KR101515926B1 - Method for object detection and apparatus thereof - Google Patents

Abstract

A method and apparatus for detecting a feature similarity-based object are disclosed. An object detection apparatus according to the present invention includes a preprocessor for performing preprocessing on an input image and then dividing the input image into blocks of a predetermined size; A first feature calculating unit for calculating HOG (Histogram of Oriented Gradients) representing a frequency distribution of gradient vectors for blocks of the input image as a first feature; A second feature calculating unit for calculating a second feature based on shape similarity for each block using the symmetry of the object included in the input image; And an object determining unit that determines the target object corresponding to the first characteristic and the second characteristic with reference to the learning data.

Description

TECHNICAL FIELD The present invention relates to an object detecting method and an object detecting method using the object detecting method.

The present invention relates to an object detection method and a detection apparatus using the same, and more particularly, to a method and apparatus for detecting an object based on a shape similarity feature.

Recently, researches have been actively conducted to apply image-based object detection technology to various industrial fields. For object detection, it is important to extract appropriate features. In general, various features such as intensity, edge, corner, color, and disparity are used to improve object detection performance. Among them, the edge is robust against the change of the external brightness, and is one of the features widely used because it contains a relatively large amount of information. As a representative feature generation method using edges, a method of extracting the size and direction of an edge and histogramming it as a feature vector is widely used.

1 shows a method of generating a feature for object detection according to the prior art.

The target object can be detected by comparing the feature vector of the acquired image with the pre-learned data. For this purpose, the features constituting the feature vector must be generated in the acquired image, Describes a method of using Histogram of Oriented Gradients (HOG) among the above-mentioned characteristics.

Referring to FIG. 1, a gradient vector for a pixel is calculated from an input acquired image, an image is blocked, and orientation information for each gradient vector is generated as a histogram for each block, use. This method can be used as a very useful feature in a specific object, that is, an object in which an external form is important, such as a pedestrian.

It is helpful to improve the object detection performance by knowing the degree of directionality of the gradient vector for each pixel in the acquired image. However, when only HOG is used as a feature vector, the accuracy of object detection may be lowered. Particularly, since the outline information of the object is used as it is, it is difficult to improve the detection performance.

SUMMARY OF THE INVENTION The object of the present invention is to provide an object detection method and an object detection method using the object similarity feature of the object to be detected as a feature vector.

The objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided an object detecting apparatus comprising: a preprocessor for performing preprocessing on an input image and dividing the input image into blocks of a predetermined size; A first feature calculating unit for calculating HOG (Histogram of Oriented Gradients) representing a frequency distribution of gradient vectors for blocks of the input image as a first feature; A second feature calculating unit for calculating a second feature based on shape similarity for each block using the symmetry of the object included in the input image; And an object determining unit that determines the target object corresponding to the first characteristic and the second characteristic with reference to the learning data.

In a preferred embodiment, the second feature calculator divides the input image into left and right halves and performs mirroring to invert left and right images of either the left or right image.

In addition, the second feature calculating unit calculates the second feature by measuring the degree of similarity of the HOG with respect to the one side image and the other side image on which the mirroring is performed.

The second feature calculating unit compares the HOG of the first block of the one-side image with the HOG of the second block located at a position corresponding to the first block of the other image to measure the similarity.

In one embodiment, the second feature calculator measures the similarity of the HOG using one of L1-norm, L2-norm, histogram intersection, and Chi-square distance.

In one embodiment, the second feature calculating unit calculates a second feature based on the shape similarity by applying different weights to the positions of the blocks according to the types of objects to be detected in the input image.

In a preferred embodiment, the first feature calculation unit includes a HOG calculation module that calculates the gradient strength and gradient direction for each of the blocks, and calculates the HOG using the gradient strength and the gradient direction.

In a preferred embodiment, the second feature calculating unit includes a mirroring module for dividing the input image into left and right halves, and performing mirroring for inverting left and right images of either the left or right image; And a similarity measurement module that measures a similarity between the HOG of the first block of the one side image and the HOG of the second block that is located at a position corresponding to the first block of the other side image, The degree of similarity is measured by applying different weights according to the type of the object to be detected and the position of the first block and the second block in the input image.

According to another aspect of the present invention, there is provided an object detecting method comprising: pre-processing an input image and dividing the input image into blocks of a predetermined size; Calculating HOG (Histogram of Oriented Gradients) representing a frequency distribution of gradient vectors for blocks of the input image as a first characteristic; A mirroring step of dividing the input image into left and right halves and performing mirroring to invert left and right images of either the left or right image; Calculating a second characteristic by measuring a degree of similarity between a HOG of a first block of the one-side image and a HOG of a second block located at a position corresponding to the first block of the other side image; And determining a target object corresponding to the first characteristic and the second characteristic by referring to the dictionary learning data.

