KR101491906B1 - Method for enhancing the fastness of feature matching with pre-determined classification - Google Patents

Abstract

The present invention relates to an image processing method, comprising: an image input processing step of processing a signal input through an image input unit into image information; a feature extraction step of extracting and confirming a feature including a corner feature from the image information; A characteristic classification step of classifying the features according to a predetermined corner feature classification criterion and a feature matching step of matching using the corner features classified in the feature classification step.

Description

METHOD FOR ENHANCING THE FASTENESS OF FEATURE MATCHING WITH PRE-DETERMINED CLASSIFICATION [0002]

The present invention relates to an image recognition method, and more particularly, to a matching speed enhancement image recognition method capable of enhancing a matching speed by eliminating an unnecessary matching process between features.

These features, especially the corner features, are extracted from the intersections of the straight lines or the vertices of the regions, and are used in various fields such as position estimation of the mobile robot using a camera, object recognition Is widely used.

In the field of position estimation, mainly the corner features extracted from the current image and the previous image are matched and tracked to detect the movement of the camera. In particular, such position estimation and the like are actually being applied and actively studied in a robot.

Movable mobile robots have been studying techniques for accurately recognizing their own positions using the surrounding environment and for creating a map for enabling them to grasp their own position based on a landmark as a reference.

In particular, information having a specific shape such as a vertex or a straight line among information acquired from the surrounding environment is called a feature, and a map created by using these features is called a feature map. Since the feature is used as a marker for recognizing the position of the robot in the environment, it is possible to create an accurate feature map such as which feature to extract and use, and whether the extracted feature has robust characteristics according to the change of surrounding environment It becomes an important factor.

However, in order to confirm the position of the mobile robot itself or to locate the moving object of the object in the case of tracking the object, the feature is extracted from the image information and compared with the data, do.

However, in the related art, unnecessary arithmetic operations have occurred when matching the feature extracted from the image information with the feature information extracted and stored in the data storage, thereby exposing limitations that make it difficult to perform rapid position recognition or object recognition.

Accordingly, the present invention allows a fast and accurate matching process to be performed through processes such as classification of a corner feature as a typical characteristic of features, thereby achieving matching speed enhancement image recognition And a method thereof.

According to an aspect of the present invention, there is provided an image processing method including: an image input processing step of processing a signal input through an image input unit into image information; a feature extraction step of extracting and confirming a feature including a corner feature from the image information; A feature classification step of classifying the corner feature extracted in the feature extraction step according to a predetermined corner feature classification criterion and a feature matching step of matching using the corner feature classified in the feature classification step .

Wherein the feature extraction step comprises: a virtual round setting step of setting a virtual round of a predetermined radius with respect to a unit pixel in the image information; a virtual round setting step of setting a virtual round in the arbitrary pixel in the image information, And a corner feature identification step of forming the virtual round around the pixel to compare the arbitrary pixel with the pixel in the virtual round to check whether or not the corner feature exists.

In the matching speed enhancement image recognizing method, the corner characteristic checking step may include: a center unit pixel setting step of setting any pixel in the image information as a center unit pixel; and a step of setting the virtual round And comparing the brightness of the center unit pixel with the brightness of the virtual round pixel disposed around the center unit pixel; and controlling the brightness of the center unit pixel and the virtual round- (Ncp), which has a brightness difference equal to or greater than a preset brightness difference stored in the storage unit, wherein the brightness of the pixel on the virtual round is continuous, (Ncp) of the virtual round consecutive pixels, A corner characteristic extraction extraction step of extracting a corner feature by comparing the number of corners with the number of corner feature judgment pixels Nc stored in the corner feature extraction judgment step; (Ndp) stored in the storage unit to determine whether or not the extracted corner feature is directed through the arc shape of the virtual round pixels constituting the number of consecutive virtual round pixels May be included.

