KR20110004719A - Fast rotation-invariant feature matching method, mobile robot unit based on the ceiling image/feature map with using the mehod, and method for recognizing the self position of the same unit - Google Patents

Abstract

PURPOSE: A fast rotation-invariant feature matching method, a mobile robot based on a ceiling image, and a position recognizing method thereof are provided to reduce operation load by accurately performing calculations. CONSTITUTION: A fast rotation-invariant feature matching method is as follows. Video information inputted through a video input part(100) is processed. A feature is extracted based on current video information. Feature matching is determined in advance based on the extracted feature and the information about a reference feature set in a storing unit. The matching state is determined by extracting a feature image from the feature in a feature matching determining step and splitting the feature image.

Description

Fast ROTATION-INVARIANT FEATURE MATCHING METHOD, MOBILE ROBOT UNIT BASED ON THE CEILING IMAGE / FEATURE MAP WITH USING THE MEHOD, AND METHOD FOR RECOGNIZING THE SELF POSITION OF THE SAME UNIT}

The present invention relates to a feature matching determination method, a robot apparatus, and a control method thereof during image processing. The present invention relates to a robot device and a method for recognizing a location thereof.

The demand for robots is not only for industrial use, but also for home use, and accordingly, research on robots is actively progressing. In particular, unlike conventional position-fixed robots, researches on mobile robots that can be moved are being actively conducted, and the mobile robots can recognize their position accurately using the surrounding environment and marks for reference. Starting from now, researches on techniques for making maps that enable one's own location are underway.

In particular, information having a specific shape, such as a vertex or a straight line, among information obtained from the surrounding environment is named as a feature, and these maps created using the feature map are called feature maps. Since the feature is used as a marker for recognizing the robot's position in the environment, it is important to create an accurate feature map such as what features to use and how to extract them, and whether the extracted features have robust characteristics according to changes in the surrounding environment. This is an important factor.

Such a location recognition of the robot compares the features in the image taken with the pre-stored feature information, and if they match, takes a method of recognizing the current location of the robot. Matching of these features is accomplished in a variety of ways. For example, SSD (sum of squared difference) is one of the typical template matching techniques, and the exact matching result can be obtained through pixel-by-pixel comparison, but the accuracy of matching is drastically reduced when a change in the rotation position occurs in the image. In addition, the pixel-to-pixel comparison has a disadvantage that the computational load increases sharply when there are many comparison images.

In addition, there are feature matching methods such as speeded up robust features (SURF) and scale-invariant feature transform (SIFT), which achieve accurate feature matching but have a sharp increase in computational load.

The present invention solves the problems as described above, high speed rotation invariant feature matching method that can achieve a fast and accurate calculation, significantly reducing the computation load, and ceiling image-based mobile robot position recognition method and ceiling image-based method having the same An object is to provide a mobile robot.

According to an aspect of the present invention, there is provided an image input processing step of processing current image information input through an image input unit, a feature extraction step of extracting a feature based on the current image information, And a feature matching determination step of determining whether to match using the reference feature information preset and stored in the storage unit, wherein in the feature matching determination step, extracting a feature image based on the feature and dividing and partitioning the feature image Provides a method for fast rotating invariant feature matching.

In the fast rotation invariant feature matching method, the feature matching determination step may include: extracting a current feature image of the feature based on the feature from the current image information; A current feature image partitioning step of forming a current partition feature image by dividing the feature at a predetermined partition angle around the feature; a partition feature variable calculating step of calculating a partition feature amount for the current partition feature image; The method may further include a reordering step of comparing and rearranging the current partition feature image, and a partition matching determination step of determining whether to match based on the rearranged current partition feature image and the reference feature information.

In the fast rotation constant feature matching method, in the current feature image segmentation step, the preset partition angle may be 22.5 degrees to 45 degrees.

In the fast rotation invariant feature matching method, the partition feature amount calculating step includes: an image digitizing step of calculating a pixel feature amount by digitizing a pixel in the current partition feature image, and a pixel for a pixel in the current partition feature image A partition feature variable calculation step of calculating and processing the feature variable to calculate the partition feature amount for each current partition feature image may be provided.

In the high speed rotation invariant feature matching method,

In the image digitizing step, the current partition feature image is digitized and digitized, and the partition feature amount may be an average value of pixel feature amounts for each current partition feature image.

In the fast rotation invariant feature matching method, the reordering step includes: a partition descriptor assigning step of comparing a partition feature with respect to the current partition feature image and assigning a partition descriptor according to a pre-arrangement criterion; A partition feature image rearrangement step of rearranging the current partition feature image may be provided as a reference.

In the fast rotation constant feature matching method, the pre-arrangement criterion is set based on the maximum partition feature amount among the partition feature amounts, and the partition descriptor in the counterclockwise direction based on the current partition feature image having the maximum partition feature amount. May be given.

In the fast rotating constant feature matching method, the reference feature information includes a reference partition feature amount and a reference matching decision amount for each feature, and the partition matching determination step includes: a partition of the rearranged current partition feature image; And a matching judgment amount calculating step of calculating a matching judgment amount by comparing a feature amount and the reference partition feature amount, and a matching judgment determining step of comparing the matching judgment amount and the reference matching judgment amount and determining whether to match. have.

