a reliability value when both the non-recognition tracking method and the object recognition tracking method are simultaneously applied.

In order to achieve this object, an object tracking system according to an embodiment of the present invention includes an object information receiving unit receiving information about an object to be tracked from video data composed of a plurality of frames, and the tracking object in a previous frame of the video data. A non-recognition tracking unit that tracks the movement of the object to be tracked using the location of the object and the current frame, an object recognition tracker that recognizes the object to be tracked for each frame of the video data and tracks the movement of the object; and and a comprehensive tracking unit generating a comprehensive tracking result based on a tracking result of the object recognition tracking unit and a tracking result of the non-recognition tracking unit.

At this time,

The object recognition tracking unit and the non-recognition tracking unit each calculate a tracking reliability value, and the comprehensive tracking unit obtains weights of a tracking result of the object recognition tracking unit and a tracking result of the non-recognition tracking unit using the tracking reliability value, respectively. , a comprehensive tracking result can be generated by reflecting this.

In addition, the comprehensive tracking unit obtains a weight of a tracking result of the non-recognition tracking unit based on the tracking reliability value of the non-recognition tracking unit, and based on a difference between the tracking reliability value of the object recognition tracking unit and the tracking reliability value of the non-recognition tracking unit A weight of a tracking result of the object recognition tracking unit may be obtained.

In addition, the object recognition tracker tracks the movement of the object to be tracked only when the tracking reliability value of the non-recognition tracker is equal to or less than a predetermined reference value, and the comprehensive tracker tracks the movement of the object recognition tracker when the object recognition tracker does not track the movement. The tracking result of the recognition tracking unit may be generated as a comprehensive tracking result.

In addition, the object recognition tracking unit determines whether the object to be tracked is a learned type of object based on the information about the object to be tracked, tracks the movement of the object to be tracked only when the object is a learned type of object, and The tracking unit may generate a tracking result of the non-recognition tracking unit as a comprehensive tracking result when the object recognition tracking unit does not track movement.

The present invention uses a non-recognition tracking technology to effectively track the movement of an object, and to track the movement of an object again even when tracking of the object is stopped, such as when the object leaves and then returns to the screen.

The present invention enables effective object tracking by supplementarily using an object recognition tracking method when the reliability is low while tracking an object using a non-recognition tracking method.

According to the present invention, the object recognition tracking method is used only when the reliability of the non-recognition tracking method is low based on the reliability value, thereby enabling effective object tracking with minimal computation.

When the non-recognition tracking method and the object recognition tracking method are simultaneously applied, the present invention maximizes tracking performance by generating a comprehensive tracking result by assigning a weight to the tracking result of each method based on a reliability value.

1 is a configuration diagram showing the internal configuration of an object tracking system according to an embodiment of the present invention.
2 is a diagram illustrating an example of a case where non-recognition tracking and object recognition tracking are applied according to an embodiment of the present invention.
3 is a flow illustrating the flow of an object tracking method according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description will be omitted. In addition, in describing the embodiments of the present invention, specific numerical values are only examples, and the scope of the invention is not limited thereby.

The object tracking system according to the present invention may be configured in the form of a server having a central processing unit (CPU) and a memory and being connectable to other terminals through a communication network such as the Internet. In addition, it may be implemented in a mobile terminal, an Internet of Things (IoT) device, etc., and independent operation is possible even in an environment where a communication network such as the Internet is not connected. However, the present invention is not limited by the configuration of the central processing unit and memory. In addition, the object tracking system according to the present invention may be physically configured as one device or implemented in a distributed form in a plurality of devices.

1 is a configuration diagram showing the internal configuration of an object tracking system according to an embodiment of the present invention.

As shown in the drawing, the object tracking system 101 according to an embodiment of the present invention includes an object information receiving unit 110, a non-recognition tracking unit 120, an object recognition tracking unit 130, and a comprehensive tracking unit 140. ). Each component may be a software module that operates in the same computer system physically, or may be configured so that two or more physically separated computer systems can operate in conjunction with each other. Embodiments belong to the scope of the present invention.