According to the present invention as described above, the accuracy of object detection can be improved by constructing a feature vector using the feature similarity information of the HOG and the object itself.

1 is an exemplary diagram for explaining a method of generating a feature for object detection according to the prior art;
2 is a block diagram showing an object detecting apparatus according to an embodiment of the present invention;
FIG. 3 illustrates an example of generating a HOG feature for pixels constituting an acquired image in an exemplary embodiment of the present invention; FIG.
Figure 4 illustrates an example of generating a shape similarity-based feature in an embodiment of the present invention.
FIG. 5 illustrates an example of generating a weight applied when generating a shape similarity-based feature in FIG. 4; FIG.
6 is a flowchart for explaining an object detection method according to an embodiment of the present invention;

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention and the manner of achieving them will become apparent with reference to the embodiments described in detail below with reference to the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Is provided to fully convey the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims. It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. In the present specification, the singular form includes plural forms unless otherwise specified in the specification.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to designate the same or similar components throughout the drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

2 is a block diagram showing an object detecting apparatus according to an embodiment of the present invention.

2, an object detection apparatus according to the present invention includes an image acquisition unit 110, a feature vector calculation unit 130, a pre-learned data storage unit , 140, and an object judgment unit (150).

The image obtaining unit 110 obtains an image including a target objet from an external device. For example, the image acquisition unit 110 may acquire an image provided from an image pickup apparatus, an image storage apparatus, and the like. The image acquiring unit 110 performs preprocessing on the acquired image.

For example, as shown in FIG. 3 (b), the image obtaining unit 110 performs color correction on an acquired image. A typical method is gamma correction. Also, the image obtaining unit 110 obtains a gradient vector for each pixel of the acquired image, and calculates the magnitude and orientation of the gradient vector (see (c) of FIG. 3). For example, methods such as Sobel, Canny, and the like can be used to obtain gradient vectors. Next, the Gaussian weight is applied to the gradient vector (see FIG. 3 (d)). This is because the strength (size) and the direction of the gradient vector are not compared for each pixel, The feature is to increase the weight, and the feature of the block edge is to lower the weight, so as to improve the object detection performance. The image acquiring unit 110 divides the acquired image into a plurality of blocks and divides the acquired blocks into a plurality of blocks.

For example, as shown in FIG. 3, the image obtaining unit 110 divides an acquired image into left and right halves, divides the obtained image into upper and lower halves, and obtains the acquired images with a total of four halves A, B, C, Can be divided. Then, the image obtaining unit 110 divides each block into blocks of a predetermined size.

Although the segment division and the block segmentation of the image are performed by the image acquisition unit 110, the subject of the performance is not limited to the image acquisition unit 110. [ For the efficiency of the arithmetic processing, a preprocessing process such as partitioning and block division may be performed in the first feature calculating unit or the second feature calculating unit, which will be described later.

The feature vector calculator 130 calculates a feature vector for each block in the acquired image. In general, the feature vector consists only of HOG (Histogram of Oriented Gradients). Here, HOG refers to a histogram representing the frequency distribution of gradient vectors corresponding to a plurality of blocks. However, the feature vector according to the exemplary embodiment of the present invention is calculated using not only HOG but also feature based on shape similarity. Here, the feature based on the shape similarity means the left-right symmetry or the up-down symmetry similarity information of the object to be detected.

The feature vector calculating unit 130 includes a first feature calculating unit 131 and a second feature calculating unit 132. [

The first feature calculating unit 131 calculates the HOG for each block based on the gradient vectors calculated for each pixel to calculate the HOG for each block.

For example, the acquired image is divided into a plurality of blocks. But the number of blocks is not limited to this, and the acquired image may be divided into various numbers of blocks.

The HOGs constituting the feature vector are calculated for each of a plurality of blocks. That is, the number of HOGs constituting the feature vector corresponds to the number of blocks. Hereinafter, the calculation of the gradient vectors used to calculate the HOG for one block will be exemplarily described.

One block includes a plurality of pixels. For example, 16 pixels (4x4 pixels) may constitute one block, and it is not limited thereto and may include various numbers of pixels. For example, the size of one block may be defined as 8x8 pixels, 16x16 pixels.

Gradient vectors are calculated for each of a plurality of pixels. However, since the pixels adjacent to the pixel corresponding to each gradient vector (hereinafter referred to as " target pixel ") are considered at the time of calculating each gradient vector, the calculation of the gradient vector is performed for the pixels positioned at the edge of the block Can be omitted.

