KR102470139B1 - Device and method of searching objects based on quad tree - Google Patents

Abstract

According to an embodiment of the present invention, an object detection method for detecting an object in an image by receiving an image includes receiving the image as an analysis target for detecting the object; analyzing a distribution of the object in the image by dividing the image into four cells and dividing the cell in which the object exists among the cells into four lower cells more than once; storing an analysis result of the image for the distribution of the image or the object; and outputting the analysis result for the distribution of the object.

Description

Quad tree-based object detection method and apparatus {DEVICE AND METHOD OF SEARCHING OBJECTS BASED ON QUAD TREE}

The present invention relates to a method and apparatus for detecting an object, and more particularly, to a apparatus and method for deriving a result by searching whether an object exists in an image and, if so, where it exists, etc. based on a quad tree technique. It is about.

As artificial intelligence technology develops, research on technology for detecting objects in space is being actively conducted. Object detection is a technique of identifying objects to be found in a video sequence and clustering pixels of these objects. Object detection, one of the fundamental and challenging problems of computer vision, has been an active research field for decades.

Korean Patent Publication No. 2002-0064888 (August 10, 2002) An object-oriented video system includes at least one video, text, audio, video packet stream, text packet stream, audio packet stream, music packet stream and/or graphic packet stream. encoding data including music and/or graphic elements, combining the packet streams into one self-contained object containing its control information, putting a plurality of the objects into a data stream and grouping one or more of the data streams in a single independent scene including a format definition as an initial packet into a series of packets. An encoder for implementing the method is provided along with a player or decoder for parsing and decoding the file, which is wirelessly streamed to a portable computing device such as a mobile phone or PDA. Object control provides rendering and interactive control over objects, allowing users to control dynamic media combinations, such as controlling received objects and dictating the shape and content of interleaved video objects.

In addition, Korea Patent Publication No. 2010-0051775 (May 18, 2010) multi-object tracking using real-time image analysis, vehicle information recognition, and data management system and processing method are multi-object tracking using a high-resolution camera, object classification, vehicle information recognition, and Provides a way to manage data. Currently, products related to license plate recognition or object analysis are used on a limited basis, but more commercialized and integrated functions are urgently required. The purpose of this system is to provide more advanced information on vehicles and pedestrians passing on general roads by using high-definition high-resolution cameras. of) information can be obtained. Object tracking and vehicle information recognition are processed in real time (15 frames/sec), and each object information consists of a series of image data, time, object classification, and vehicle recognition information (in case of a vehicle). The surveillance area is 8*15M (based on resolution of 1600*1200 pixels) and is designed to be installed and operated on general roads.

Korean Patent Publication No. 2002-0064888 (2002.08.10) Korean Patent Publication No. 2010-0051775 (2010.05.18)

An embodiment of the present invention intends to propose a method of approximately quickly storing the distribution of objects detected by various methods. Also, for this purpose, we propose a quad-tree based object search technique in the space. According to an embodiment of the present invention, a quad tree-based object search function is intended to be used in the field of view of a robot or the identification of a fire area in which an obstacle location must be identified in real time.

Among embodiments, an object detection apparatus for detecting an object in an image by receiving an image includes an image input unit for receiving the image, which is an analysis target for detecting the object; an object detection unit dividing the image into four cells and dividing the cell in which the object exists among the cells into four sub-cells one or more times to analyze the distribution of the object in the image; a memory for storing an analysis result of the image or distribution of the object; and a result output unit outputting a result of the analysis of the distribution of the object.

And, when the object detection unit divides the image two or more times, four cells are generated during the first division of the image, and in this case, the division depth of the first division is 1, and the division depth of the first division is 1. The level of the divided cell is 1, and when the image is divided n times, each divided cell is divided into 4 sub-cells n times, the division depth of the n-th division is n, and the n-th division The level of the lower cell created by division is n.

In addition, each of the cell and the sub-cells has a position value corresponding to the division depth in which the cell and the sub-cells are divided and positions of the cell and the sub-cells in the image.

When the segmentation of the image is completed, the object detection unit generates data indicating whether the object exists in each of the cells and sub-cells.

In addition, a quad tree corresponding to the structure of the cell and the sub-cell is generated, and the data for each cell and each sub-cell is stored together with the quad tree.

In addition, when the object detection unit receives a request for information on the area of the object in the image, the location of the object is derived through the location values of the cell and the sub-cell where the object exists.

In another embodiment of the present invention, an object detection method for receiving an image and detecting an object in the image includes receiving the image, which is an analysis target for detecting the object, as input; analyzing a distribution of the object in the image by dividing the image into four cells and dividing the cell in which the object exists among the cells into four lower cells more than once; storing an analysis result of the image for the distribution of the image or the object; and outputting the analysis result for the distribution of the object.