The object information receiving unit 110 receives information about an object to be tracked from video data composed of a plurality of frames. Moving image data is provided by arranging a plurality of still image frames in time series and reproducing the still image quickly according to the lapse of time, thereby providing a moving image. In order to analyze such video data, it is necessary to recognize objects included in the video. For example, when a CCTV detects an intruder, the intruder must be selected as an object to be tracked and movement must be identified.

The information about the object to be tracked received by the object information receiving unit 110 may include location information of where the object to be tracked is located in the image or range information indicating the range in which the object to be tracked is located, and may further include properties of the object to be tracked. information may be included. For example, when an object to be tracked is a face of a person, by providing attribute information that the object to be tracked is a face, more accurate recognition is possible when tracking an object through object recognition.

In order to receive information about an object to be tracked in the object information receiving unit 110, when a user tracks a desired object using an application or the like, a user's mouse click or touch input is received to track the object at the corresponding location. Alternatively, when applied to an intelligent CCTV, etc., an object in which an abnormal behavior such as an intruder is detected may be automatically identified, selected as an object to be tracked, and information on the object may be received by the object information receiving unit 110.

The non-recognition tracking unit 120 tracks the movement of the object to be tracked by using the position of the object to be tracked in the previous frame of the video data and the current frame. Recognizing an object in each frame of a video and tracking the location of the object has been the most conventional object tracking technology, but the artificial intelligence network that recognizes objects focuses on accurately detecting and classifying objects. is adjusted, so the tracking performance for tracking a moving object is not high. For example, in the case of tracking a person's face, when the front of the face is exposed, it is possible to recognize that the corresponding part is a person's face, but in the next scene, if the face is rotated to the side or back, In the frame of the scene where the next face is turned backwards, the tracking fails because the face cannot be recognized.

Therefore, the non-recognition tracking unit 120 designs a network focused on tracking rather than recognition, and when an object exists in a certain position or range in the previous frame, how the object moves in the next frame. By checking, it is possible to accurately determine the movement of objects between frames. That is, if the conventional object recognition-based tracking technology enables tracking by utilizing recognition technology, the technology of detecting the movement of an object in the previous frame and the current frame in the non-recognition tracking unit 120 is optimized for object tracking. Therefore, it can be said that it is a technology that maximizes tracking performance.

The tracking technology using the location information of the object in the previous frame and the current frame information in the non-recognition tracking unit 120 can be implemented through artificial intelligence, and various artificial intelligence algorithms are being developed to implement this, In the present invention, it does not matter which method is applied. In general, a technology for tracking an object without recognizing it is called visual object tracking (VOT).

The method of tracking an object in the non-recognition tracking unit 120 enables relatively accurate tracking even when an object moving within the screen is deformed or rotated, and maximizes tracking performance by tracking using a network optimized for tracking. On the other hand, there is a problem that there is no way to track when an object goes out of the screen and then comes back in.

The object recognition tracker 130 recognizes the object to be tracked for each frame of the video data and tracks the movement of the object. As described above, the tracking method of the non-recognition tracking unit 120 cannot track an object moving out of the screen. If the object is recognized every frame, when the object that has moved out of the screen returns, it recognizes the object again There are advantages to discovering. Therefore, the object tracking system of the present invention utilizes the frame-by-frame object recognition-based tracking technology together with the non-recognition tracking unit 120 tracking technology to compensate for each other's disadvantages.

In order for the object recognition tracking unit 130 to recognize an object for each frame, a process of learning characteristics of the object using learning data of the object to be recognized in advance is required. If a human face is selected as an object to be tracked, it is necessary to learn various human face images as learning data so that the human face part and other objects can be distinguished in the image. Therefore, the object recognition tracking unit 130 checks information about the object to be tracked, determines whether the object to be tracked is a learned type of object, and tracks the movement of the object to be tracked only when the object is a learned type of object. can do.

The non-recognition tracking unit 120 and the object recognition tracking unit 130 may each calculate a tracking reliability value along with a result of tracking an object. Both the object tracking algorithm using the position of the object in the previous frame and the current frame of the non-recognition tracking unit 120 and the object recognition-based tracking algorithm for each frame of the object recognition tracking unit 130 have a tracking reliability value according to the tracking result. can provide. The tracking reliability value is a numerical value indicating how accurate the tracking result is, and may be calculated to have a value between 0 and 1.