In the fast rotation invariant feature matching method, the matching determination amount may be an average value of an absolute value of the difference between the partition feature of the rearranged current partition feature image and the reference partition feature.

According to another aspect of the present invention, an image input processing step of processing current image information input through an image input unit of a ceiling image-based mobile robot device, a feature extraction step of extracting a feature based on the current image information, and extraction A feature matching determination step of determining whether to match using the stored feature and reference feature information preset and stored in the storage unit, and the position of the ceiling image-based mobile robot apparatus based on the matching determined in the feature matching determination step. A ceiling image-based mobile robot device location recognition method may include a location recognition step of recognizing, and in the feature matching determination step, extracting and partitioning a feature image based on the feature and determining whether to match.

In the ceiling image-based mobile robot device position recognition method, the feature matching determination step includes: extracting a current feature image of the feature based on the feature from the current image information; A current feature image partitioning step of forming a current partition feature image by dividing an image by a preset partition angle around the feature, a partition feature variable calculating step of calculating a partition feature amount for the current partition feature image, and the partition And realigning the current partition feature image by comparing feature amounts, and determining whether to match based on the rearranged current partition feature image and the reference feature information.

According to yet another aspect of the present invention, a ceiling image-based mobile robot device that moves in position by a driver, comprising: an image input unit for acquiring current image information of a ceiling, and preset reference information for the ceiling image; A storage unit and a control unit extracting a feature from the current image information of the ceiling and comparing the feature and the reference feature information to determine whether to match, and extracting a feature image based on the feature and partitioning the partition to determine matching. A ceiling image-based mobile robot device may be provided.

The fast rotation invariant feature matching method, the ceiling image-based mobile robot apparatus having the same, and the position recognition method thereof according to the present invention having the configuration as described above have the following effects.

First, the fast rotation invariant feature matching method according to the present invention, the ceiling image-based mobile robot device using the same, and the position recognition method thereof, when the current image information obtained through the image input unit 100 has only a rotation change without a change in size Accurate and rapid feature matching can be made.

Secondly, the fast rotation invariant feature matching method according to the present invention, the ceiling image-based mobile robot device using the same, and the location recognition method thereof can be used in an embedded system according to the reduction of the computational load. You can also

Third, the fast rotation invariant feature matching method according to the present invention, the ceiling image-based mobile robot device using the same, and the position recognition method thereof can enable more accurate position recognition of the ceiling image-based mobile robot according to a feature matching decision made quickly. You may.

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.

Hereinafter, a fast rotation invariant feature matching method, a ceiling image-based mobile robot device, and a position recognition method thereof will be described with reference to the accompanying drawings. Here, the fast rotation invariant feature matching method of the present invention can be used in various image processing fields. In the present embodiment, the present invention will be described based on the case used for the position recognition method of the ceiling image-based mobile robot apparatus 10. As an example for describing the high speed rotation invariant feature matching method of the present invention, the high speed rotation invariant feature matching method of the present invention is not limited thereto.

FIG. 1 is a schematic block diagram of a ceiling image-based mobile robot apparatus 10 according to an embodiment of the present invention, and FIG. 2 illustrates a fast rotation invariant feature matching method according to an embodiment of the present invention, and A flowchart showing a schematic control process for a ceiling image-based mobile robot device location recognition method is shown. According to the high-speed rotation invariant feature matching method and the ceiling image-based mobile robot device location recognition method, a ceiling as another example of the present invention is shown. An image based mobile robot apparatus 10 is provided.

First, the ceiling image-based mobile robot device 10 according to an embodiment of the present invention includes an image input unit 100, a storage unit 400, and a control unit 300 to perform a predetermined control process, and to encode an encoder. Including the sensing unit 200, the calculating unit 500, and the driving unit 600, autonomous driving may be achieved by recognizing a predetermined moving position through a fast rotational invariant feature matching method and a ceiling image based mobile robot device position recognition method based thereon. Can be.

The controller 300 may be in electrical communication with other components to receive an input signal and apply a control signal to each component. The image input unit 100 may be configured in various ways. In the present embodiment, the image input unit 100 may be configured as a monocular camera. The image input unit 100 is implemented as a monocular camera and acquires a ceiling image. The image input unit 100 may not easily recognize an accurate distance from an object due to the configuration of the monocular camera. The calculation unit 500 performs a predetermined calculation process according to the control signal of the control unit 300, and the storage unit 400 pre-stores various preset values and stores necessary image information according to the control signal of the control unit 300. Store location information. The driver 600 is implemented by an electric motor or the like and is driven according to a control signal of the controller 300 to provide a driving force for moving the ceiling image-based mobile robot device 10 to a desired position. The encoder detecting unit 200 detects a rotation speed and a 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 in the rotation speed of each driving wheel. It may be through.