The disclosed technology may have the following effects. However, it does not mean that a specific embodiment must include all of the following effects or only the following effects, so it should not be understood that the scope of rights of the disclosed technology is limited thereby.

An embodiment of the present invention may provide a method of approximately quickly storing the distribution of objects detected by various methods. Also, for this purpose, a quad tree-based object search technique within the space can be presented. If the object detection technique according to the embodiment of the present invention is applied, it is possible to improve the quad tree-based object search function in the field of view of the robot, which should be able to identify the position of an obstacle in real time, or in the detection of a fire area.

In addition, through an embodiment of the present invention, it is possible to quickly search for an approximate location of an object in a specific space, and for this purpose, a technique of searching for and storing an object using a quad tree data structure can be used. In addition, according to an embodiment of the present invention, an approximate distribution and area of an object may be obtained without searching all pixel spaces in an image space.

1 is a diagram illustrating an object detection apparatus according to an embodiment of the present invention.
2 is a diagram illustrating a processor in an object detection device according to an embodiment of the present invention.
3 is a flow chart illustrating an object detection method according to an embodiment of the present invention.
4 is a diagram illustrating a process of cell arrangement and 2D conversion of an image according to an embodiment of the present invention.
5 is a diagram for explaining tree transformation according to an embodiment of the present invention.
6 is a diagram illustrating a search target space on a coordinate plane according to an embodiment of the present invention.
7 is a diagram illustrating a quad-tree based object search technique according to an embodiment of the present invention.
8 is a diagram illustrating a search target space in a tree structure of depth 1 according to an embodiment of the present invention.
9 shows a saturation tree according to an embodiment of the present invention.
10 is a diagram showing a test screen of an object search method according to an embodiment of the present invention.
11 is a graph showing the average time taken for searching for each tree level according to an embodiment of the present invention.
12 is a graph showing the average time taken for searching for each tree level according to another embodiment of the present invention.
13 is an experimental image for an object search technique according to an embodiment of the present invention.
14 is a graph showing an average search time for each image when an object is detected according to an embodiment of the present invention.

Since the description of the present invention is only an embodiment for structural or functional description, the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, since the embodiments may be variously modified and may have various forms, it should be understood that the scope of the present invention includes equivalents capable of realizing the technical idea. In addition, since the object or effect presented in the present invention does not mean that a specific embodiment should include all of them or only such effects, the scope of the present invention should not be construed as being limited thereto.

Meanwhile, the meaning of terms described in this application should be understood as follows.

Terms such as "first" and "second" are used to distinguish one component from another, and the scope of rights should not be limited by these terms. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element.

It should be understood that when an element is referred to as being “connected” to another element, it may be directly connected to the other element, but other elements may exist in the middle. On the other hand, when an element is referred to as being "directly connected" to another element, it should be understood that no intervening elements exist. Meanwhile, other expressions describing the relationship between components, such as “between” and “immediately between” or “adjacent to” and “directly adjacent to” should be interpreted similarly.

Expressions in the singular number should be understood to include plural expressions unless the context clearly dictates otherwise, and terms such as “comprise” or “having” refer to an embodied feature, number, step, operation, component, part, or these. It should be understood that it is intended to indicate that a combination exists, and does not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

In each step, the identification code (eg, a, b, c, etc.) is used for convenience of explanation, and the identification code does not describe the order of each step, and each step clearly follows a specific order in context. Unless otherwise specified, it may occur in a different order than specified. That is, each step may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

The present invention can be implemented as computer readable code on a computer readable recording medium, and the computer readable recording medium includes all types of recording devices storing data that can be read by a computer system. . Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices. In addition, the computer-readable recording medium may be distributed to computer systems connected through a network, so that computer-readable codes may be stored and executed in a distributed manner.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs, unless defined otherwise. Terms defined in commonly used dictionaries should be interpreted as consistent with meanings in the context of the related art, and cannot be interpreted as having ideal or excessively formal meanings unless explicitly defined in the present application.

1 is a diagram illustrating an object detection apparatus 100 according to an embodiment of the present invention.

The object detection device 100 corresponds to a computing device that can be connected to other devices through a network, and can perform a function of receiving and analyzing an image and detecting an object in the image. In particular, the object detection device 100 may maintain a network connection with a device that captures or generates an image.