Wherein the directionality confirmation step comprises: a step of determining a virtual round-shaped virtual round continuous pixel number (Ncp) for a center unit pixel of the corner feature extracted at the corner feature extraction determination step to the directional confirmation pixel number (Ndp), comparing the number of directional verifying pixels in the directional verifying pixel number comparing step with the number of directional verifying pixels (Ndp) in the virtual verifying step, Determining whether the extracted corner feature exists as a directionality; determining whether the extracted corner feature exists as a directionality; determining whether the extracted corner feature exists as a directionality; The center of the pixel which constitutes the number of pixels is the directional Set to the central pixel, and may comprise an orientation setting step of setting the line segment to form a center line connecting the center of the unit pixels of the center pixel and the directional characteristics of the extracted corner by bangwiseon representing the orientation.

The matching speed enhancement image recognizing method may further comprise a direction alignment step of aligning the diagonal line set in the orientation setting step so as to match the preset reference line.

In the matching speed enhancement image recognition method, the corner feature classification step may include: comparing and determining the brightness of the virtual round pixels in which the brightness of the central unit pixel of the extracted corner feature is the number of consecutive virtual round pixels, A corner discrimination judgment step of discriminating the corners of the corners according to the characteristics of the corners of the corners and the dark corners of the corners of the corners, Step < / RTI >

The matching speed enhancement image recognizing method according to the present invention having the above-described configuration has the following effects.

First, in the matching speed enhancement image recognition method according to the present invention, the feature classification step according to a predetermined classification criterion is performed with respect to the corner feature among the features, so that unnecessary registration process is excluded, .

 Second, the matching speed enhancement image recognition method according to the present invention can be applied to various matching methods for corner features such as FAST and AGAST, thereby extending the usability.

Thirdly, the matching speed enhancement image recognizing method according to the present invention may include the orientation information on the corner feature extracted in the preset feature classification reference to achieve the matching speed enhancement through the segmented classification reference.

Fourth, the matching speed enhancement image recognizing method according to the present invention includes a step of aligning a corner feature having azimuth information with respect to the azimuth direction, so that a strong and fast matching process is also performed on the rotation state of the corner feature extracted from the image information It can be done.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

1 is a schematic block diagram of a matching speed increasing recognition method according to an embodiment of the present invention.
2 is a schematic control flowchart of a matching speed increasing recognition method according to an embodiment of the present invention.
3 is a flowchart of an image input processing step of the matching speed increasing recognition method according to an embodiment of the present invention.
FIG. 4 and FIG. 5 are schematic diagrams of an image obtained by the image input unit of the matching speed enhancement recognition method according to an embodiment of the present invention and an image obtained by performing the distortion correction.
6 is a flowchart of a schematic feature extraction step of the matching speed enhancement recognition method according to an embodiment of the present invention.
7 is a schematic diagram illustrating a virtual round used in the feature extraction step of the matching speed enhancement recognition method according to an exemplary embodiment of the present invention on the image information.
FIG. 8 is a flowchart illustrating a detailed process of calculating the number of consecutive pixels of the matching speed increasing recognition method according to an exemplary embodiment of the present invention.
9 is a flowchart of a corner feature extraction determination step of the matching speed enhancement recognition method according to an embodiment of the present invention.
10 is a flowchart of a direction confirming step of the matching speed increasing recognition method according to an embodiment of the present invention.
11 is a flowchart of alignment alignment steps of a matching speed enhancement recognition method according to an embodiment of the present invention.
12 is a flowchart of a feature classification step of the matching speed enhancement recognition method according to an embodiment of the present invention.
13 is a schematic diagram showing a direction alignment process for a corner feature having directionality of the matching speed enhancement recognition method according to an embodiment of the present invention.
FIG. 14 is a schematic diagram of a classification criterion type according to a preset corner feature classification criterion such as the contrast and azimuth of the matching speed enhancement recognition method according to an embodiment of the present invention.
FIG. 15 is a diagram showing a classification criterion type according to a preset corner feature classification criterion such as contrast and orientation, and a corner feature extracted from actual image information in the matching speed enhancement recognition method according to an embodiment of the present invention.
16 is a diagram showing types of corner features extracted from image information of a ceiling of a matching speed increasing recognition method according to an embodiment of the present invention.

Hereinafter, the matching speed increasing recognition method of the present invention will be described with reference to the drawings.