2 is a flowchart illustrating a schematic control process of the ceiling image-based mobile robot apparatus 10 according to an embodiment of the present invention. The fast rotation invariant feature matching method of the present invention is included in the bar image-based ceiling image. Based on the ceiling image-based mobile robot device position recognition method of the mobile robot device 10 will be described. Here, the fast rotation invariant matching method includes an image input processing step (S10), a feature extraction step (S20), and a feature matching step (S30), and the ceiling image-based mobile robot device position recognition method is located in the fast rotation invariant matching method. The case where the recognition step S40 is further added.

First, current image information input through the image input unit 100 is processed in the image input processing step (S10). That is, the controller 300 receives the ceiling image (or image information) from the image input unit 100, processes the image information, and takes current image information. Before the image input processing step S10 is executed, the ceiling image-based mobile robot apparatus 10 is provided. In the flowchart of FIG. 2, notation is omitted. 3 shows an example of a more specific flow of the image input processing step (S10), wherein the controller 300 applies a control signal to the image input unit 100 so that the image input unit 100 implemented as a monocular camera is presently present. The image information on the ceiling (ceiling) at the position of the ceiling image-based mobile robot device 10 is acquired and input to the controller 300 (S11). The controller 300 determines whether the input image or the image information is distorted (S12), and if image distortion exists in the image or the image information, the controller 300 transmits the image distortion to the operation unit 500 to correct the image. Perform the step (S13). 4 and 5 illustrate schematic examples of an image before and after image correction. The image input unit 100 implemented as a monocular camera extracts a feature from a ceiling image and continuously tracks the image or image. Distortion occurs in the image information acquired to secure a wide field of view so that information can be obtained. The calculation unit 500 corrects the distortion by processing the image or the image information input from the image input unit 100 according to the control signal of the control unit 400. When 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).

After the image input processing step S10 is completed, a feature extraction step S20 is performed to extract a feature included in the current image information based on the current image information obtained in step S10. In the feature extraction step S20, the controller 300 extracts a feature existing in the current image information based on the image in which image distortion and the like are corrected. Features present in the current image information include corner features, outlines, straight features, lighting features, and the like. 6 shows an example of a feature extraction step according to an embodiment of the present invention. First, a corner feature is extracted from image information (S21), an outline is extracted from image information (S22), and the obtained contours are obtained. The straight line feature is extracted from (S23). After the linear feature is extracted, the present embodiment includes a step S24 of extracting the lighting feature after the door feature extraction, and the feature extraction can be variously extracted according to the design specification.

In step S20, corner features, outline extraction, and straight line extraction are performed from the current image information. The corner features are formed as an angled portion or a portion where two lines cross each other in the current image information. After contour extraction is performed from the current video information, various straight lines are obtained. Among these straight lines, a straight line having a predetermined length or more is set as a straight line feature as a feature. While the straight line feature may be confused with adjacent straight lines, the corner feature is clearly distinguishable from the surrounding part, and the corner feature can be immediately used as a marker for position recognition. For example, the extraction of corner features may be performed by using corner extraction methods such as Harris corner detection or Features from Accelerated Segment Test (FAST). Since the lighting extraction step is the same as a general image information processing process, a detailed description thereof will be omitted.

In the embodiment of the present invention, the feature extraction step (S20) is a corner feature extraction step (S21), contour extraction step (S22), linear feature extraction step (S23) and illumination feature extraction step (S24) for extracting various features This is just an example for explaining the process of utilizing various features in the position recognition method of the ceiling image-based mobile robot device, the fast rotation constant feature matching method and the ceiling image-based mobile robot device of the present invention And the location recognition method thereof are not limited thereto. That is, in the case of the fast rotation invariant feature matching method of the present invention, various modifications are possible depending on the design specification, such as a configuration of extracting only a corner feature that is a feature of a point unit in the feature extraction step.

Thereafter, the controller 300 executes the feature matching step S30 of determining whether to match using the features extracted in the step S20 and the reference feature information preset and stored in the storage 400. As shown in FIG. 7A, the feature matching determination step (S30) according to the present embodiment includes a current feature image extraction step (S31), a current feature image partition step (S33), a partition feature variable calculation step (S35), and a rearrangement step. (S37) and the partition matching determination step (S39).

The current feature image extraction step S31 extracts a current feature image for a feature based on the feature extracted from the acquired current image information. The current feature image, which is a peripheral image centered on a feature of a predetermined size, is extracted based on the feature extracted at step S20, for example, a corner feature.

