KR101648208B1

KR101648208B1 - Method and apparatus for recognizing and tracking object by using high resolution image

Info

Publication number: KR101648208B1
Application number: KR1020150127253A
Authority: KR
Inventors: 김동기
Original assignee: 김동기
Priority date: 2015-09-08
Filing date: 2015-09-08
Publication date: 2016-08-12

Abstract

A method and apparatus for object recognition and tracking using high resolution images are disclosed. The object recognition and tracking method includes the steps of setting an initial setting value of a dividing step, dividing an image into a plurality of first dividing blocks by performing a first dividing step based on an initial setting value, Dividing an image into a plurality of second sub-blocks based on the first change sub-block by performing a second sub-step for the first change sub-block, And determining a second change division block in which image information is changed among the second split blocks of the first change block.

Description

TECHNICAL FIELD The present invention relates to a method and apparatus for recognizing and tracking an object using a high resolution image,

The present invention relates to an object recognition and tracking method, and more particularly, to a method and apparatus for object recognition and tracking using a high-resolution image.

[0003] Conventional CCTV (closed circuit television) cameras used in existing security monitoring systems have been widely used in full-HD (high definition) (resolution: 1080P, 1920 (width) x 1080 (length)). Currently, CCTV camera's resolution application technology is more advanced and can be served from terrestrial digital television (Digital TV) to Ultra-HD (4k)

Ultra-HD (resolution: 1080P * 4K, 3840 (landscape) x2160 (portrait)) images consist of 33,177,600 pixels. When an RGB color is applied to an Ultra-HD image, many color values of 99,532,800 are combined. Conventionally, an object recognition and tracking system using an image processor has a problem of processing big data of 15 frames per second as a limited central processing unit (CPU) resource. Typically, the object recognition system performs object recognition processing using a low resolution image of D1 (resolution: 720x480) or CIF (resolution: 352x240) which downsamples a high resolution image. Therefore, there are limitations on object recognition distance, limit of shape recognition, limitation of color recognition, and so there is a limitation in operating object recognition system efficiently.

KR 10-2009-0109719

One aspect of the present invention provides an object recognition and tracking method using a high-resolution image.

Another aspect of the present invention provides an object recognition and tracking apparatus using a high-resolution image.

An object recognition and tracking method according to an aspect of the present invention includes the steps of setting an initial setting value of a dividing step, dividing an image into a plurality of first dividing blocks by performing a first dividing step based on the initial setting value Determining a first change partitioning block in which image information is changed among the plurality of first partitioning blocks, performing a second partitioning step on the first change partitioning block, And dividing the plurality of second divided blocks into a plurality of second divided blocks, and determining a second changed divided block in which the image information among the plurality of second divided blocks is changed, wherein the second dividing step includes: And can be divided into unit blocks smaller in size than the division step.

If the block background model based on the block background model learning is generated, the second partitioning step is a final partitioning step based on the initial setting value, and the plurality of second partitioning blocks are allocated to the prediction moving position of the object And may correspond to a corresponding region of interest (ROI) region.

If the block background model based on the block background model learning is not generated, the second segmentation step is a next segmentation step of the first segmentation step, and the plurality of second segmentation blocks are divided into the first change And may correspond to a split block.

The object recognition and tracking method may further include extracting an object by detecting a pixel change in the second change division block when the second division step is the last division step based on the initial setting value have.

Also, the initial setting value may be determined based on the resolution of the image and the number of objects included in the image.

In addition, the change of the image information is determined based on an inter-frame bit reference operation for an area not including an object in a previous frame, and for an area including the object in the previous frame, Can be determined.

According to another aspect of the present invention, an image processing apparatus for performing object recognition and tracking includes a processor, and the processor sets an initial setting value of a dividing step and performs a first dividing step based on the initial setting value The first divided block is divided into a plurality of first divided blocks, a first changed divided block whose image information is changed among the plurality of first divided blocks is determined, and a second divided block is performed on the first changed divided block, 1 change sub-block, and to determine a second change sub-block whose image information is changed among the plurality of second sub-blocks, wherein the second sub- May divide the image into unit blocks smaller in size than the first division step.

In addition, the processor may be configured to detect a pixel change in the second change division block to extract an object if the second division step is the last division step based on the initial setting value.