FIG. 1 shows a position-aware mobile robot apparatus 10 (hereinafter referred to as a mobile robot apparatus) for executing a matching speed increasing recognition method according to an embodiment of the present invention, which is an example of implementing the matching speed increasing recognition method of the present invention The present invention is not limited to the position recognition of the mobile robot apparatus. The mobile robot apparatus 10 includes an image input unit 100, an encoder sensing unit 200, a control unit 300, a storage unit 400, an operation unit 500, and a driver 600. The controller 300 electrically communicates with other components to receive an input signal and apply a control signal to each component. The image input unit 100 may have a variety of configurations. In the present exemplary embodiment, the image input unit 100 may include a monocular camera, but may be implemented as a stereo camera as an embodiment of the present invention. When the image input unit 100 is implemented as a monocular camera, an image obtained through the image input unit 100, for example, a ceiling image, is easily distinguished from an object by the image input unit 100, not. The operation unit 500 performs a predetermined operation process in accordance with a control signal of the control unit 300. The storage unit 400 stores various predetermined values in advance and stores necessary image information or location information in accordance with a control signal of the control unit 300. [ Stores information. The driving unit 600 is implemented by an electric motor or the like and is driven according to a control signal of the control unit 300 to provide a driving force capable of moving the matching speed increasing recognition method 10 to a desired position. The encoder sensing unit 200 senses the rotation speed and the rotation angle of a driving wheel (not shown) driven by the driving unit 300, and the rotation angle is calculated and derived according to the difference of the rotation speeds of the respective driving wheels Lt; / RTI >

The mobile robot apparatus 10 can perform a more accurate and rapid position recognition process by using the matching speed enhancement recognition method of the present invention in performing the conventional position recognition process and simultaneously performing the feature matching process for position recognition have.

FIG. 2 is a flowchart showing a schematic control process of the matching speed increasing recognition method according to an embodiment of the present invention. When the image input unit 100 is used as a monocular camera, The input processing step S10 may be further executed. The control unit 300 receives the ceiling image (or image information) from the image input unit 100 and processes the image information (S10). As described above, before the image input processing step S10 is executed, the matching speed increasing recognition method 10 is provided, and the notation is omitted in the flow chart of FIG.

3 shows an example of a more specific flow of the image input processing step S10. The control unit 300 applies a control signal to the image input unit 100, and the image input unit 100 implemented by the single- The control unit 300 can acquire image information about an object or an external environment such as a ceiling at the position of the speed enhancement recognition method 10 and input the information to the control unit 300 at step S11. The control unit 300 determines whether the input image or image information is distorted (S12). If there is image distortion in the image or the image information, the control unit 300 transmits the image or image information to the operation unit 500 to perform image correction Step S13 may be performed.

4 and 5 show schematic examples of images before and after image correction. An image input unit 100 implemented by a monocular camera extracts characteristics of an input image, for example, a ceiling image, and continuously The image information obtained to obtain a wide field of view is distorted. The operation unit 500 processes the image or image information inputted from the image input unit 100 according to the control signal of the control unit 400 to correct the distortion phenomenon. If it is determined that the image correction is completed or the image correction is unnecessary, the controller 300 outputs the corrected image or image information to the controller 300 (S14). The correction in the image input processing step (S10) is an example, and the present invention is not limited to this. In other words, it is possible to simply acquire the peripheral image through the image input unit 100, and to convert the peripheral image into the image information.

Then, the feature extraction step S20 is executed, and a predetermined extraction step is executed in accordance with a control signal for executing the feature extraction step of the control unit 300. [ More specifically, the control unit 300 extracts features existing in the current image based on the corrected image, such as image distortion. In some cases, the feature may be a door feature, an illumination feature, and the like. However, S20 may include a step of extracting a corner feature. However, when a ceiling image is used to recognize the position of a mobile robot as an example of the present invention, among the features obtained in the feature extraction step performed in the matching speed enhancement recognition method, And the like.

The feature extraction step S20 may include a door feature extraction step S25, an illumination feature extraction step S27 and an arbitrary shape feature extraction step S29 as shown in FIG. 6, Step S20 is centered on the corner feature extraction step S21 for extracting the corner feature according to a predetermined corner classification classification criterion.