The fast rotation invariant matching method of the present invention is applied when there is no change in the size of the image. That is, even though the ceiling image-based mobile robot device moves, the distance between the ceiling and the ceiling image-based mobile robot device 10 is constant, so that an image input through the image input unit 100 of the ceiling image-based mobile robot device 10 is based on the ceiling image. Even when the mobile robot apparatus 10 moves, the size of the image does not change but only a rotation change with respect to the image is performed. FIG. 8 illustrates a comparison of current image information input through the image input unit 100 of the ceiling image-based mobile robot device 10 and current image features included therein. (A), (b) and (c) of FIG. 8 show image information acquired when the ceiling image-based mobile robot apparatus 10 moves through the image input unit of the ceiling image-based mobile robot apparatus 10, and (d) , (e), (f) represent current feature images acquired from the current image information of each of Figs. 8A, 8B, and 8C. (D), (e), and (f) of FIG. 8 are acquired based on corner features among current image information. When a plurality of features exist in the same current image information, a plurality of current feature images may be acquired. It is obvious. (A), (b), and (c) of FIG. 8 are images obtained when the ceiling image-based mobile robot apparatus 10 rotates by approximately 90 degrees clockwise, respectively. FIGS. ), (f) represents the center of the corner feature in the current feature image obtained from the features shown in FIGS. 8 (a), (b) and (c). e) and (f) are obtained in the size of 21 pixels x 21 pixels around these center points. As such, since the distance between the ceiling image-based mobile robot device 10 and the ceiling is almost constant, it can be seen that the image acquired by the ceiling image-based mobile robot device 10 has only a rotation change, which is an image of the present invention. Even if the information forms a high-speed rotation state, it is possible to accurately and quickly determine whether the feature is matched, and moreover, to quickly recognize the current position of the ceiling image-based mobile robot device.

As described above, in step S31 the current feature image (as shown in Fig. 8 (Fig. 8) is centered on a feature such as a corner feature obtained from the current image information as shown in Figs. 8A, 8B, and 8C. d), (e), (f)) are extracted. In the present embodiment, the size of the current feature image is set to an image size of 21 pixels × 21 pixels. However, this is an example for explaining the present invention, and the present invention is not limited thereto. In order to derive a more accurate feature matching result, a high resolution image input is performed. Various modifications are possible, such as taking the configuration of extracting the current feature image around the feature or adjusting the size of the current feature image. However, if the size of the image is large, the accuracy of matching determination can be improved.However, as the size increases, the computational load increases, so that the computational load may be excessive. Since is lowered, it is preferable to take an image of the feature surroundings at an appropriate size.

If the current feature image is extracted based on the feature extracted from the current image information in step S31, the current feature image segmentation step of forming a current partition feature image by dividing the current feature image by a preset partition angle around the feature. (S33) is executed. When the center point of the feature is indicated by reference numeral O in the current feature image, the current feature image having a size of 21 × 21 pixels around the center point O is divided into equal intervals by a preset partition angle α. FIG. 9 is a schematic state diagram for such a current feature image segmentation step S33. The current feature image of 21 × 21 pixels in FIG. 9 is a preset segment angle (aa to hh around the center point O). By dividing into α), a plurality of partition feature images can be obtained. The plurality of partition feature images are divided at equal intervals based on the center point O of the current feature image, and each partition feature image is allocated the same number or substantially the same number of pixels. In the present exemplary embodiment, the preset partition angle α is set to 22.5 degrees so that one current feature image includes 16 partition feature images in total. Here, the preset partition angle α may have various values according to the design specification, but the preset partition angle α has a value ranging from 22.5 degrees to 45 degrees so that approximately 8 pieces of the current feature image are present. It is preferable to configure to have from 16 to 16 partition feature images. For example, if the preset parcel angle α is smaller than 22.5 degrees, a large number of parcel feature images can be obtained for one current feature image, further improving accuracy in the following matching process. Although the number of pixels for the image included in the partition feature image partitioned by the preset partition angle α is also small, the robustness to noise is weakened, and the preset partition angle α is too large, that is, If the preset partition angle α has a value larger than 45 degrees, the number of pixels in the partitioned feature image is increased, so the robustness against noise is increased but the accuracy is lowered. Therefore, the preset partition angle α is an appropriate range of values. It is desirable to be designed to have.

In Fig. 9, each partition feature image partitioned by lines aa to hh with respect to the current feature image is denoted by numerals "0" to "15", and the numerals composed of these numerals are descriptors and are arbitrarily herein. Listed. However, the descriptors for the partitioned plurality of partition feature images are sequentially arranged in the order of the numbers. In this embodiment, the descriptors are arranged in the counterclockwise direction.

After partitioning from the current feature image partitioning step S33 to a plurality of partition feature images for the current feature image, the controller 300 performs a partition feature variable calculating step S35, and in the partition feature variable calculating step S35. Each partition feature amount is calculated from a plurality of partitioned current partition feature images. Here, the partition feature amount represents the characteristics of each of the partitioned plurality of current partition feature images, which are used as a reference for rearranging the plurality of current partition feature images in the reordering step described below. In one embodiment according to the present invention, the partition feature amount is an average value of grayscales for the current partition feature image. However, this is an example for explaining the present invention, and the partition feature can be applied to various criteria according to the design specification. Various current processing features such as average, standard deviation, and variance may be used in the range for smoothing, and each current partition feature video may be rearranged in order to rearrange a plurality of current partition feature images such as other RGB modes besides grayscale mode. There are a variety of choices within the scope of establishing the criteria for quantifying.