In an embodiment of the present invention, the object detection apparatus 100 proposes a technique for roughly and quickly storing the distribution of objects detected by various methods, in particular, a quad tree-based object search technique in space. Through the proposed technique, the object detection device 100 divides the image into 4 cells and searches the distribution of objects to derive a quad tree result used to determine the approximate distribution of objects in the image and calculate the area where the objects are distributed. can do. Details will be given later.

According to an embodiment of the present invention, the object detection device 100 may include an image input unit 110, an object detection unit 120, a memory 130, and a result output unit 140.

Of course, the object detection device 100 does not have to include all of the above-described components at the same time, and may selectively include some of the above components according to embodiments, and further include other components other than the above components. You may.

The image input unit 110 receives an image of a space to be analyzed and object detected. The image input unit 110 may include an image generating module that directly generates an image, such as a camera, or a communication module capable of receiving an image generated by another device through a network. The image input unit 110 may receive an image through a wireless communication network or may receive an image through a closed circuit.

When the video input unit 110 includes a communication module, the communication module includes an environment for connecting to other devices through a network, and may include, for example, an adapter for LAN (Local Area Network) communication. .

The object detection unit 120 may execute image analysis and object detection procedures according to an embodiment of the present invention, manage the memory 130 read or written in this process, and Synchronization time between volatile memory and non-volatile memory can be scheduled.

The object detection unit 120 may control the overall operation of the object detection device 110 in the image, and is electrically connected to the image input unit 110, the memory 130, and the result output unit 140 so as to flow data between them. can control. The object detection unit 120 may be implemented as a central processing unit (CPU) of the object detection device 110 .

In an embodiment of the present invention, the object detection unit 120 implements image analysis and object detection technology using a quad tree. A quadtree is a type of hierarchical data structure, and has a characteristic of saving execution time due to hierarchical characteristics that depend on the number of blocks of an image rather than the size of an image.

Here, hierarchical data structure is becoming an increasingly important representation technology in the fields of computer graphics technology, image processing, computational geometry, geographic information systems, and robotics. In the embodiment of the present invention, 'Region Quadtree' technology, which has been studied the most among quad trees, is used.

More specific operation of the object detection unit 120 will be described in detail with reference to FIG. 3 hereinafter.

The memory 130 is implemented as a non-volatile memory such as a solid state disk (SSD) or a hard disk drive (HDD) to store an image input through the image input unit 110, and the image is analyzed for object detection. Results can be saved. In addition, the memory 130 may include an auxiliary memory device used to store all data necessary for the object detection device 110, and may include a main memory device implemented as a volatile memory such as RAM (Random Access Memory). have. As such, the memory 130 may be implemented as volatile and non-volatile memory, and if implemented as a non-volatile memory, it may be implemented to be connected through a hyperlink.

The result output unit 140 may output an image analyzed through the object detection unit 120 or a result of object detection. Data obtained by analyzing the image for the location and area of the detected object may be output through the result output unit 140 . The result output unit 140 responds to the user's request by providing an image of the part where the object is detected, an image in which the object of the detection target is photographed, and information about the area of the area where the object is detected in the image corresponding to the user's input signal. Correspondingly, it can be output.

2 is a diagram illustrating an object detection unit 130 in an object detection device according to an embodiment of the present invention.

The object detection unit 130 of the object detection device 100 according to an embodiment of the present invention includes an image analysis unit 210, an image conversion unit 220, a quad tree processing unit 230, an area calculation unit 240, and a control unit. (250).

The object detection unit 130 also does not have to include all of the above-described components at the same time, and may selectively include some of the above components according to embodiments, and further include other components other than the above components. You may.

According to the aforementioned regional quad tree technology, the image analyzer 210 divides a space into four equally sized quadrants, that is, four cells. The image conversion unit 220 converts whether or not an object is detected in each cell into data according to whether an object exists in each cell. And the quad tree processing unit 230 makes and stores the data of each cell as a tree. The area calculation unit 240 may calculate the area of the object in the image using the size and number of cells determined to have the object. The controller 250 may control the number of times of image segmentation or the depth and level of cells in accordance with the received user signal in relation to object detection.

When an image to be analyzed is input, the image analyzer 210 divides the corresponding image into 4 cells, and divides the image into 4 cells again if it is determined that an object exists. The depth of the cell generated by the first division becomes 1, and if the cell generated by the first division is divided once more, the depth of the cell generated by the second division becomes 2. As the division is repeated, smaller cells are created and the depth of the corresponding cells increases.

As the division of the image is repeated, when each cell of the internal node of the tree corresponding to the image is space-divided again, each cell has four child nodes. Also, when the divided cells are divided again, child nodes may also have child nodes.