The corner feature extraction step S21 includes a virtual round setting step S22 and a corner feature identification step S23. In the virtual round setting step S22, a unit pixel constituting image information according to a control signal of the controller 300 (VR) of a preset radius. FIG. 7 shows a virtual round (VR) formed for any unit pixel in the image information in the virtual round setting step S22 of the corner feature extraction step S21, wherein the preset radius of the virtual round (VR) Unit 400 as preset data. In the present embodiment, the preset radius is formed by 3 pixels, and FIG. 7 shows an example of a virtual round (VR) when 3 pixels are set as the preset radius. A virtual circle having a radius of 3 pixels can be formed around any unit pixel indicated by the center unit pixel C as shown in partly enlarged view (a) of FIG. In this case, since the left and right upper and lower sides forming the virtual round VR are formed by three pixels of a straight line and are arranged diagonally between the respective sides, the virtual round VR of the pre-set radius of 3 pixels according to the present embodiment A total of 16 pixels form the circumference of the virtual round (VR). Therefore, when the preset radius of the virtual round is adjusted, the number of pixels on the circumference constituting the virtual round (VR) can also be changed.

Then, the controller 300 executes the corner feature identification step S23. In the corner feature identification step S23, any pixel in the image information and a virtual round (VR) centering on the thus selected pixel And compares the arbitrary pixel with the pixel on the virtual round (VR) to confirm whether or not the corner feature exists.

More specifically, the cotter characteristic confirmation step S23 includes a center unit pixel setting step S231, a pixel brightness confirmation step S233, a continuous pixel number calculation step S235, a corner feature extraction determination step S237, (S239).

In the center unit pixel setting step S231, any pixel in the image information is set as the center unit pixel (C). Then, in the pixel brightness check step S233, a virtual round VR having a radius of a predetermined radius is arranged in the outer periphery of the center unit pixel C according to the control signal of the controller 300, The brightness and the brightness of the virtual round (VR) pixels arranged around the center unit pixel C are compared. That is, as shown in FIG. 7, a certain pixel in the image information is set as a center unit pixel C, and a virtual round (VR) having a preset radius around the center is set. The virtual round (VR) occupies The brightness of the pixel, that is, the brightness of the pixel on the virtual round (VR) and the brightness of the center unit pixel (C). Here, for example, any pixel formed with the center unit pixel can be applied to all the pixels in the image information of a single frame, and can be structured so that all of the pixels in the frame are scanned.

The consecutive pixel number calculation step S235 is executed in accordance with the control signal of the control unit 300. In the consecutive pixel number calculation step S235, the brightness of the center unit pixel C is calculated in the pixel brightness confirmation step S233 If there is a brightness difference of the virtual round phase pixel, the number of consecutive virtual round continuous pixels Ncp is calculated (S235).

More specifically, the consecutive pixel number calculation step S235 may include a brightness difference existence determination step S2351, a brightness difference filtering step S2353, and a consecutive pixel number confirmation step S2355, as the case may be.

8, in step S2351, the controller 300 determines whether there is a brightness difference between the center unit pixel C and the virtual round pixels. In step S2351, The control unit 300 proceeds to step S2353. In step S2353, the control unit 300 executes the brightness difference filtering step. The brightness difference between the center unit pixel C and the virtual round pixels in the brightness difference filtering step S2353 is referred to as Ld and the preset brightness difference Lds ) To determine whether the brightness difference Ld is equal to or greater than the predetermined brightness difference Lds. The unnecessary noise or the like can be removed through the brightness difference filtering step S2353. The determination of brightness difference and the brightness difference filtering are performed for each pixel on the central unit pixel and the virtual round, respectively.

Then, in the consecutive pixel number checking step S2355, the controller 300 determines whether the brightness difference Ld from the center unit pixel is greater than the preset brightness difference Lds, And confirms the number Ncp (S2355).

On the other hand, if the brightness difference does not exist in step S2351, or if the brightness difference Ld between the central unit pixel and the virtual round-shaped pixel is equal to or smaller than the preset brightness difference Lds in step S2353, (Step S2357) may be executed to confirm whether or not the corner feature checking step has been executed for any of the pixels in the image information composed of the remaining pixels The control flow proceeds to step S231 to repeat the predetermined corner feature identification step. On the contrary, in step S2357, the remaining pixels are not present and all the pixels in the image information, or the pixels within the preset range, If the feature checking process is completed, the control flow may proceed to step S30.