The partition feature variable calculating step S35 according to an embodiment of the present invention includes an image digitizing step S351 and a partition feature variable calculating step S353. In the image digitizing step S351, the pixel in the current partition feature image is selected. The numerical feature is calculated, the pixel feature amount is calculated, and in the partition feature variable calculation step S353, the pixel feature amount for the pixel in the current partition feature image is calculated and processed to calculate the partition feature amount for each current partition feature image. Here, the pixel feature amount represents a numerical value for a plurality of pixels included in the current partition feature image, and the partition feature amount represents a pixel feature amount for a plurality of pixels included in each current partition feature image. It represents the value derived by calculating. As illustrated in FIG. 10, a switch to the gray scale mode for the current feature image or the plurality of current partition feature images is performed in the image digitization step S351. The image conversion to the gray scale mode is performed to quantify a range of 0 to 255 values for each pixel, and calculate the pixel feature amounts for the plurality of pixels included in the plurality of current partition feature images.

Then, in the partition feature variable calculating step S353, partition feature quantities for the plurality of current partition feature images are calculated. The partition feature amount is obtained from the pixel feature amounts for the pixels included in the individual current partition feature images. In this embodiment, the partition feature amounts use the average value of the pixel feature amounts for each pixel. That is, the partition feature amount for each current partition feature image is calculated by summing the pixel feature quantities calculated for the pixels included in each current partition feature image by the number of pixels in the current partition feature image, and dividing them by the number of pixels in the current partition feature image. can do. Such partition feature amounts are computed separately for each current partition feature image to which the start position is arbitrarily determined and given a descriptor that increases in the counterclockwise direction.

In this way, after the partition feature variable calculating step S35 is completed, the control unit 300 executes the reordering step S37. In the rearranging operation S37, the controller 300 rearranges the plurality of current partition feature images by comparing the calculated partition feature amounts with respect to the plurality of current partition feature images. A reordering step S37 according to an embodiment of the present invention includes a partition descriptor providing step S371 and a partition feature image rearranging step S373, and the partition descriptor providing step S371 includes a plurality of current partition feature images. Comparing the partition feature quantities and assigning partition descriptors according to a pre-arrangement criterion, the partition feature image reordering step S373 rearranges the plurality of current partition feature images based on the assigned partition descriptors. In the partition descriptor assignment step S371, the controller 300 compares the partition feature amounts for each current partition feature image. As shown in FIG. 11, each partition feature amount for a plurality of current partition feature images (see FIG. 11A) included in the current feature image, that is, pixels in gray scale mode for each current partition feature image. A diagram showing a segment feature amount, which is an average value for each current segment feature image, of the pixel feature amount, which is a star value, for each current segment feature image, is shown. As shown in FIG. 11 (b), a diagram in which segment feature amounts for each current segment feature image represented by each descriptor are listed is shown, and the controller 300 may display each current segment that can be represented by such a diagram. Compare feature images. FIG. 7B shows a more detailed flowchart of the reordering step S37. The reordering step S37 includes a partition descriptor assigning step S371 and a partition feature image rearranging step S373 as described above, and the partition descriptor is provided. Step S371 includes a current partition feature video partition feature variable comparison step S3711, a first partition descriptor assignment step S3713, and a sequential partition descriptor assignment step S3715. Comparing the current partition feature image in the partition feature amount comparison step (S3711), the partition feature values calculated for the plurality of current partition feature images are compared to determine which current partition feature image has the maximum feature value of partition feature or which current partition feature image. It is possible to determine whether it has the partition feature amount of this minimum value. Then, according to the comparison result made in the current partition feature video section feature quantity comparison step (S3711), the first partition descriptor assigning step (S3713) of assigning a first partition descriptor to a specific current partition feature image according to a predetermined partition descriptor assignment criterion is performed. Is performed. Sequential partition descriptor assignment step (S3715) of assigning the partition descriptors sequentially in the order adjacent to each other in the counterclockwise direction according to a predetermined criterion, for example, after the highest priority partition descriptor is given to the specific current partition feature image. Run After the partition descriptor assignment step S371 is completed, the partition feature image rearranging step S373 is performed to rearrange the plurality of current partition feature images based on the newly assigned descriptor to the partition descriptors sequentially assigned from the highest partition descriptor. Accordingly, values for each current partition feature image may be rearranged.

Thereafter, the controller 300 executes the partition feature image rearrangement step S373. Here, the controller 300 rearranges the current partition feature images compared through the partition feature amounts according to a pre-arrangement criterion. As shown in FIG. 11B, a total of 16 images of the current feature images in the present embodiment are shown. It is confirmed that the partition feature amount of the current partition feature image indicated by the descriptor 11 among the current partition feature images has the largest value. The highest partition value of the new partition descriptor is assigned to the current partition feature image having the largest partition feature amount. That is, as shown in FIG. 12, "0", which is the highest value of the new partition descriptor, is assigned to the current partition feature image indicated by 11 of the previous descriptor, and sequentially in the counterclockwise direction based on the new partition descriptor "0". Gives an incremental partition descriptor for each current partition feature image. Here, the incremental partition descriptor is given in a counterclockwise direction, which is intended to match the descriptor assignment order in the current feature image segmentation step, and the descriptor or partition descriptor is assigned in clockwise order in a range that maintains the same assignment order. It can be modified according to design specifications, such as can be achieved. FIG. 13 illustrates rearrangement of the partition feature amounts of the current partition feature image rearranged in the order of partition descriptors based on a newly assigned partition descriptor according to an embodiment of the present invention. Therefore, the current partition feature image having the highest partition feature amount is arranged at the head according to the pre-arrangement criterion.