The method and apparatus for recognizing and tracking an object using a high resolution image according to an embodiment of the present invention can quickly and accurately acquire information on the existence of an object and movement of an object in the image while reducing the amount of calculation for unnecessary image processing.

1 is a block diagram showing an image processing apparatus according to an embodiment of the present invention.
2 is a conceptual diagram illustrating an operation of an HBR object recognition unit according to an embodiment of the present invention.
3 is a flowchart illustrating a parallel processing method of an HBR object recognition unit according to an embodiment of the present invention.
4 is a flowchart illustrating a first job management step according to an embodiment of the present invention.
5 is a flowchart illustrating an operation of an HBR object recognition unit according to an embodiment of the present invention.
6 is a conceptual diagram illustrating a fusion object recognition unit according to an embodiment of the present invention.
7 is a flowchart illustrating an operation of the parallel operation processing unit according to the embodiment of the present invention.
8 is a conceptual diagram illustrating an operation method according to an embodiment of the present invention.
9 is a conceptual diagram illustrating a forward operation thread and a reverse operation thread according to an embodiment of the present invention.
10 is a conceptual diagram showing the result of the arithmetic processing according to the embodiment of the present invention.

The following detailed description of the invention refers to the accompanying drawings, which illustrate, by way of illustration, specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention are different, but need not be mutually exclusive. For example, certain features, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the invention in connection with an embodiment. It is also to be understood that the position or arrangement of the individual components within each disclosed embodiment may be varied without departing from the spirit and scope of the invention. The following detailed description is, therefore, not to be taken in a limiting sense, and the scope of the present invention is to be limited only by the appended claims, along with the full scope of equivalents to which such claims are entitled, if properly explained. In the drawings, like reference numerals refer to the same or similar functions throughout the several views.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

Background discrimination is used in object recognition method in existing high resolution image. The shape extraction by the difference of the background image is a method of separating the background from the foreground. The foreground can be an object of interest to us, that is, a vehicle in which a person is an object of interest in a crime prevention image and illegally parked in an illegal parking control system. On the other hand, the background may be a wall, a picture frame, a refrigerator in an indoor case, a building, a parked car, a tree, a traffic light, a street, and the like. In other words, the foreground is a dynamic object of interest to the user, and the background may be a floating object whose motion is small or has a pattern and should not be processed as an object. Under the assumption that the background does not change in the shape extraction method based on the difference of the background image, the object can be extracted based on the image change such as the change of the pixel value.

Hereinafter, an embodiment of the present invention discloses a method and apparatus for recognizing and tracking an object using a high-resolution image. By using block-based incremental object recognition method, high-resolution images are effectively processed in parallel, and complementary combining based on sensor fusion is used in a high density environment to duplicate the object recognition step, Can be reduced.

1 is a block diagram showing an image processing apparatus according to an embodiment of the present invention.

FIG. 1 illustrates an image processing apparatus for performing an object recognition and tracking method using a high-resolution image.

1, an image processing apparatus includes an image collecting unit 100, a background learning processing unit 105, an HBR (hierarchical block recognize) object recognizing unit 110, a fusion object recognizing unit 115, an object recognizing unit An object tracking unit 125, a parallel operation processing unit 130, a GPU operation unit 135, an object information storage unit 140, an object information transfer unit 145, a rotary camera 150, an infrared sensor / / Thermal camera 155, a video card graphics processor unit (GPU) 160, and a processor (not shown).

The image collecting unit 100 receives a high resolution image from a digital camera (for example, a rotary camera, an infrared sensor / a radar / thermal camera), and transmits information on the received image to the background learning processing unit 105 Or < / RTI >

The background learning processing unit 105 may be implemented to extract a primary object type based on background differencing. The background learning processing unit 105 may perform learning on the background model in units of seconds or minutes as a learning model of the background image. Also, the background learning processing unit 105 may be implemented for periodic updating of the learned background pixels during object recognition processing.

The HBR object recognizing unit 110 recognizes a high resolution image in at least four stages (1 stage division, 2 division) in a progressive block unit (for example, 8x8 blocks, 16x16 blocks, 32x32 blocks, 64x64 blocks, NxN blocks) Step division, three-step division, and four-step division).