Then, the controller 300 executes a corner feature extraction determination step S237. In the corner feature extraction determination step S237, the controller 300 stores the virtual round continuous pixel number Ncp in the storage unit 400 The corner feature is extracted by comparing the number of corner feature determining pixels Nc, which is one of the preset data, to be the corner feature (S237). That is, the number of corner feature determining pixels Nc is a virtual round-shaped pixel, and the pixels having the brightness difference Ld between the center unit pixel C and the virtual round pixels are equal to or greater than the preset brightness difference Lds, Means a minimum value of the number of consecutive pixels for judging whether or not an area in a virtual round based on a unit pixel corresponds to a corner feature. Therefore, if the number of consecutive pixels Ncp is greater than the number Nc of corner feature determination pixels in step S2371, a region forming a virtual round around the corresponding central unit pixel can be confirmed and set as a corner feature, Ncp) is equal to or less than the number Nc of corner feature determination pixels, an area forming a virtual round around the corresponding central unit pixel may be identified as a non-corner feature.

Meanwhile, the corner feature checking step S23 may further include a direction confirming step S239 for confirming whether the corner feature extracted in the step S237 is present or not.

In the directionality confirmation step S239, the controller 300 determines whether or not the number of virtual round consecutive pixels Ncp and directionality VR in the virtual round (VR) for the center unit pixel C of the corner feature extracted in the corner feature extraction determination step S237 And compares the number Ndp of confirmation pixels to determine whether or not the extracted corner feature has directionality. The directionality check pixel number Ndp is a value obtained by dividing the number of virtual round consecutive pixels Ncp of the corner feature extracted into one of the preset data stored in the storage unit 400 into a sufficient arc shape, Means a minimum number criterion for successive pixels to determine whether the feature has directionality.

More specifically, the directionality confirmation step S239 may include a directionality confirmation pixel number comparison step S2391, a directionality determination step S2393, and a direction setting step S2395. In the directionality confirmation pixel number confirmation step S2391 The control unit 300 determines the number of virtual round consecutive pixels Ncp and the number of directional confirmation pixels Ndp in the virtual round (VR) for the center unit pixel C of the corner feature extracted in the corner feature extraction determination step S237 ). Here, the directional confirmation pixel number Ndp, which is one of the preset data stored in the storage unit 400, is a value obtained by dividing the number of consecutive virtual round consecutive pixels Ncp in the virtual round VR into a constant arcuate shape It is used as a criterion for judging whether or not it is possible to set various numbers according to design specifications.

After the directionality confirmation pixel count confirmation step S2393 is executed, the controller 300 executes the directionality determination step S2393. The directionality determination step S2393 includes a succession determination step S2393a and a hunting determination step S2393b ). In the continuous determination step S2393a, the control unit 300 compares the number of virtual round continuous pixels (Ncp) on the virtual round (VR) with respect to the center unit pixel C of the extracted corner feature to the directional confirmation pixel number Ndp , If the number of consecutive virtual consecutive pixels Ncp is less than the number Ndp of directional confirmation pixels in the consecutive determination step S2393a, the controller 300 determines that the extracted extracted corner feature does not exist, The process goes to step S2392b to confirm that there is no directionality. On the other hand, if the number of consecutive virtual consecutive pixels Ncp is equal to or greater than the number Ndp of directional confirmation pixels in the consecutive determination step S2393a, the controller 300 determines that the extracted extracted corner feature exists as a directional presence possibility, . The controller 300 confirms whether it is an arc or a circle in the direction confirmation step S2393b, where Nvr represents the number of pixels constituting the virtual round formed for the predetermined radius. That is, in step S2393b, the control unit 300 checks whether the number of virtual round consecutive pixels Ncp is less than the total number of virtual rounds Nvr so that the number of virtual round consecutive pixels Ncp is greater than the total number of virtual round pixels Nvr The control unit 300 determines that the extracted corner feature has directionality (S2394a). If the virtual round continuous pixel number Ncp is equal to the virtual round pixel total number Nvr, The extracted corner feature has a characteristic of a point and it is determined that the characteristic (see Fig. 14 (d)) that no directionality exists has no directionality (S2394b).