The image conversion unit 220 converts images into data according to whether an object exists or not. In particular, data values of corresponding cells are calculated as 0 or 1 according to whether an object exists in each cell in the image. For example, if there is an object in the corresponding cell, the data value of the corresponding cell becomes 1, and if there is no object, the data value of the corresponding cell becomes 0. Accordingly, the image to be analyzed may be converted into a form of a two-dimensional data sheet in which a value of 0 or 1 is written in each cell.

The quad tree processing unit 230 may store an image converted into a set or array of 2D data as a tree. In the embodiment of the present invention, since an image is divided into 4 cells, and all cells subject to division are divided into 4 sub-cells, each node in the tree also has 4 child nodes. Therefore, a tree generated according to an embodiment of the present invention becomes a quad tree.

In order to specify the location of each cell, a cell location value is designated for each cell. Depth 1 means that the image is segmented once, and through depth 1 segmentation, four cells of level 1 are created. Through division of depth 1, each cell has cell position values of 1, 2, 3, and 4.

Also, depth 2 means that the image is divided once more and the image is divided a total of two times, and means that each of the four cells of level 1 is divided into four sub-cells once more. Through division of depth 2, a total of 16 cells, 4 cells of level 2 are created for each cell. Cells of level 2 generated through division of depth 2 have cell position values from 5 to 20.

In particular, cells of level 2 generated by dividing a cell of level 1 having a position value of 1 into a depth of 2 have cell position values of 5, 6, 7, and 8, respectively. Level 2 cells generated by dividing a cell having a position value of 2 into a depth of 2 have cell position values of 9, 10, 11, and 12, respectively. Level 2 cells generated by re-division of level 1 cells with position value 3 to depth 2 are 13, 14, 15, 16, and level 2 cells generated by re-division of level 1 cells with position value 4 to depth 2. It has cell position values of 17, 18, 19, and 20.

In addition, child cells generated through re-division of depth 3 have cell position values from 21 to 84. The position values of the lower level cells are sequentially designated according to the descending order of the position values of the upper level cells.

The division depth and level of each cell are applied as they are to the tree level during quad tree conversion. Accordingly, cells of level 1 generated by splitting depth 1 are converted to each node of tree level 1, and cells of level 2 generated by splitting depth 2 are converted to each node of tree level 2.

The position value increases as the division is repeated, and is numbered cumulatively from the division of depth 1. For example, the cell position value of the cell (2-4-3) in the second position in the image in the depth 1 segmentation, the fourth in the cell in the depth 2 re-segmentation, and the third position in the cell in the depth 3 re-segmentation is the cell position value of the level 1 cell. The sum of the number + the number of level 2 cells + the number of level 3 cells is 4 + 16 + (16 + 12 + 3) = 51. The position value of the corresponding cell is 51 when expressed in decimal and 243 when expressed in quaternary.

The position values of the re-divided lower level cells (child cells) are designated in descending order of the position values of the higher level cells (parent cells). And the position values of the cells are numbered in order from upper left to lower right. Depths after 4, 5, ? , n are also numbered by the same method.

The area calculation unit 240 may calculate the area of the object by multiplying the size and number of cells in which the object is determined to exist, that is, the data value is 1. However, since an object does not always have a shape that is easy to calculate an area for, such as a triangle or a circle, it is not easy to obtain the area as the shape of the object becomes more complex. If an object is searched for in an image in units of divided cells through the technique proposed in the embodiment of the present invention, it is easy to calculate the area of an object because the area is obtained by multiplying the depth and the number of cells in which the object exists.

Cells of depth 1 have an area of 1/4 of the entire image, cells of depth 2 have an area of 1/16 of the entire image, and cells of depth 3 have an area of 1/64 of the entire image. By multiplying the number of existing cells by the area, the area of the object in the image can be obtained.

The controller 250 may control the number of re-divisions according to the depth setting. In addition, the control unit 250 may determine the type of result to be output through the result output unit 140 according to a user signal or a preset bar. For example, the control unit 250 outputs a result such as information about the presence or absence of an object, information about the location of an object in an image, information about an area of an object in an image, information about the date and time of an object existing in a space, etc. ) to determine the result data to be output.

Meanwhile, in an embodiment of the present invention, the terminal node stores whether an object exists in a corresponding quadrant or not. In addition, the terminal node may be referred to as BLACK if it exists, and may be referred to as WHITE if it does not exist, depending on whether objects exist in the entire corresponding quadrant. Also, all internal nodes may be referred to as GRAY.

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

As described above, an embodiment of the present invention proposes a technique for roughly quickly storing the distribution of objects detected by various methods. As described above, we propose a quad tree-based object search technique that divides the image into four cells of the same size to search the distribution of objects and represents the distribution of objects as a quad tree data structure.