In addition, the HBR object recognition unit 110 may be implemented to recognize information on a block having sequentially changed image information. For example, based on the background information learned by the background learning processing unit 105, a block including the changed image information can be determined from among blocks (for example, 8x8 blocks) divided into one-stage division. The block including the changed image information among the blocks divided into the one-stage division may be expressed by the changed one-stage block (or the one-stage change block).

Thereafter, it is possible to perform the division into two-stage division for the changed first-stage block. The block including the changed image information among the blocks divided into the two-stage division may be expressed by the changed two-stage block (or the two-stage change block). That is, the changed 2-level block can more specifically indicate the image change portion within the changed 1-level block as a block included in the changed 1-level block.

In addition, the HBR object recognition unit 110 may gradually divide the block information step by step and determine the shape of the object by considering the information of pixels changed for each block using the learned background information in the background learning processing unit 105 . When the progressive block information refinement method is used in the HBR object recognition unit 110, the amount of computation can be reduced.

The fusion object recognizing unit 115 can be implemented for the purpose of supplementing the limitation of the GPU processing. Specifically, a fusion object recognition unit can be used to supplement the processing limit of the GPU, such as when a large number of objects are generated in a high density region or when an image processing is performed in a general-purpose PC. The fusion object recognizing unit 115 can reduce the amount of computation based on the changed information of the block using the information of the object received from the infrared sensor as an auxiliary means.

The object recognition unit 120 extracts only the corresponding block using the machine learning data based on the object information determined by the HBR object recognition unit 110 and the object information determined by the fusion object recognition unit 115, To perform the final object recognition.

The object tracking unit 125 may be implemented to track an object by calculating a moving distance and direction of the object and controlling a camera having a controller for Pan, Tilt, and Zoom.

The parallel operation processing unit 130 may be implemented to resolve the bottleneck of the operation by referring to bits between the current frame (the n-th frame) and the previous frame (the (n-1) -th frame). The parallel operation processing unit 130 can use the performance of the GPU as much as possible with effective parallel processing. The parallel operation processing unit 130 may further use the inner bit of the previous frame as the object recognition information in order to increase the reliability of the operation information according to the inter-frame bit reference (Outer bit) .

The GPU computing unit 135 may be implemented to share an API (Application Program Interface) that can be used with a GPU for each graphics card manufacturer.

The object information storage unit 140 may be implemented to store object information (object position, object size, object color, object orientation).

The object information transmitting unit 145 may be implemented to transmit the object information through the wired / wireless network of the central control center.

The rotary camera 150 may be implemented to obtain general image information for the object and the infrared sensor / radar / thermal camera 155 may be configured to receive infrared image information / radar image information / thermal image information for the object And can be implemented to obtain.

The video card GPU 160 may be implemented for processing of received video information.

2 is a conceptual diagram illustrating an operation of an HBR object recognition unit according to an embodiment of the present invention.

In FIG. 2, a method of searching for an object while gradually decreasing the image segment size in the HBR object recognition unit is disclosed.

Specifically, the HBR object recognition unit can divide the image into a plurality of divisional steps, and an initial setting value for the number of divisional steps can be set according to the image resolution and the recognition zone environment. For example, the initial setting value may be set to three divided steps for HD-level video, four divided steps for full HD video, and six divided steps for Ultra-HD (4k) video.

Also, if the number of objects in the captured image is large (for example, an urban area image), two or three steps may be added to the initial setting value. For example, the image capturing apparatus may set an initial setting value after first determining the number of objects in the image.

Referring to FIG. 3, the first image may be unprocessed high-resolution image data received from the image capturing apparatus. Hereinafter, in the embodiment of the present invention, high-resolution image data is assumed to be HD-level image data for convenience of explanation.

The per-block object recognition can be performed progressively through a plurality of segmentation steps (for example, three-step segmentation) on the image data. Specifically, the image is down-sampled into 4x4 blocks, 8x8 blocks, and 16x16 blocks, and the per-block object recognition is performed gradually. Gaussian pyramid is used for downsampling to reduce the computation speed, and background differencing of pixels can be used for block-by-block conversion.

The first level block indicates an individual block when the image is divided into one level (for example, 4x4 blocks), and the two level block indicates that the image is divided into two levels (for example, 8x8 blocks) , The third block indicates the individual block when the image is divided into three stages (for example, 16x16 blocks).