For example, in the YOLO algorithm, which is one of the object recognition-based tracking methods in the object recognition tracking unit 130, the probability that a recognized object belongs to the class (classification) of the object is obtained, and the accuracy of the area where the object is located is determined. A reliability value can be obtained by obtaining an Intersection over Union (IOU) value representing , and multiplying the two. Since the probability that the recognized object belongs to the object class and the accuracy of the area both have a value between 0 and 1, the final tracking reliability value has a value between 0 and 1. In this example, tracking reliability can be calculated by Equation 1 below.

[Formula 1]

The comprehensive tracking unit 140 generates a comprehensive tracking result based on the tracking result of the object recognition tracking unit 130 and the tracking result of the non-recognition tracking unit 120 . When the tracking reliability of the tracking result of the unrecognized tracking unit 120 is high, the tracking result of the unrecognized tracking unit 120 may be used entirely to generate a comprehensive tracking result, or the tracking of the unrecognized tracking unit 120 When the reliability is low and the tracking reliability of the object recognition tracking unit 130 is high, the tracking result of the object recognition tracking unit 130 may be used as a comprehensive tracking result. Also, in the middle, a tracking result summed by assigning a weight to each may be generated as a comprehensive tracking result.

To this end, the comprehensive tracking unit 140 obtains the weights of the tracking result of the object recognition tracking unit 130 and the tracking result of the non-recognition tracking unit 120, respectively, using the tracking reliability value, and comprehensive tracking by reflecting them. can produce results. In addition, based on the tracking reliability value of the non-recognition tracking unit 120, a weight of a tracking result of the non-recognition tracking unit 120 is obtained, and the tracking reliability value of the object recognition tracking unit 130 and the non-recognition tracking unit The weight of the tracking result of the object recognition tracking unit 130 may be obtained based on the difference between the tracking reliability values of (120). The weight should have a value between 0 and 1, and when only one tracking result is used, the weight of the corresponding method may be set to 1 and the weight of the other method may be set to 0.

More specifically, the comprehensive tracking unit 140 may obtain the weight of the tracking result of the unrecognized tracking unit 120 by Equation 2 below.

[Formula 2]

At this time, wt is the weight of the tracking result of the unrecognized tracking unit 120.

ct is the tracking reliability of the unrecognized tracking unit 120 (value between 0 and 1)

,

는 신뢰도를 구분하기 위한 소정의 기준 값

,

Is a predetermined reference value for distinguishing reliability

,

는 추적 알고리즘을 무엇으로 적용하는지에 따라서, 변경될 수 있다.Depending on the tracking method, since the meaning of the reliability value may be slightly different,

,

may change depending on what the tracking algorithm is applied to.

In addition, the comprehensive tracking unit 140 may obtain the weight of the tracking result of the object recognition tracking unit 130 by Equation 3 below.

[Formula 3]

At this time, wd is the weight of the tracking result of the object recognition tracking unit 130.

ct is the tracking reliability of the unrecognized tracking unit 120 (value between 0 and 1)

cd is the tracking reliability of the object recognition tracking unit 130 (a value between 0 and 1)

는 신뢰도를 구분하기 위한 소정의 기준값

is a predetermined reference value for distinguishing reliability

는 추적 알고리즘을 무엇으로 적용하는지에 따라서 변경될 수 있다.Since the meaning of the reliability value may differ depending on the tracking method,

may change depending on what the tracking algorithm is applied to.

Unlike the conventional method, the present invention is to increase the accuracy of tracking by performing tracking in the non-recognition tracking unit 120 and the object recognition tracking unit 130, respectively, so it occupies a lot of computational resources compared to the conventional method. will do Therefore, the object recognition tracking unit 130 does not track an object when the tracking reliability value of the non-recognition tracking unit 120 is greater than or equal to a predetermined reference value, and performs tracking only when the tracking reliability value falls below the reference value. can That is, while the object is moving within the screen, the tracking method of the non-recognition tracking unit 120 is applied, but when the object moves out of the screen or moves at high speed and the reliability of tracking in the non-recognition tracking unit 120 decreases, object recognition tracking The unit 130 may execute tracing and collect and utilize the tracing results. This maximizes tracing performance while minimizing the use of computational resources.