According to this technique, the detection target image is divided into 4 cells of the same size to determine whether there is an object in each cell, and if an object exists in a specific cell, the cell is divided into 4 cells of the same size to search for the distribution of the object. .

This process proceeds to the depth of the quad tree that the user wants to search, and the distribution of objects, that is, whether or not an object exists in a specific space is stored as a quad tree. The user can use it to figure out the approximate distribution of objects through the stored result, the quad tree.

Hereinafter, an object detection method according to this technique will be described with reference to FIG. 3 .

The object detection apparatus 100 receives an image that is an analysis target for object detection (S310). The image to be analyzed may be input from another external device or may be an image directly captured and generated by a photographing module of the object detection device 100 .

The object detection apparatus 100 quad-divides an image to be analyzed (S320). Each time an image is segmented, 4 cells are generated. During the first division of the image, 4 cells are generated, and during the second division, each cell is further divided into 4 sub-cells. Primary division means division of depth 1, and cells generated through this become cells of level 1. Secondary division means division of depth 2, and cells generated through this become cells of level 2.

When division is performed up to a user-requested depth or a preset depth, data for each cell is calculated according to whether an object exists in each cell (S330).

The object detection apparatus 100 determines whether an object exists in each cell according to the data value, and determines whether to re-divide the cell in which the object exists according to the result (S340). The object detection apparatus 100 further creates four sub-cells for each cell by re-dividing a cell in which an object exists among cells and a target for re-division is designated (S350).

When cell division and re-division and data calculation are completed, the object detection apparatus 100 generates a quad tree using cells for each depth and level (S360).

Then, when information on the area of the object in the image is requested, the object detection apparatus 100 may calculate the area using the area and number of cells in which the object exists, and calculates the location value of the cells in which the object exists. Through this, the location of the object can also be derived (S370).

The object detection apparatus 100 may output an object detection result including information about the existence of an object, the area and/or location of the object (S380).

4 is a diagram illustrating a process of cell arrangement and 2D conversion of an image according to an embodiment of the present invention. 5 is a diagram for explaining tree transformation according to an embodiment of the present invention.

Figure 4 (a) shows that one analysis target image is divided and arranged into a plurality of cells. And FIG. 4(b) shows a data sheet displaying the presence or absence of an object in each cell with respect to each of the divided cells in FIG. 4(a). That is, referring to (b) of FIG. 4 , it can be seen that the image has been converted into a data sheet through segmentation and data conversion corresponding to the presence or absence of objects in each cell.

According to the data sheet shown in (b) of FIG. 4, if an object exists in a corresponding cell, the data value of the corresponding cell is 1, and if no object exists in the corresponding cell, the data value of the corresponding cell is displayed as 0.

And FIG. 4(c) shows the position value of each cell of various levels by repeating re-segmentation according to the presence or absence of an object for each cell.

5 shows a quad tree in which cells having various levels divided into various depths are generated according to the depth, level, and structure. The level of each cell corresponds to the tree level of the quad tree. The quad tree can simultaneously store and display data values for each cell along with location values for each cell. In this case, each node of the quad tree may be written together with the location value of each cell (including sub-cells).

In addition, whether or not an object exists in each cell may be indicated by the color of a node of the quarter tree corresponding to each cell. For example, a node of a tree may be displayed in black or white, and information on the presence or absence of an object in a corresponding cell is displayed according to the color of the node. A black node may mean that an object exists in a cell corresponding to the corresponding node, and a white node may mean that an object does not exist in a corresponding cell.

The same applies to the process of converting the data sheet to a quad tree. If an object exists in a specific cell, a node corresponding to the corresponding cell may be displayed in black, and if an object does not exist in another specific cell, a node corresponding to the corresponding cell may be displayed in white.

Referring to FIG. 5 , since an object exists in a cell with a position value of 1 and additional division is required, a node corresponding to a cell with a position value of 1 is displayed in gray. In addition, no object exists in cells with position values 5 and 6, and objects exist in cells with position value 8. And the cell with position value 7 is re-divided. Objects do not exist in cells with position values of 29, 30, and 31, which are lower cells of the cell with position value 7, and objects exist in lower cells with position value 32. Cells with position values 2, 3, 4, 13, 14, 15, 16, 65, 66, 67, and 68 can also guess whether or not there is an object in the cell according to the color of each node.

Referring to (c) of FIG. 4 , according to the quad tree-based object search technique according to an embodiment of the present invention, a space photographed as an image is divided into four spaces of the same size. Then, it is determined whether there is an object in each space or not.