 The pre-arrangement criterion according to the present embodiment is to assign the highest priority partition descriptor based on the largest partition feature value, that is, the maximum partition feature of each partition feature of the plurality of current partition feature images. For example, a pre-arrangement criterion that represents a parcel descriptor granting criterion in the design specification is consistent with the design specification, where a consistent application criterion exists, such as assigning a new highest parcel descriptor to the current parcel feature image with the minimum parcel feature values. Accordingly, various modifications may be made.

14 and 15 show the respective partitions for each current feature image A and B made through the partition feature calculation (S35) and the reordering step (S37) of the feature matching step (S30) according to an embodiment of the present invention. Each diagram listing the feature quantities in order of their respective compartment descriptors is shown. Here, the current feature image B of FIG. 15 is a rotated image 90 degrees counterclockwise with respect to the current feature image A of FIG. 14. The segment feature quantity-compartment descriptor diagram for the current feature image A is indicated by a solid line and the segment feature quantity-compartment descriptor diagram for the current feature image B is indicated by a dotted line. In FIG. 16, the partition feature variables-compartment descriptor diagrams of the current feature image A and the current feature image B of FIGS. 14 and 15 are superimposed, and the partition feature amounts and the dotted lines of the current feature image A indicated by a solid line are displayed. It is possible to confirm that the error of the partition feature variable is small in the current partition feature image indicated by each partition descriptor of the partition feature variable for the current feature image B. In this case, the error of the partition feature amount generated when the current feature image is rotated is because the region partitioning each current feature image is slightly different. The matching is determined by comparing the matching judgment amount and the reference matching decision amount, which are the same to almost the same current partition feature image, and are assigned to each rearranged current partition feature image. It can be seen that the partition descriptors take the same arrangement for the same to nearly identical rearranged current partition features.

The partition matching determination step S39 is performed after the rearranging step S37 as described above. The partition matching determination step S39 is performed based on the current partition feature image rearranged in the partition matching step S39 and reference feature information preset and stored in the storage unit 400. It is determined whether a feature extracted from the current image information and represented by the current feature image including the surrounding environment matches a previously stored reference feature. Here, the reference feature information is a value preset and stored in the storage unit 400, and the reference partition feature amount Sref and the reference matching determination amount Dref for each feature in the feature map pre-stored in the storage unit 400. The reference partition feature amount means a partition feature amount obtained for a feature previously stored in the feature map, and the reference matching determination amount determines whether a feature obtained from current image information matches the reference feature previously stored in the feature map. Means a preset reference value.

As shown in FIG. 17, in this embodiment, the partition matching determination step S39 includes a matching determination amount calculation step S391 and a matching determination determination step S393, and the matching determination amount calculation step S391 is rearranged. The matching determination amount is calculated by comparing the partition feature amount of the current partition feature image with the reference partition feature amount, and in the matching determination determination step (S393), the calculated matching determination amount and the reference matching determination amount are compared to determine whether to match. . The matching judgment amount is used as a criterion for determining whether or not the feature obtained from the current image information is matched with the reference feature. In this embodiment, the matching judgment amount D is equal to the partition feature amount S of the rearranged current partition feature image. The average value of the absolute values of the differences of the reference partition feature quantities Sref is used. That is, the match determination amount D according to an embodiment of the present invention may be expressed as follows.

Where N denotes the number of current partition feature images in which the current feature image is partitioned by a preset partition angle, and Δvi is the partition feature amount S for the current partition feature image indicated by each partition descriptor, and The difference value of the reference partition feature amount Sref with respect to the reference feature corresponding to the current feature image is shown. The predetermined value is determined by adding the absolute values of the difference values of the partitioned feature amounts S of the rearranged current partition feature images and the corresponding reference partition feature quantities Sref, and dividing them by the number of partition feature images as the matching judgment amount. While guaranteeing the accuracy of, can significantly reduce the computational load. However, this is only one example of a matching judgment amount used to determine whether a feature obtained from current image information matches a previously stored reference feature, and the matching judgment amount of the present invention is a simple difference value, an absolute value of the difference value, and a difference value. There are many choices depending on the design specification, such as the square of, and the RMS of the difference.

When the matching determination amount D is calculated in the matching determination amount calculating step S391, the controller 300 compares the matching determination amount D and the reference matching determination amount Dref to determine whether the feature is matched. The matching determination decision step S393 is executed. In this embodiment, the reference matching judgment amount Dref is set to a value of 30 to 50, which can be modified according to the design specification. Therefore, when the controller 300 determines that the matching determination amount D is less than or equal to the reference matching determination amount Dref in step S393, the control unit 300 transmits the control flow to step S395 to determine that both are matched. However, since the matching determination amount obtained from the error between the feature obtained from the current image information and the feature previously stored in the feature map is smaller than the reference matching determination amount as the reference value, both may be determined to be matched. On the other hand, when the controller 300 determines that the matching determination amount D is greater than the reference matching determination amount Dref in step S393, the control unit 300 transmits the control flow to step S397 to determine that both are not matched. . In this case, the controller 300 transmits the control flow to step S397 to determine that both are inconsistent.