For example, in the case of a first-stage block (hereinafter referred to as a first-stage change block) 200 in which a first-stage block of a plurality of first-stage blocks is changed, a plurality of two- have. (Hereinafter referred to as a 2-step change block) 220 among a plurality of 2-step blocks corresponding to the 1-stage change block. Similarly, a plurality of three-stage blocks corresponding to the two-stage change block can be determined in the same manner. A three-step change block (hereinafter referred to as a three-step change block) 240 among a plurality of three-step blocks corresponding to the two-step change block may be determined. The area in which the object is included in the image can be determined through the iterative block search method.

In the last step, the region of the changed pixel can be extracted except for the background in the region containing the object. In the case of FIG. 2, the changed pixel region can be extracted except the unchanged pixel region in the three-step change block. In addition, if block background learning is performed in the iterative structure and block background learning model exists, the recognition step of the intermediate block change is omitted and the ROI (Region Of Interest) region of the predicted calculated movement position of the object is considered. (For example, 16x16 blocks).

That is, the image processing apparatus sets an initial setting value of the dividing step and divides the image into a plurality of first dividing blocks by performing the first dividing step based on the initial setting value. In addition, the image processing apparatus may determine a first change division block in which image information is changed among a plurality of first division blocks, perform a second division step on the first change division block, Can be divided into a plurality of second divided blocks. Thereafter, the image processing apparatus can determine the second change division block in which the image information among the plurality of second split blocks is changed. And the second dividing step may be a step of dividing the image into unit blocks smaller in size than the first dividing step.

3 is a flowchart illustrating a parallel processing method of an HBR object recognition unit according to an embodiment of the present invention.

Referring to FIG. 3, in a first job management step S300, a change in image information in a block is searched for and position information (or block information) of a block to be calculated may be input to a job waiting queue.

In the job waiting queue step (S310), the job waiting queue determines a block to be operated based on a FIFO (First In First Out).

In the second job management step S320, the block information determined by the job waiting queue is transmitted to the thread in standby, which is a pool of the threads.

In the job execution step (S330), an operation on a block to be processed corresponding to the block information determined in the second job management step (S320) can be processed using a GPU related command. In the job execution step S330, operations for a plurality of job target blocks may be performed, and operations for a plurality of job target blocks may be performed based on the bit unit parallel processing (S340).

In operation S330, when the operation is completed, the operation result information about the operation target block may be transmitted to the first operation management step S300.

In the first job management step (S300), information on a block (next block) further subdivided according to the result of the pixel variation amount of the operation target block included in the operation result information on the received operation target block is input to the job waiting queue . Specifically, according to the result of the pixel change amount for the specific one-stage block (one block of 4x4 blocks for the image), a plurality of two-stage blocks (one block of 16x16 blocks for the image, (I.e., four blocks corresponding to the four blocks) can be determined. Information on each of the plurality of two-level blocks may be determined in the first job management step (S300) and transmitted to the job waiting queue step (S310).

Each of the plurality of two-level blocks becomes a work target block, and the pixel variation amount for each of the plurality of two-level blocks can be calculated through the second job management step (S320) and the job execution step (S330).

4 is a flowchart illustrating a first job management step according to an embodiment of the present invention.

4, the image is divided into 4x4 blocks (1-step partition), 8x8 blocks (2-step partition), 16x16 blocks (3-step partition), 32x32 blocks (4-step division), and the object recognition process is progressively performed, the object recognition process is started by progressing from the 1-stage division to the 2-stage division.

Step 1 division is performed and information on the first-level block can be input to the job waiting queue (step S410).

Specifically, a data structure for 16 blocks including position information for each of 4x4 blocks (16 blocks) based on one-stage partitioning is input to the job waiting queue.

The operation result for the plurality of first-level blocks based on the one-level partitioning is received and the amount of change is detected (step S420).

As described above, based on the data structure of 16 blocks in the work waiting queue, information on each of 16 blocks is transmitted to a thread in a standby working pool of a thread pool, An operation on the change amount of the 16 blocks as the target block can be performed. The calculation result of the variation of the 16 blocks can be transmitted to the first operation stage and the variation amount for each of the 16 blocks can be detected based on the calculation result of the variation of the 16 blocks.

The block background model is learned (step S430).

The part corresponding to the background excluding the object in the image is determined, and the image processing apparatus can determine the block background model by performing learning on the background.

A change in the block background model is detected (step S440).