2 is a diagram illustrating an example of a case where non-recognition tracking and object recognition tracking are applied according to an embodiment of the present invention.

As shown in the figure, when there is an object (human face in the figure) to be recognized on the screen, the object information receiving unit 110 first receives information about the location and attributes of the object.

When an object moves as in (b), the non-recognition tracking unit 120 and the object recognition tracking unit 130 respectively track the movement of the object, and synthesize them to generate a comprehensive tracking result. As in (b), since the reliability of the tracking result of the non-recognition tracking unit 120 is highly likely to appear when the object moves within the screen, in this case, a comprehensive tracking result can be generated only with the tracking result of the non-recognition tracking unit 120. can

If the object to be tracked moves out of the screen as in (c), it becomes difficult for the non-recognition tracking unit 120 to track the object. Even if the object enters the screen again as in (d), it is difficult to track the object's movement because the object's location information in the previous frame does not exist in the non-recognition tracking unit 120. Therefore, in this case, the object recognition tracking unit 130 recognizes the object again in the corresponding frame so that the tracking can be continued.

3 is a flow illustrating the flow of an object tracking method according to an embodiment of the present invention.

An object tracking method according to the present invention is a method of deriving a data boundary in an object tracking system having a central processing unit and a memory, and can be driven in such a computing system.

Accordingly, the object tracking method includes all of the characteristic configurations described for the object tracking system described above, and contents not described in the following description can also be implemented with reference to the description of the object tracking system described above.

In the object information receiving step (S301), information about an object to be tracked is received from video data composed of a plurality of frames. Moving image data is provided by arranging a plurality of still image frames in time series and reproducing the still image quickly according to the lapse of time, thereby providing a moving image. In order to analyze such video data, it is necessary to recognize objects included in the video. For example, when a CCTV detects an intruder, the intruder must be selected as an object to be tracked and movement must be identified.

The information about the object to be tracked received in the object information receiving step (S301) may include location information of the object to be tracked within the image or range information indicating the range in which the object to be tracked is located. information may be included. For example, when an object to be tracked is a face of a person, by providing attribute information that the object to be tracked is a face, more accurate recognition is possible when tracking an object through object recognition.

In the non-recognition tracking step (S302), the movement of the object to be tracked is tracked using the position of the object to be tracked in the previous frame of the video data and the current frame. Recognizing an object in each frame of a video and tracking the location of the object has been the most conventional object tracking technology, but the artificial intelligence network that recognizes objects focuses on accurately detecting and classifying objects. is adjusted, so the tracking performance for tracking a moving object is not high. For example, in the case of tracking a person's face, when the front of the face is exposed, it is possible to recognize that the corresponding part is a person's face, but in the next scene, if the face is rotated to the side or back, In the frame of the scene where the next face is turned backwards, the tracking fails because the face cannot be recognized.

Therefore, in the non-recognition tracking step (S302), a network focused on tracking rather than recognition is designed, and when an object exists in a certain position or range in the previous frame, how the object moves in the next frame By checking, it is possible to accurately determine the movement of objects between frames. That is, if the conventional object recognition-based tracking technology enables tracking by utilizing recognition technology, the technology for detecting the movement of an object in the previous frame and the current frame in the non-recognition tracking step (S302) is optimized for object tracking. Therefore, it can be said that it is a technology that maximizes tracking performance.

In the non-recognition tracking step (S302), the tracking technology using the location information of the object in the previous frame and the current frame information can be implemented through artificial intelligence. Various artificial intelligence algorithms are being developed to implement this, In the present invention, it does not matter which method is applied. In general, a technology for tracking an object without recognizing it is called visual object tracking (VOT).