18 and 19 illustrate an embodiment of the partition matching determination step S39 according to an embodiment of the present invention. First, FIG. 18A illustrates a legend for displaying a partition feature variable-compartment descriptor diagram for a current feature image for a feature obtained from current image information and a reference feature image for a pre-stored reference feature, where the current The segment feature quantity-compartment descriptor plot for the feature image is indicated by a solid line, and the segment feature quantity-compartment descriptor diagram for the reference feature image is indicated by a dashed line. FIG. 18B shows a partition feature-partition descriptor diagram for the current feature image and the reference feature image obtained through the partition, the partition feature amount calculation and the reordering step for the feature image as described above. Here, it is understood that the difference (Δvi = S-Sref) between the partition feature amount of the current partition feature image represented by each partition descriptor image and the reference partition feature amount for the reference partition feature image has a significant value for each partition descriptor. Can be. In the case of this embodiment, the matching judgment amount D calculated is obtained with a value of 105.9, which is considerably larger than the reference matching judgment amount Dref in the range of about 30 to 50 which is preset and stored, about 2 to 3 times. The controller 300 determines that the feature obtained from the current image information does not match the reference feature, and terminates step S39.

In FIG. 19 (a), a display legend of the image of the current feature and the reference feature and each segment feature amount is shown in FIG. 19 (a), and FIG. A segment feature quantity-compartment descriptor diagram is shown for the current feature image and the reference feature image obtained through the segment feature quantity calculation and rearrangement steps. Here, the difference (Δvi = S-Sref) between the partition feature amount of the current partition feature image represented by each partition descriptor image and the reference partition feature amount for the reference partition feature image has almost a small error for each partition descriptor. It can be seen that the plot of the partition feature quantities closely matches the reference plot feature plot. In the case of this embodiment, the matching judgment amount D calculated is obtained with a value of 18.6, which is considerably smaller than the reference matching judgment amount Dref in the range of about 30 to 50 which is preset and stored, about 0.35 to 0.6 times. The controller 300 determines that the feature obtained from the current image information matches the reference feature, and terminates step S39.

FIG. 20 is a schematic diagram showing whether matching is performed based on the fast rotational invariant matching method as described above, each reference feature image indicated by L1, L2, and L3 and the fast rotational invariant matching method as described above. The current feature image including the feature determined to be matched by A is indicated by a, b, c, d, and e. Here, the current feature image including each feature may be identified as an image in which rotation deformation occurs when viewed in comparison with each reference feature.

Through the simplified structure determination method as described above, robustness and fastness of feature matching can be achieved even with a large number of features.

In addition, as described above, the matching result obtained in step S30 of the fast rotational invariant matching method as described above is used as determination data for position recognition of the ceiling image-based mobile robot device. That is, the ceiling image-based mobile robot device position recognition method executes step S40 after step S30 of the fast rotational invariant matching method. Step S40 is a position recognition step, and the control unit 300 matches the feature in the current image information with the reference feature in the matching determination decision step S393 of the partition matching determination step S39 of step S30 of the fast rotational invariant matching method. If it is determined (step S395), the control unit 300 to grasp the position information of the current ceiling image-based mobile robot device 10 by using the position information on the reference feature that is preset and stored in the storage unit 400 Can be updated. On the other hand, the control unit 300 determines that the characteristics in the current image information and the reference feature are not matched in the matching determination determination step (S393) of the section matching determination step (S39) of the step S30 of the high speed rotation invariant matching method to determine that they are not matched. In the case (step S397), the control unit 300 grasps the position information of the current ceiling image-based mobile robot device 10 based on the position information obtained during the previous position recognition process and the movement amount calculated by the encoder detection unit 200. To update.

The above embodiments are examples for describing the present invention, but the present invention is not limited thereto. When the image obtained through the image input unit has only a rotation change without a change in size, a fast rotation invariant feature matching method and a ceiling image mobile robot including and utilizing the fast rotation invariant feature matching method to determine whether to match the feature or to perform position recognition based on the feature. Various modifications are possible in the range of providing a device location recognition method and a ceiling image-based mobile robot device.

1 is a schematic block diagram of a ceiling image-based mobile robot device 10 according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating a schematic control process for a fast rotation invariant feature matching method and a ceiling image based mobile robot device position recognition method according to an embodiment of the present invention.

3 is a flowchart illustrating an image input processing step (S10) of a fast rotation invariant feature matching method and a ceiling image-based mobile robot device position recognition method including the same according to an embodiment of the present invention.

4 and 5 are schematic diagrams of an image obtained by the image input unit of the ceiling image-based mobile robot device according to an embodiment of the present invention and an image of which distortion is corrected.

6 is a flowchart illustrating a feature extraction step of a fast rotational invariant feature matching method and a position recognition method of a ceiling image-based mobile robot apparatus having the same according to an embodiment of the present invention.

7A is a schematic flowchart of a feature matching determination step S30 according to the present embodiment.

7B is a more detailed flowchart of the reordering step S37 according to the present embodiment.