Detection of changes in the block background model can be performed from a two-step partition. When a change occurs in the learned block background model, the next step division is not performed, and the last step division is performed on the ROI (Region Of Interest) region corresponding to the predicted movement position of the object.

When the last step division is performed on the area corresponding to the ROI corresponding to the predicted movement position of the object, the amount of change may be calculated in a plurality of last step blocks corresponding to the block in which the change is detected. For example, a data structure for 1024 blocks including position information for each of 32x32 blocks divided on the basis of the last stage division is input as a data structure to a job wait queue, and a plurality of The variation amount for each of the last stage blocks can be calculated. If such a method is used, the intermediate step is not performed because it proceeds directly to the four-step division without going through the two-step division and the three-step division, so that the unnecessary arithmetic amount can be reduced.

When the change of the block background model is not detected and the change amount is detected, the data structure including the position information for each of the 16x16 blocks divided based on the 2 step division, which is the next step of the 1 step division, In other words, when a change in the block background model is not detected and a change amount is detected, the dividing step is sequentially applied to each of the plurality of divided blocks in the next division step corresponding to the detected divided block The amount of change can be calculated.

5 is a flowchart illustrating an operation of an HBR object recognition unit according to an embodiment of the present invention.

Referring to FIG. 5, a step of dividing a block is specified (step S500).

An initial setting value which is the maximum value of the dividing step according to the resolution is set and the dividing step of the block is designated in consideration of the initial setting value. As described above, the HBR object recognizer can divide an image into a plurality of divided steps according to an initial setting value. The initial setting value for the number of division steps can be set according to the image resolution and recognition zone environment. For example, the initial setting value may be set to three divided steps for HD-level video, four divided steps for full HD video, and six divided steps for Ultra-HD (4k) video.

It is determined whether or not the division step according to the initial setting value is completed (step S505).

That is, when the division step set to the initial set value (count) is performed to the end, a contour extraction algorithm (for example, a CannyEdge algorithm) is used for the block having the changed image information in the step S560 And recognizes the object. If the set division step is not completed, the image processing procedure according to the designated division step proceeds.

The image is downsampled to the designated division step (step S510).

Downsampling based on a Gaussian pyramid algorithm is performed on the image.

A pixel change is detected in a plurality of divided N-level blocks in a designated division step (step S515).

The changed pixels are determined based on the background differencing. For example, if the pixel difference is 50 or more, it can be regarded as a changed pixel.

N step change blocks are detected and the initial set values are changed (step S520).

An N step change block may be determined based on the pixel change value calculated in step S515. In addition, the arithmetic processing unit can change the initial set value in consideration of the image resolution and the recognition zone environment.

The block background model is learned (step S525).

As described above, it is possible to omit the intermediate division step through learning of the block background model and extract the object through the operation for the final division step.

It is determined whether there is a changed block in the area of the block background model (step S530).

If there is a changed block in the area of the block background model, the process proceeds to step S540 without comparing any more blocks.

If there is no changed block in the area of the block background model, the presence or absence of the N-level change block determined in step S515 is determined. If the above-described image processing is performed on a plurality of N + And the (N + 1) -step change block among the plurality of (N + 1) -step blocks may be determined.

A final segmentation block for a prediction custom area (ROI) is determined by predicting the moving position of the object (step S540).

If there is a changed block in the area of the block background model, the last divided block for the ROI is determined by predicting the moving position of the object without performing the intermediate partitioning step.

A change pixel is detected in the last division step block (step S545), and a last division step change block in which a change pixel exists in the last division step block is determined (step S550).

That is, the changed pixels can be determined using the difference of the background image after the downsampling process only in the customary ROI region. Based on the information about the changed pixel, the last divided block is determined.

A contour line is extracted in the last division step change block (step S555).

Contours can be extracted in the last split-step change block to extract the object.

6 is a conceptual diagram illustrating a fusion object recognition unit according to an embodiment of the present invention.

Referring to FIG. 6, the fusion object recognizing unit receives sensing information sensed by a fusion sensor (for example, a radar) through a receiving apparatus.

The fusion object recognition unit identifies the object based on the sensing information (step S600).

The fusion object recognition unit can determine the presence or absence of the object by considering the size and shape of the object based on the information of the object received from the reception device of the fusion sensor.