If the controller 300 determines that the directionality of the corner feature extracted in the directionality determination step S2393 exists, the control unit 300 may execute the orientation setting step S2395. The center of the pixel constituting the virtual round continuous virtual pixel number Ncp of the corner feature extracted in the orientation setting step S2395 is set to the directional central pixel OC in S2395a, A central line segment C-OC connecting the pixel C and the directional central pixel OC is formed (S2395b), and the center line segment is set as a diagonal line indicating the azimuth (S2395c).

After the direction confirmation step S239 has been executed, the apparatus may further include a direction alignment step S240 for executing a separate orientation alignment process. That is, the step S239 is completed and the controller 300 executes the step S240 to perform the alignment using the calculated orientation line with respect to the extracted corner feature. The directional information derived from the step S239, (S241). When it is confirmed in the directionality confirmation step S241 that directionality exists, the diagonal line obtained for the extracted corner feature is set as a preset reference line, as shown in the figure, For example, in this embodiment, a diagonal line alignment step (S243) is performed to align the 12 o'clock orientation line. Through such an alignment process, a fast and robust registration process can be achieved. Fig. 13 shows a schematic line showing a direction alignment process for a directional corner feature. In (a), the orientation alignment process is not performed. In (b), however, the orientation line is aligned with the 12 o'clock orientation line By aligning it with the rotation transformation as in (c), it is possible to make the matching process faster and also to have a strong characteristic against rotation.

After the feature extraction step S20 including the corner feature identification step S23 is completed, the control unit 300 executes the feature classification step S30. In the feature classification step S30 of the present invention, And a corner feature classifying step S31 for classifying according to a predetermined corner feature classifying criterion. The corner feature classifying step S31 includes a corner lightness classifying step S311 and a direction classifying step S313. In the corner lightness classifying classifying step S311, the controller 300 classifies the center unit pixels (See Fig. 14 (b)) and the dark corner (Fig. 14 (b)) by comparing the brightness of the pixels constituting the virtual round continuous consecutive pixel number Ncp in the virtual round, ). If the brightness of a pixel forming the virtual round continuous pixel number Ncp is darker than the brightness of the central unit pixel C, the extracted corner feature is classified as a bright corner feature, and conversely, If the brightness of the pixel forming the continuous pixel number Ncp is brighter than the brightness of the center unit pixel C, the extracted corner feature can be classified as a dark corner feature at step S311.

After step S311 is executed, the controller 300 executes step S313 (S313a, S313b). The controller 300 determines whether or not the orientation of the extracted corner feature is set and determines whether or not the orientation feature corner feature and the non- Direction) corner feature. That is, when it is determined that directionality exists in the corner feature extracted in step S239 and that the orientation is set, the control unit 300 classifies the extracted corner feature into the orientation corner feature, and if the extracted corner feature does not have directionality The controller 300 classifies the corner feature as a non-edge corner feature.

Through this process, the corresponding corner features can be classified into four types of corner features, namely, the name-bearing corner feature, the name-non-defective corner feature, the cancer-orientation corner feature, and the cancer- , S316, S317, S318). Figs. 14 and 15 show a schematic diagram of the classification criterion type according to the preset corner feature classification criterion such as the contrast and the azimuth. The orientation of the center unit pixel is oriented with respect to the center unit pixel in the virtual round, and when the consecutive pixels having a brightness difference but a predetermined number or more of the consecutive pixels form an arc, It is classified as cancer. According to such a corner feature classification, FIG. 16 (a) shows a bright corner feature having a bearing in the case of the corner feature in the image information, that is, a bright / dark corner feature and a dark corner feature having a bearing, (B) shows an example of a dark corner feature without orientation, i.e., a cancer / non-corner corner feature.

After the feature extraction and classification, in particular, the corner feature extraction and classification step, is performed, the controller 300 performs the matching process using the corner features extracted in the image information and the existing features, And the like. Although the present embodiment has been described with reference to the process executed in the position recognition process of the mobile robot apparatus as an example only, the present invention is not limited thereto. That is, various applications, implementations, and modifications can be made within the scope of taking the structure of extracting, classifying, and matching the corner features. For example, it is possible to track the movement through identification of objects other than the location recognition, A variety of applications are possible.