FIG. 8 is a schematic diagram illustrating current image information and current feature images obtained by a fast rotation invariant feature matching method and a ceiling image-based mobile robot device position recognition method including the same according to an embodiment of the present invention.

9 is a schematic state diagram of the feature image segmentation step S33 of the present invention.

FIG. 10 is a schematic diagram illustrating an image converted to a gray scale mode for a current feature image or a plurality of current partition feature images in an image digitization step S351 of the present invention.

11 is a diagram of a plurality of current partition feature images and a partition feature variable-compartment descriptor diagram containing a current feature image.

12 shows a current feature image to which a new partition descriptor is assigned.

13 shows the partition feature-block descriptor diagrams rearranged in order of new partition descriptors.

14 and 15 illustrate, in order of each partition descriptor, the respective partition feature values for each current feature image A and B, which are generated through the partition feature variable calculation and reordering step of the feature matching step, according to an embodiment of the present invention. Each line listed is.

FIG. 16 is a partition feature variable-compartment descriptor diagram displayed superimposed on the current feature image A and the current feature image B of FIGS. 14 and 15.

17 is a more detailed diagram of the partition matching determination step.

18 and 19 are diagrams showing an embodiment of the partition matching determination step S39 according to an embodiment of the present invention.

20 is a schematic diagram showing whether registration is performed based on the fast rotational invariant matching method of the present invention.

* Description of the symbols for the main parts of the drawings *

10 ... Ceiling-based mobile robot device 100 ... Image input unit

200 encoder encoder 300 controller

400 storage 500 operation

Claims (12)

An image input processing step of processing current image information input through the image input unit; A feature extraction step of extracting a feature based on the current image information; And a feature matching determination step of determining whether to match using the extracted feature and reference feature information preset and stored in the storage unit. And in the feature matching determination step, extracting and partitioning a feature image based on the feature to determine whether to match. The method of claim 1, The feature matching determination step is: Extracting a current feature image of the feature based on the feature from the current image information; A current feature image partitioning step of dividing the current feature image into a preset partition angle around the feature to form a current partition feature image; A partition feature variable calculating step of calculating a partition feature variable for the current partition feature image; Reordering the current partition feature video by comparing the partition feature amounts; And a partition matching determination step of determining whether to match based on the rearranged current partition feature image and the reference feature information. 3. The method of claim 2, And in the present feature image segmentation step, the preset segment angle is between 22.5 degrees and 45 degrees. 3. The method of claim 2, The partition feature quantity calculating step is: An image digitizing step of calculating pixel feature amounts by digitizing pixels in the current partition feature image; And a partition feature variable calculation step of calculating and processing a pixel feature amount for a pixel in the current partition feature image to calculate a partition feature amount for each current partition feature image. The method of claim 4, wherein In the image digitizing step, the current partition feature image is digitized by digitizing, And said partition feature amount is an average value of pixel feature amounts for each current partition feature image. 3. The method of claim 2, The reordering step is: A partition descriptor assigning step of comparing a partition feature with respect to the current partition feature image and granting a partition descriptor according to a pre-arrangement criterion; And a partition feature image rearranging step of rearranging the current partition feature image based on the assigned partition descriptor. The method of claim 6, The pre-arrangement criterion is set based on the maximum partition feature amount among the partition feature amounts, And a partition descriptor in a counterclockwise direction based on the current partition feature image having the maximum partition feature amount. 3. The method of claim 2, The reference feature information includes a reference partition feature amount and a reference matching judgment amount for each feature, The partition matching determination step is: A matching judgment amount calculating step of calculating a matching judgment amount by comparing the partition feature amount of the rearranged current partition feature image with the reference partition feature amount; And a matching determination determination step of comparing the matching determination amount with the reference matching determination amount to determine whether to match. The method of claim 8, And the matching determination amount is an average value of an absolute value of the difference between the partition feature amount of the rearranged current partition feature image and the reference partition feature amount. An image input processing step of processing current image information input through an image input unit of a ceiling image-based mobile robot device; A feature extraction step of extracting a feature based on the current image information; A feature matching determination step of determining whether to match using the extracted feature and reference feature information preset and stored in the storage unit; And a location recognizing step of recognizing the location of the ceiling image-based mobile robot device based on the matching determined in the feature matching determination step. In the feature matching determination step, a ceiling image-based mobile robot device position recognition method for extracting a feature image based on the feature and partitioning to determine a match. The method of claim 10, The feature matching determination step is: Extracting a current feature image of the feature based on the feature from the current image information; A current feature image partitioning step of dividing the current feature image into a preset partition angle around the feature to form a current partition feature image; A partition feature variable calculating step of calculating a partition feature variable for the current partition feature image; Reordering the current partition feature video by comparing the partition feature amounts; And a partition matching determination step of determining whether to match based on the rearranged current partition feature image and the reference feature information. In the ceiling image-based mobile robot device for moving the position by the drive unit, A video input unit for acquiring current image information of the ceiling; A storage unit for preset storing reference characteristic information on the image of the ceiling; And extracting a feature from the current image information of the ceiling and comparing the feature with the reference feature information to determine whether to match. Ceiling image-based mobile robot device characterized by.

Images