In the object recognition tracking step (S303), the movement of the object is tracked by recognizing the object to be tracked for each frame of the video data. As described above, the tracking method of the non-recognition tracking step (S302) cannot track an object moving out of the screen. If the object is recognized every frame, when the object that has moved out of the screen returns, it recognizes the object again There are advantages to discovering. Therefore, the object tracking system according to the present invention simultaneously utilizes the frame-by-frame object recognition-based tracking technology along with the non-recognition tracking step (S302) to compensate for each other's disadvantages.

In order to recognize an object for each frame in the object recognition tracking step (S303), a process of learning characteristics of the object using learning data of the object to be recognized in advance is required. If a human face is selected as an object to be tracked, it is necessary to learn various human face images as learning data so that the human face part and other objects can be distinguished in the image. Therefore, in the object recognition tracking step (S303), information about the object to be tracked is checked to determine whether the object to be tracked is a learned type of object, and the movement of the object to be tracked is tracked only when the object is a learned type of object. can do.

In the non-recognition tracking step (S302) and the object recognition tracking step (S303), a tracking reliability value may be calculated together with a result of tracking the object, respectively. The object tracking algorithm using the position of the object in the previous frame and the current frame in the non-recognition tracking step (S302) and the object recognition-based tracking algorithm for each frame in the object recognition tracking step (S303) both track reliability values according to the tracking results. can provide. The tracking reliability value is a numerical value indicating how accurate the tracking result is, and may be calculated to have a value between 0 and 1.

The comprehensive tracking step (S304) generates a comprehensive tracking result based on the tracking result of the object recognition tracking step (S303) and the tracking result of the non-recognition tracking step (S302). If the tracking reliability of the tracking result of the non-recognized tracking step (S302) is high, the tracking result of the non-recognized tracking step (S302) may be used entirely to generate a comprehensive tracking result, or the tracking of the non-recognized tracking step (S302) When the reliability is low and the tracking reliability of the object recognition tracking step (S303) is high, the tracking result of the object recognition tracking step (S303) may be used as a comprehensive tracking result. Also, in the middle, a tracking result summed by assigning a weight to each may be generated as a comprehensive tracking result.

To this end, the comprehensive tracking step (S304) obtains the weights of the tracking result of the object recognition tracking step (S303) and the tracking result of the non-recognition tracking step (S302) using the tracking reliability value, respectively, and comprehensive tracking by reflecting them. can produce results. In addition, based on the tracking reliability value of the non-recognition tracking step (S302), the weight of the tracking result of the non-recognition tracking step (S302) is obtained, and the tracking reliability value of the object recognition tracking step (S303) and the non-recognition tracking step The weight of the tracking result of the object recognition tracking step (S303) can be obtained based on the difference between the tracking reliability values of (S302). The weight should have a value between 0 and 1, and when only one tracking result is used, the weight of the corresponding method may be set to 1 and the weight of the other method may be set to 0.

Unlike the conventional method, the present invention performs tracking in the non-recognition tracking step (S302) and the object recognition tracking step (S303) to increase the accuracy of tracking, so it takes up more computational resources than the conventional method. will do Therefore, in the object recognition tracking step (S303), when the tracking reliability value of the non-recognition tracking step (S302) is greater than or equal to a predetermined reference value, the object is not tracked, and tracking is performed only when the tracking reliability value falls below the reference value. can That is, while the object is moving within the screen, the tracking method of the non-recognition tracking step (S302) is applied, but when the object moves out of the screen or moves at high speed and the tracking reliability decreases in the non-recognition tracking step (S302), object recognition tracking is performed. In step S303, tracking may be executed, and the tracking results may be synthesized and utilized. This maximizes tracing performance while minimizing the use of computational resources.

The object tracking method according to the present invention may be produced as a program for a computer to execute and recorded on a computer-readable recording medium.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CDROMs and DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like.

Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter or the like as well as machine language codes such as those produced by a compiler. The hardware device may be configured to act as one or more software modules to perform processing according to the present invention and vice versa.

Although the above has been described with reference to the embodiments, those skilled in the art can variously modify and change the present invention without departing from the spirit and scope of the present invention described in the claims below.