The feature extracting and feature classification process of the present invention, particularly, the process of extracting the corner feature extracted in the image obtained through the extraction and classification process of the corner feature and matching the extracted feature to the feature stored in the storage unit 400 And precise matching can be achieved quickly by excluding an unnecessary matching calculation process. Further, by performing the alignment process using the orientation of the corner feature, it is possible to execute a matching process that is robust against the rotation state. In addition, the corner feature extraction and classification according to the present invention is performed in a preprocessing manner before the matching process, and can be applied to matching methods using corner features such as FAST and AGAST in the related art. At the same time, The orientation information of the corner feature can be used to calculate the orientation and the position of the mobile robot apparatus and the tracking object in the field of the mobile robot apparatus and the object tracking apparatus using the image processing, It can also be used as information.

The embodiments are illustrative of the present invention, and the present invention is not limited thereto, and various modifications are possible within a range including extraction and classification of corner features.

10 ... position recognition mobile robot apparatus 100 ... image inputting section
200 ... Encoder detection unit 300 ... Control unit
400 ... storage unit 500 ... operation unit
600:

Claims (6)

A feature extracting step of extracting and confirming a feature including a corner feature from the image information, a feature extracting step of extracting a feature including the corner feature from the image information, A feature classification step of classifying the feature points according to a corner feature classification criterion; and a feature matching step of matching using corner features classified in the feature classification step,
Wherein the feature extracting step comprises: a virtual round setting step of setting a virtual round of a predetermined radius with respect to a unit pixel in the image information; a virtual round setting step of forming the virtual round around any pixel in the image information and the arbitrary pixel And a corner feature checking step of comparing the arbitrary pixel with the pixel in the virtual round to check whether or not the corner feature is present,
The corner feature checking step comprises:
A center unit pixel setting step of setting any pixel in the image information as a center unit pixel;
A pixel brightness checking step of arranging the virtual round on the outer periphery of the center unit pixel and comparing the brightness of the center unit pixel and the brightness of the virtual round pixel disposed around the center unit pixel;
Wherein when the brightness difference between the center unit pixel and the virtual round pixel exists in the pixel brightness check step, the brightness of the pixel in the virtual round is set to be greater than the preset brightness difference stored in the storage unit, A consecutive pixel number calculation step of calculating a consecutive virtual round continuous pixel number (Ncp)
A corner feature extraction determining step of comparing the virtual round continuous pixel number (Ncp) with a corner feature determining pixel number (Nc) stored in a storage unit to extract a corner feature;
The virtual round continuous pixel count Ncp for the center unit pixel of the corner feature extracted in the corner feature extraction determination step is compared with the direction verifying pixel count Ndp stored in the storage unit to calculate the number of virtual round continuous pixels And determining a direction of the extracted corner feature based on whether or not the virtual round-shaped pixel has an arcuate shape. delete delete The method according to claim 1,
Wherein the directional confirmation step comprises:
Comparing the directional confirmation pixel number (Ndp) with the number of consecutive virtual round pixels (Ncp) in the virtual round for the center unit pixel of the corner feature extracted in the corner feature extraction determination step;
If it is determined that the number of directional consecutive pixels Ncp is greater than or equal to the number Ndp of directional confirmation pixels and the number of pixels forming the virtual round is equal to or less than the number of pixels forming the virtual round, A directionality determination step of determining,
Wherein when the extracted corner feature is determined to be directional, the center of the pixel forming the virtual round continuous pixel number in the virtual round is set as a directional center pixel, and the directional center pixel and the extraction And forming a center line connecting central unit pixels of the corner feature, and setting the line segment as a diagonal line indicating an azimuth. 5. The method of claim 4,
And aligning the orientation line set in the orientation setting step so as to match the preset reference line. 5. The method of claim 4,
Wherein the feature classification step includes a corner feature classification step of classifying the extracted corner feature according to a predetermined corner feature classification criterion,
Wherein the corner feature classifying step comprises:
A corner contrast determination step of classifying the brightness of the virtual round pixels in which the brightness of the center unit pixel of the extracted corner feature is the number of consecutive virtual round pixels as the bright corner feature and the dark corner feature, ,
Determining whether or not the orientation of the extracted corner feature is set, and classifying it into a bearing feature corner feature and a non-directional corner feature feature.

Images