The gradient vectors for the plurality of pixels are used to calculate the HOG. The rank of HOG means the quantized interval for the range of gradient directions (0 to 180 degrees). The frequency of the HOG means the cumulative value of the gradient intensities according to the class of the HOG.

For example, the HOG may be implemented to have eight ranks, the number of ranks of the HOG is not limited thereto, and the HOG may have various ranks.

For example, for the pixels constituting any one block, 12 gradient vectors having a gradient direction of 0 degrees and 4 gradient vectors having a gradient direction of 90 degrees can be calculated. At this time, the HOG for this block will be generated as the frequency distribution of pixels for the gradient direction of each class.

The second characteristic calculating unit 132 calculates the second characteristic based on the shape similarity based on the symmetry of the object included in the acquired image.

4 (f), the second feature calculating unit 132 may invert the left and right images of the left image or the right image in the input image divided into left and right half sections, And calculates the second characteristic by measuring the degree of similarity of the HOG with respect to the one side image and the other side image on which the mirroring is performed.

At this time, the second feature calculating unit 132 compares the HOG of the first block of the one-side image with the HOG of the second block located at the position corresponding to the first block of the other image, , And the degree of similarity may be measured by any one of L1-norm, L2-norm, histogram intersection, and Chi-square distance.

Referring to FIG. 4 (f), the second feature calculating unit 132 first performs mirroring on the regions B and D. FIG. Then, HOG is calculated for each block to compare the HOG for each block. The comparison target corresponds to the order of blocks in the block as shown in the figure. In other words, the A zone and the B zone, the C zone and the D zone are compared. In each zone, the HOGs of blocks corresponding to each other are compared according to the positions of the blocks.

At this time, the degree of similarity of the HOG is calculated by applying different weights to the positions of the blocks according to the types of objects to be detected in the acquired image. Thus, the performance of the second feature based on the shape similarity can be improved, and a description thereof will be specifically described with reference to FIG.

FIG. 5 is a view showing an example of generating a weight to be applied when generating a shape similarity-based feature in FIG.

Referring to FIG. 5, the weight includes a weight (a) and an intra-region weight (b) in the horizontal line.

For example, when the object to be detected is a pedestrian, the highest weighting line for each horizontal line is displayed in yellow. In other words, since the outline of the pedestrian is likely to be around yellow, the histogram comparison value in this vicinity is used as important data for object detection. Therefore, it is used to calculate the feature similarity based on the degree of HOG.

Also, as shown in FIG. 5, the weight p is applied differently according to the position d of the horizontal line, or the weight is applied to each vertical block differently. The highest weighted line can be changed and applied to the shape of the object.

In the case where the object to be detected is a pedestrian, the degree of change of the HOG value is large according to the walking level and pose of the pedestrian in the lower half zone (zones C and D) (Areas A and B) do not change relatively strongly. Therefore, when used as a feature value, the HOG comparison value of the upper half body is relatively high (q) and the HOG comparison value of the lower half body is low (r) to fully reflect the overall shape characteristic of the pedestrian. This can also be changed depending on the type and type of the object.

The second feature calculating unit 132 divides the input image into left and right halves and performs mirroring to invert left and right images of either the left or right image, A mirroring module for performing mirroring; And a similarity measuring module for measuring a similarity between the HOG of the first block of the one side image and the HOG of the second block located at a position corresponding to the first block of the other side image.

At this time, the similarity measurement module measures the similarity by applying different weights according to the type of the object to be detected and the position of the first block and the second block in the input image.

The pre-learning data storage unit 140 stores pre-learning data for various objects such as a human body, a vehicle, and the like. The pre-learning data storage unit 140 delivers the pre-learning data to the object determining unit 150 as reference information for determining a target object.

The pre-learning data includes pattern information and identification information of objects. Here, the pattern information refers to information to be compared with the feature vector, and the identification information refers to information for indicating the intrinsic characteristics of the objects such as the name, type, and size of the objects.

On the other hand, the pre-learning data can be generated through iterative learning of a unique pattern of objects. At this time, a machine learning algorithm such as Adaboost, SVM (Support Vector Machine) and the like can be applied to generate the prior learning data.

The object determining unit 150 determines what the target object corresponds to the feature vector by referring to the prior learning data. Then, the object determining unit 150 outputs the determination result for the target object. For example, when the pattern information matching the feature vector is included in the prior learning data, the object determining unit 150 can output the detection success notice information and the identification information of the target object corresponding to the pattern information as a determination result have. Alternatively, when the pattern information matching the feature vector is not included in the prior learning data, the object determining unit 150 may output the detection failure notification information as the determination result. The determination result may be provided to an external display device, a storage device, or the like.