The fusion object recognition unit maps the information of the determined object to the coordinate system of the image to vectorize the position of the object (step S610).

And transmits the object information to the object recognition unit (step S620).

The fusion object recognition unit may transmit object information including object position, object size and shape information to the object recognition unit.

7 is a flowchart illustrating an operation of the parallel operation processing unit according to the embodiment of the present invention.

The parallel operation processing unit is a means for resolving the reference of bits in a frame generating a bottleneck which is an inhibiting factor of the parallel operation processing for the most effective parallel processing of the GPU, the n-th frame) and the previous frame (n-1 < th > frame).

Also, in order to improve the reliability of the calculation result by the bit reference between the current frame (n-th frame) and the previous frame (n-1-th frame), an intra-frame bit reference operation method is applied only to an area including the object being tracked . That is, the intra-frame bit reference operation can be used to increase the reliability of the object only in the area including the object in the previous frame (n-1th frame).

The in-frame bit reference operation can also be used to maximize the parallel computation performance of the GPU by executing the forward operation thread, the forward operation thread, and the common block operation processing in order to remove the bottleneck elements as much as possible

Referring to FIG. 7, the parallel operation processing using the GPU can be divided into an operation method based on inter-frame bit reference and an operation method based on intra-frame bit reference.

The operation method based on inter-frame bit reference is a means for processing a bottleneck due to an operation method based on intra-frame bit reference, and is a method for performing operations by bit reference between a current frame (n-th frame) and a previous frame Can be performed.

The intra-frame bit-by-frame arithmetic operation is performed by performing in-frame bit reference operation only on the object location area included in the previous frame (n-1th frame) as a means for improving object recognition reliability of the result of operation between frames, The reliability of the extraction can be enhanced.

An operation method based on a specific inter-frame bit reference is disclosed.

And requests a full frame operation (step S700).

The full frame operation can be performed based on the entire pixel information (or bit information) of the current frame (n-th frame) and the previous frame (n-1) frame.

The inter-frame bit is referred to (step S705).

The parallel processing operation is performed on the basis of the GPU instruction by referring to the bit information of the previous frame and the bit information of the current frame (step S710).

The result of the inter-frame parallel processing operation is processed (step S715).

An operation method based on a specific in-frame bit reference is disclosed.

In order to improve the reliability of the calculation result between frames, an operation in a frame may be requested only in an object location area of a previous frame (n-1 < th > frame) (step S720).

The forward calculation thread and the reverse calculation thread are executed (step S725).

For the purpose of computation reduction, a thread that computes in the forward direction and a thread that performs the computation in the reverse direction are separated and executed for the minimum in-frame bit reference. The forward operation thread and the reverse operation thread can be executed around the operation end block by setting a specific operation end block. Hereinafter, the forward calculation thread and the reverse calculation thread are specifically disclosed.

The bit in the thread-specific frame is referred to (step S730), and a parallel processing operation using the GPU instruction is performed (step S735).

The forward operation processing thread and the reverse operation processing thread perform operation processing on the common block that can not be processed (step S740).

The start, end, and error of the calculation are managed to process the calculation result (step S745).

8 is a conceptual diagram illustrating an operation method according to an embodiment of the present invention.

In Fig. 8 (A), a calculation method based on inter-frame bit reference is disclosed.

In Fig. 8 (B), a calculation method based on in-frame bit reference is started.

Referring to FIG. 8A, a bit reference between the current frame (N-th frame) and the previous frame (N-1-th frame) is started. A comparison between the bits in the corresponding area between the current frame and the previous frame can be performed.

Referring to FIG. 8B, a bit reference is started within the current frame (Nth frame). A comparison of bits between adjacent regions in the current frame can be performed.

The following table is an algorithm that represents a data reference structure that can not be parallelized.

<Table>

A table describes the structure in which the result of the current data operation is a dependent structure of the result value of the previous data, so that the parallel operation processing can not be used.

9 is a conceptual diagram illustrating a forward operation thread and a reverse operation thread according to an embodiment of the present invention.

Referring to FIG. 9, a thread for performing calculation processing on the forward (900) side and a thread for performing an arithmetic processing on the reverse side (930) may be separately executed for a minimum in-frame bit reference for the purpose of calculation reduction. The forward operation processing thread may process the operation up to the first operation end block 910 and the reverse operation processing thread may perform operation processing until the second operation end block 920 so that the bottleneck due to the bit reference in the frame may be resolved.