If an object exists in a specific space, that is, a partial area within a specific cell, the object detection apparatus 100 divides the corresponding space into four spaces of the same size and searches for it. This process is repeated until the quad tree depth that the user wants to search, that is, the division depth, and the distribution of objects—whether an object exists in a specific space or not is stored as a quad tree.

According to an embodiment of the present invention, the object detection apparatus 100 may receive a raster image as a spatial input. Also, depending on the actual use environment, if pixels with specific RGBA values are converted to True or False and stored as a 2D array, the desired part of the raster image can be saved as an object.

In the experiment, the transparency of the raster image is 0, that is, the transparent pixels are converted to False, and the transparency is from 1 to 255, that is, the opaque pixels are converted to True. Save as In a two-dimensional array, False means that an object does not exist at a pixel with a corresponding index, and True may mean that an object exists at a pixel with a corresponding index.

<Table 1> below shows an algorithm implemented using a recursive call to a quad tree-based object search technique in space. And <Table 2> shows the description of the variables used in the algorithm of <Table 1>.

According to <Table 1> below, the quadtree_search() function calculates the range of x coordinates to be searched, the range of y coordinates to be searched, the coordinates where the object was found at the upper level, the upper level, and the space to be searched are a one-dimensional array It receives as an argument which index of the quad tree managed by .

Pixel search is performed from top to bottom and from left to right through the for statement in row 10 of <Table 1>. If an object is found in the space to be searched (row 11 of <Table 1>), the 1D array value of the quad tree at the corresponding index is set to True (row 12 of <Table 1>).

Also, if the tree level currently being searched has not reached the tree level set by the user (line 13 of <Table 1>), the quadtree_search() function is recursively called according to lines 15 to 18 of <Table 1>. This means that, from a logical point of view, as the division depth increases and the level of the tree or lower cells increases, the space in which the object is found is divided into four spaces of the same size, and each space is searched.

<Table 1>

<Table 2>

6 is a diagram illustrating a search target space on a coordinate plane according to an embodiment of the present invention. 7 is a diagram illustrating a quad-tree based object search technique according to an embodiment of the present invention, and FIG. 8 is a diagram showing a search target space in a tree structure with a depth of 1 according to an embodiment of the present invention.

However, if the coordinates searched at the upper level are searched once again at the lower level, the search time is increased, which is inefficient because the search is performed redundantly. To prevent this redundant search problem, the coordinates that were being searched at the upper level (current_x, current_y) and the coordinates to be searched at the current level (temp_x, temp_y) were distinguished (row 1 of <Table 1>), and Through lines 2 to 9 of 1>, coordinates to be searched can be reset.

First, if an object is not found in space ① or space ① while searching the entire space in the image in FIG. 6 and an object is found in space ② or ④, the function recurses according to rows 15 to 18 of <Table 1>. is called Here, ① called by line 15 of <Table 1>, ③ called by line 17 of <Table 1>, the quadtree_search() function corresponding to the space, because the current_x coordinate exceeded max_x (2 in <Table 1>) row) The data value is stored as False in the quad tree array without searching (row 3 of <Table 1>) and returned (row 4 of <Table 1>).

Second, if an object is found in ① or ③ space while searching the entire space in the image in FIG. 6, the function is recursively called according to rows 15 to 18 of <Table 1>. Here, ② or ④ space has not been searched yet (row 8 of <Table 1>), so start searching from (min_x, min_y) of each space (row 9 of <Table 1>).

Third, if an object is found in ① or ② space while searching the entire space in the image in FIG. 6, the function is recursively called according to rows 15 to 18 of <Table 1>. Here, if an object is found in ① space, ③ in space, ② if an object is found in space, ④ in space, the current_x coordinates left of the search are completed. Therefore, the search can be started from the min_y coordinates of ③ or ④ space (row 6 of <Table 1>).

Fourth, if an object is found in ③ or ④ space while searching the entire space in the image of FIG. 6, the function is recursively called according to rows 15 to 18 of <Table 1>. Here, if an object is found in ③ space, ① space is also searched, and if an object is found in ④ space, search is completed up to current_x coordinates in ② space as well. Therefore, move the temp_x coordinate by one pixel to the right and start searching from the min_y coordinate of ① or ② space (row 7 of <Table 1>).

6 and 7 show examples of searching a space in a quad tree having a tree level of 1. The first space node shown in FIG. 7(c) and FIG. 8 corresponds to the fourth case, and the second and fourth space nodes correspond to the second case.

and Figure 9 illustrates a saturation tree according to an embodiment of the present invention. Referring to FIG. 9, when the quadtree_search() function is executed, a one-dimensional array of Boolean data type logically containing quad tree information as shown in FIG. 9 is returned.