101: object tracking system
110: object information receiving unit 120: non-recognition tracking unit
130: object recognition tracking unit 140: comprehensive tracking unit

Claims (11)

an object information receiving unit which receives information about an object to be tracked from video data composed of a plurality of frames;
a non-recognition tracking unit for tracking the movement of the object to be tracked using a position of the object to be tracked in a previous frame of the video data and a current frame;
an object recognition tracker for recognizing the object to be tracked for each frame of the video data and tracking the movement of the object; and
A comprehensive tracking unit that generates a comprehensive tracking result based on a tracking result of the object recognition tracking unit and a tracking result of the non-recognition tracking unit.
including,
The object recognition tracking unit and the non-recognition tracking unit each calculate a tracking reliability value,
The comprehensive tracking unit
Using the tracking reliability value, weights of the tracking result of the object recognition tracking unit and the tracking result of the non-recognition tracking unit are respectively obtained, and a comprehensive tracking result is generated by reflecting the weights,
Obtaining a weight of a tracking result of the unrecognized tracking unit based on a tracking reliability value of the unrecognized tracking unit;
Calculating a weight of a tracking result of the object recognition tracking unit based on a difference between a tracking reliability value of the object recognition tracking unit and a tracking reliability value of the non-recognition tracking unit
Characterized by an object tracking system. delete delete According to claim 1,
The object recognition tracking unit
Tracking the movement of the object to be tracked only when the tracking reliability value of the unrecognized tracking unit is equal to or less than a predetermined reference value;
The comprehensive tracking unit
Generating a tracking result of the non-recognition tracking unit as a comprehensive tracking result when the object recognition tracking unit does not track movement
Characterized by an object tracking system. According to claim 1,
The object recognition tracking unit
Confirm whether the object to be tracked is a learned type based on the information about the object to be tracked;
Track the movement of the object to be tracked only if it is a learned type of object,
The comprehensive tracking unit
When the object recognition tracker does not track movement, generating a tracking result of the non-recognition tracker as a comprehensive tracking result
Characterized by an object tracking system. In the object tracking method of an object tracking system having a memory and a central processing unit,
an object information receiving step of receiving information about an object to be tracked from video data composed of a plurality of frames;
a non-recognition tracking step of tracking the movement of the object to be tracked using a position of the object to be tracked in a previous frame of the video data and a current frame;
An object recognition tracking step of recognizing the object to be tracked for each frame of the video data and tracking the movement of the object; and
A comprehensive tracking step of generating a comprehensive tracking result based on a tracking result of the object recognition tracking step and a tracking result of the non-recognition tracking step.
and include
The object recognition tracking step and the non-recognition tracking step each calculate a tracking reliability value,
The comprehensive tracking step is
Using the tracking reliability value, weights of the tracking result of the object recognition tracking step and the tracking result of the non-recognition tracking step are respectively obtained, and a comprehensive tracking result is generated by reflecting them,
Based on the tracking reliability value of the non-recognized tracking step, weights of tracking results in the non-recognized tracking step are obtained;
Obtaining a weight of a tracking result of the object recognition tracking step based on a difference between a tracking reliability value of the object recognition tracking step and a tracking reliability value of the non-recognition tracking step.
An object tracking method characterized by delete delete According to claim 6,
The object recognition tracking step is
Tracking the movement of the object to be tracked only when the tracking reliability value in the non-recognized tracking step is equal to or less than a predetermined reference value;
The comprehensive tracking step is
If the object recognition tracking step does not track movement, generating a tracking result of the non-recognition tracking step as a comprehensive tracking result.
An object tracking method characterized by According to claim 6,
The object recognition tracking step is
Confirm whether the object to be tracked is a learned type based on the information about the object to be tracked;
Track the movement of the object to be tracked only if it is a learned type of object,
The comprehensive tracking step is
If the object recognition tracking step does not track movement, generating a tracking result of the non-recognition tracking step as a comprehensive tracking result.
An object tracking method characterized by A computer-readable recording medium on which a program for causing a computer to function to execute the method of any one of claims 6, 9, and 10 is recorded.

Images