10 is a conceptual diagram showing the result of the arithmetic processing according to the embodiment of the present invention.

In Fig. 10, the result of the arithmetic processing by the bit reference between the current frame (n-th frame) and the previous frame (n-1 < th > frame) and the result of arithmetic processing within the frame are schematized.

In FIG. 10 (A), the change of the object in the previous frame (n-1) -th frame and the current frame (n-th frame) is schematized.

In FIG. 10 (B), the appearance of the object changes as a result of the parallel operation processing based on the GPU of the current frame (n-th frame) and the previous frame (n-1) frame. Referring to FIG. 10B, it can be confirmed that the object position of the previous frame ((n-1) th frame) overlaps with the object position of the current frame (n-th frame).

In order to solve the duplicate recognition of the object position, the parallel associating process in the frame can be performed only in the object moving position area. In FIG. 9C, the result of performing the parallel associating process in the frame only for the object moving position area is started.

Such an object recognition and tracking method using a high-resolution image can be implemented in an application or implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, and the like, alone or in combination.

The program instructions recorded on the computer-readable recording medium may be ones that are specially designed and configured for the present invention and are known and available to those skilled in the art of computer software.

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

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

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it will be understood by those skilled in the art that various changes and modifications may be made therein without departing from the spirit and scope of the invention as defined in the appended claims. It will be possible.

Claims

Object recognition and tracking methods,
Setting an initial set value of the dividing step;
Dividing an image into a plurality of first divided blocks by performing a first dividing step based on the initial setting value;
Determining a first change partitioning block in which image information among the plurality of first partitioning blocks is changed;
Dividing the image into a plurality of second split blocks based on the first change split block by performing a second split step on the first change split block; And
And determining a second change partitioning block in which image information among the plurality of second partitioning blocks is changed,
The second dividing step divides the image into unit blocks smaller in size than the first dividing step,
When the block background model based on the block background model learning is generated, the second segmentation step is the last segmentation step based on the initial setting value, and the plurality of second segmentation blocks are corresponding to the predicted movement position of the object Corresponds to a region of interest (ROI) region,
If the block background model based on the block background model learning is not generated, the second segmentation step is a next segmentation step of the first segmentation step, and the plurality of second segmentation blocks are divided into the first change segmentation block Respectively,
The change of the image information is determined based on an inter-frame bit reference operation for an area not including an object in a previous frame, and an area including the object in the previous frame is determined based on an intra-frame bit reference operation &Lt; / RTI >

delete

The method according to claim 1,
Further comprising the step of detecting an object by detecting a pixel change in the second change division block when the second division step is the last division step based on the initial setting value.

The method according to claim 1,
Wherein the initial setting value is determined based on the resolution of the image and the number of objects included in the image.

delete

1. An image processing apparatus for performing object recognition and tracking,
Wherein the image processing apparatus includes a processor,
The processor sets an initial setting value of the dividing step,
Dividing an image into a plurality of first divided blocks by performing a first dividing step based on the initial setting value,
Determining a first change division block in which image information of the plurality of first divided blocks is changed,
Dividing the image into a plurality of second sub-blocks based on the first change sub-block, performing a second sub-step on the first change sub-block,
And a second change division block in which image information among the plurality of second split blocks is changed,
The second dividing step divides the image into unit blocks smaller in size than the first dividing step,
When a block background model based on a block background model learning is generated, the second segmentation step is a final segmentation step based on an initial set value, and the plurality of second segmentation blocks correspond to a predicted movement position of the object Corresponds to a region of interest (ROI) region,
If the block background model based on the block background model learning is not generated, the second segmentation step is a next segmentation step of the first segmentation step, and the plurality of second segmentation blocks are divided into the first change segmentation block Respectively,
The change of the image information is determined based on an inter-frame bit reference operation for an area not including an object in a previous frame, and an area including the object in the previous frame is determined based on an intra-frame bit reference operation And the image processing apparatus.

delete

8. The method of claim 7,
Wherein the processor is configured to extract an object by detecting a pixel change in the second change division block when the second division step is the last division step based on the initial setting value.

8. The method of claim 7,
Wherein the initial setting value is determined based on the resolution of the image and the number of objects included in the image.

delete