The one-dimensional array stores the tree as a saturated tree, and the proposed method does not distinguish between GRAY and BLACK nodes.

Therefore, if the value of index i in a one-dimensional array is True, searching for index i*4+n (where n is a natural number between 1 and 4) will search the space of the next level, that is, subcells of the next level. can Also, if the size of the one-dimensional array is smaller than i*4+n, i means a terminal node.

Therefore, the approximate area of the searched object can be calculated by adding all the spatial areas of the values of true indexes corresponding to the terminal nodes in the one-dimensional array. <Table 3> below shows an example of an algorithm for calculating the area of a searched object.

<Table 3>

10 is a diagram showing a test screen of an object search method according to an embodiment of the present invention.

In the experiment according to the embodiment of the present invention, an experiment program was created using Java to load an image in a user interface (UI) environment, and the experiment environment was an Intel i7-5960X 3.00 GHz, 16 GB memory.

When an image containing an object is received in the left white box 1010 of FIG. 10 and the conversion button 1030 in the middle is pressed, the image is analyzed and the object is searched to save the object information as a quad tree, and the right white box 1020 The distribution of objects is drawn in a grid and the result is output.

For reference, this experiment was conducted in a space of 512*512 size. The quad tree level used for search is determined by the value of the combo box 1001 at the top left, and can be set from level 1 to level 7.

The area of the object in the image can be calculated using the algorithm disclosed in <Table 3> and then output to the upper right corner. It is assumed that a delay of 0.001 seconds occurs in searching each pixel, and it is exemplarily assumed that a sleep function is implemented. This is because, if no delay occurs in the experimental environment, the time required to search for each level is around 0.004 seconds, resulting in similar results, making it difficult to compare.

First, the user can bring up an image by pressing the fifth icon 1006 at the top left. In the experimental environment, transparent pixels with an RGBA transparency of 0 are converted to False, and opaque pixels with a transparency of 1 to 255 are converted to True.

Then, pixels with true data values, that is, cells containing objects are displayed in black in the white box on the left. In addition, an object can be drawn in the white box on the left through the first to third icons 1004 on the left, and an image can be initialized through the fourth icon 1005.

11 is a graph showing the average time taken for searching for each tree level and the number of times of searching according to an embodiment of the present invention.

The best case for object search in an image, that is, the case in which object search takes the least amount of time, is when an object exists in the first pixel of the upper left corner of all spaces in the segmented image. According to the graph shown in FIG. 11 , when searching for all pixels and for each level is repeated 100 times, the average of the time taken for searching and the number of pixels to be searched are displayed.

On the other hand, in the worst case of object search in an image, that is, the case where object search takes the longest time, there is no object in the space or the object exists in the last pixel at the bottom right of all divided spaces. The average search time for each level was similar to that of the exhaustive search in FIG. 11, and the number of pixels to be searched was 262144, all the same.

12 is a graph showing an average time taken for a search for each tree level and an average number of searches according to another embodiment of the present invention.

12 shows the average time required to search for an object by analyzing each image for each tree level and the average of the number of searches. Referring to FIG. 12 , as the tree level increases, the search is performed more precisely, but it can be seen that the search time and the number of pixels to be searched increase. Nevertheless, it can be seen that the tree search of level 7 was able to complete the search about 100 seconds faster than the exhaustive search of FIG. 11 .

13 is an experimental image for an object search technique according to an embodiment of the present invention. 14 is a graph showing an average search time for each image when an object is detected according to an embodiment of the present invention.

13 shows 12 test subject images randomly downloaded from pixabay. The content of the experiment is to repeat the search for each image 100 times for each tree level and check the average of the search time and the number of pixels to be searched.

14 shows the average time required to search each tree level for each image and the average number of searches. When the objects are wide or evenly distributed, such as in videos 3 and 11, the search time is short. time appeared long.

This is a result related to the search algorithm proposed through the embodiment of the present invention sequentially searching from the upper left corner to the lower right corner. That is, one of the disadvantages of the sequential search algorithm is that the time complexity is O(N) in the worst case, i.e., the case where the most time is required, and the larger the size of the image or cell compared to the size of the object to be detected, the greater the inefficiency. It can be understood in the same context as the disadvantages.

Through experiments, in the worst case, that is, in the case where the object search takes the most time, the same number of pixels as the total search is searched for a similar time. However, in the best case, that is, in the case where the least amount of time is required, assuming that the level of the tree to be searched is l, the object distribution can be identified by searching 4^l pixels. In addition, it can be seen that the difference increases as the division depth or sub-cell level decreases.