6 is a flowchart illustrating an object detection method according to an embodiment of the present invention.

Referring to FIG. 6, in step S10, an image including a target object is acquired, and a preprocessing process is performed on the acquired image in S20. For example, a preprocessing process such as calculating a gradient vector for each color correction pixel, applying a Gaussian weight to a gradient vector, and dividing an acquired image may be performed on the acquired image, and a specific description thereof has been described with reference to FIGS. 2 to 5 It will be omitted in the following.

In step S30, a feature vector for object detection is calculated. More specifically, the step S30 may include calculating HOG (Histogram of Oriented Gradients) representing a frequency distribution of gradient vectors for blocks of the input image as a first characteristic, A mirroring step of dividing the left side image and the right side image of the one-side image into half sections and performing mirroring to invert left and right images of either the left image or the right image; And calculating a second characteristic by measuring a degree of similarity between the HOGs of the second block at a position corresponding to the first block.

The degree of similarity may be measured by applying different weights according to the type of the object to be detected and the position of the first block and the second block in the input image.

In step S40, referring to the pre-learning data, a target object corresponding to the first characteristic and the second characteristic is determined.

Meanwhile, the object detection method according to the present invention can be implemented as a computer-readable code on a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording media storing data that can be decoded by a computer system. For example, there may be a ROM (Read Only Memory), a RAM (Random Access Memory), a magnetic tape, a magnetic disk, a flash memory, an optical data storage device and the like. The computer-readable recording medium may also be distributed and executed in a computer system connected to a computer network and stored and executed as a code that can be read in a distributed manner.

 It will be understood by those skilled in the art that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. The scope of the present invention is defined by the appended claims rather than the detailed description, and all changes or modifications derived from the scope of the claims and their equivalents should be construed as being included within the scope of the present invention.

110: image acquiring unit (preprocessing unit) 130: feature vector calculating unit
131: First feature calculating unit 132: Second feature calculating unit
140: dictionary learning data storage unit 150: object determination unit

Claims (10)

A preprocessor for performing preprocessing on the input image and dividing the input image into blocks of a predetermined size;
A first feature calculating unit for calculating HOG (Histogram of Oriented Gradients) representing a frequency distribution of gradient vectors for blocks of the input image as a first feature;
A second feature calculating unit for calculating a second feature based on shape similarity for each block using the symmetry of the object included in the input image; And
And an object determining unit that determines a target object corresponding to the first characteristic and the second characteristic by referring to the prior learning data,
The second feature calculating unit may calculate the second feature based on the shape similarity by applying different weights to the positions of the blocks according to the types of objects to be detected in the input image
/ RTI > The apparatus according to claim 1,
Mirroring is performed to invert the left and right images of either the left image or the right image in the input image divided into left and right half sections
/ RTI > 3. The apparatus according to claim 2,
And calculating the second characteristic by measuring the degree of similarity of the HOG with respect to the one side image and the other side image on which the mirroring is performed
/ RTI > 4. The apparatus according to claim 3,
And comparing the HOG of the first block of the one side image with the HOG of the second block located at the position corresponding to the first block of the other side image to measure the similarity
/ RTI > 4. The apparatus according to claim 3,
The similarity of HOG is measured by using one of L1-norm, L2-norm, histogram intersection and Chi-square distance
/ RTI > delete 2. The apparatus according to claim 1,
And a HOG calculation module for calculating a gradient strength and a gradient direction for each of the blocks and calculating the HOG using the gradient strength and the gradient direction
/ RTI > The apparatus according to claim 1,
A mirroring module for dividing the input image into left and right halves and performing mirroring for inverting left and right images of either the left or right image; And
And a similarity measuring module for measuring a similarity between the HOG of the first block of the one-side image and the HOG of the second block located at a position corresponding to the first block of the other image,
The similarity measurement module measures the similarity by applying different weights according to the type of object to be detected and the position of the first block and the second block in the input image
/ RTI > Performing preprocessing on an input image and dividing the input image into blocks of a predetermined size;
Calculating HOG (Histogram of Oriented Gradients) representing a frequency distribution of gradient vectors for blocks of the input image as a first characteristic;
A mirroring step of dividing the input image into left and right halves and performing mirroring to invert left and right images of either the left or right image;
Calculating a second characteristic by measuring a degree of similarity between a HOG of a first block of the one-side image and a HOG of a second block located at a position corresponding to the first block of the other side image; And
Determining a target object corresponding to the first characteristic and the second characteristic by referring to the prior learning data
Gt; 10. The method of claim 9, wherein the calculating the second characteristic comprises:
And measuring the degree of similarity by applying different weights according to the type of the object to be detected and the position of the first block and the second block in the input image
In object detection method.

Images