In addition, through experiments using the images of FIG. 13 , it can be seen that, when searching according to the embodiment of the present invention, approximate distribution of objects can be searched more quickly than exhaustive searching. Therefore, compared to searching the entire space, it can be confirmed that the proposed method is more efficient in terms of search time and number of searches because it confirms the approximate distribution of objects.

In this way, according to the embodiment of the present invention, since the distribution of objects is searched only up to the tree level designated by the user, it is suitable for quickly grasping the distribution of objects. In addition, in the case of the object detection method proposed in the embodiment of the present invention, since the speed of object detection is fast, it can be used in the field of view of a robot that needs to determine the next movement path by identifying the location of an obstacle in real time, or in identifying a fire area.

Although the above has been described with reference to preferred embodiments of the present invention, those skilled in the art will variously modify and change the present invention within the scope not departing from the spirit and scope of the present invention described in the claims below. You will understand that it can be done.

100: object detection device
110: video input unit
120: object detection unit
130: memory
140: result output unit

Claims (16)

In the object detection device for receiving an image and detecting an object in the image,
an image input unit that receives the image, which is an analysis target for detecting the object;
a control unit receiving a user signal to control the number of divisions of the image or the depth of a cell;
The image is divided into 4 cells, all cells in the image are searched, and the cell in which the object is present within the area of each cell among the cells is divided into 4 sub-cells more than once to determine the distribution of the object in the image. Object detection unit to analyze;
a memory for storing an analysis result of the image or distribution of the object; and
A result output unit outputting the analysis result for the distribution of the object;
The object detection unit repeats the segmentation of the image by the segmentation depth specified by the user who requested the analysis result for the distribution of the object;
When the object detection unit divides the image two or more times, four cells are generated during the first division of the image, in this case, the division depth of the first division is 1, and the division depth generated by the first division The level of the cell is 1, and when the image is divided n times, each divided cell is divided into four sub-cells n times, the division depth of the n-th division is n, and the n-th division The level of the lower cell generated by is n,
Each of the cells and the sub-cells has a position value corresponding to the division depth at which the cell and the sub-cells are divided and the positions of the cells and the sub-cells in the image, and the position value increases as division is repeated. The position values of the cells are assigned to the position values of the lower cells in the order of decreasing position values of the cell levels.
When the segmentation of the image is completed, the object detection unit generates two-dimensional data indicating whether the object exists through dataization for each cell and each sub-cell,
Creating a quad tree corresponding to the structure of the cell and the sub-cell and storing the data for each cell and the sub-cell together with the quad tree;
When information on the area of the object in the image is requested, the area of the object is calculated using the number and size of the cell and sub-cells in which the object exists, and the cell and sub-cells in which the object exists The object detection device, characterized in that for deriving the location of the object through the location value of the cell.
delete delete delete delete delete delete delete In the object detection method for detecting an object in an image by receiving an image,
receiving the image as an analysis target for detecting the object;
receiving a user signal to control the number of divisions of the image or the depth of a cell;
The image is divided into 4 cells, all cells in the image are searched, and the cell in which the object is present within the area of each cell among the cells is divided into 4 sub-cells more than once to determine the distribution of the object in the image. analyzing;
storing an analysis result of the image for the distribution of the image or the object; and
Outputting the analysis result for the distribution of the object,
Analyzing the distribution of the object may include repeating the segmentation of the image as much as the segmentation depth specified by the user who requested the analysis result of the object distribution, and when dividing the image two or more times, During the first division, four cells are generated, in this case, the division depth of the first division is 1, the level of the cells generated by the first division is 1, and the image is divided into nth division, each The divided cell is divided into four sub-cells n times, the division depth of the n-th division is n, and the level of the sub-cell generated by the n-th division is n,
Each of the cell and the sub-cells
The cell and the sub-cells have a position value corresponding to the division depth and the position of the cell and the sub-cell in the image, the position value increases as division is repeated, and the position value of the cells is The position values of the lower cells are specified in the order of the lower level position values,
When the segmentation of the image is completed, the step of analyzing the distribution of the object generates data indicating whether the object exists in each of the cells and sub-cells,
Generating a quad tree corresponding to the structure of the cell and the sub-cell and storing the data for each cell and each sub-cell together with the quad tree,
When information on the area of the object in the image is requested, the area of the object is calculated using the number and size of the cell and sub-cells in which the object exists, and the cell and sub-cells in which the object exists The object detection method further comprising deriving the location of the object through the location value of the cell.
delete delete delete delete delete delete delete

Images