KR102283904B1 - Object recognition systen of image based on deep learning and recognition method therefor - Google Patents

Abstract

The object recognition system converts a reflective image separated from an image acquired from a camera into a imitation real image that mimics a real image, and a real image separated from the image acquired from the camera is converted into a imitation reflective image that mimics a reflective image and an image conversion unit.

Description

An object recognition system for images based on deep learning and a method for recognizing the same

The present invention relates to an image object recognition system based on deep learning and a method for recognizing the same.

When acquiring an image inside a space limited by a camera located in the upper center of an internal space of a storage box such as a refrigerator or a vending machine, objects at the edge of the camera lens angle of view are close to the camera lens and are often obscured by a large object.

1 is an exemplary view of a camera image obtained for a narrow interior space.

As can be seen from FIG. 1 , in the case of a product 'B' hidden behind a product marked 'A', there is a difficulty in object recognition during deep learning.

When there are many such obscured objects, when deep learning is performed in a system for recognizing an object that is a product in a corresponding space using an image obtained from a camera, correct recognition of the obscured objects is difficult.

1. China Utility Model Registration Publication 208781326 (2019.04.23) 2. Domestic Patent Publication No. 10-2016-0041965 (2016.04.18) 3. Domestic Registered Patent Publication No. 10-1991424 (2019.06.20)

The present invention is an invention aimed at solving the technical problems as described above, and when the product is stored in an internal space such as a refrigerator or a vending machine, an image of the product covered by a reflection image using a mirror is also secured and the corresponding image is obtained. An object of the present invention is to provide an image object recognition system based on deep learning capable of recognizing a product and a method for recognizing the same.

The object recognition system of the present invention converts a reflected image separated from an image obtained from a camera into a imitation real image that imitates a real image, and a real image separated from the image acquired from the camera is a imitation reflection that imitates a reflected image It includes; an image converter that converts the image into an image. The reflection image before conversion used by the image conversion unit corresponds to at least a part of the total reflection image, and the actual image before conversion used by the image conversion unit corresponds to at least a part of the entire actual image. do.

In addition, the object recognition system of the present invention may include an object region detection unit configured to detect a region in which an object exists from each of the reflected image, the real image, the imitation real image, and the imitation reflected image; a feature point extraction unit for extracting feature points of each of the reflection image, the real image, the imitation real image, and the imitation reflection image; and an object detection unit configured to detect an object present in the image obtained from the camera using information on a region in which the object detected by the object region detection unit exists and the characteristic point extracted by the characteristic point extraction unit. do.

Specifically, the feature point extraction unit is characterized in that the feature points of the reflection image and the feature points of the imitation reflection image are matched with each other, and the feature points of the real image and the feature points of the imitation real image are matched with each other.

In addition, the object detection unit may include: a first matching ratio that is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflection image with respect to the region detected as the region where the object exists by the object region detector; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image. It is preferable to detect the region as an object of the first object group.

In addition, the object detection unit may include: a first matching ratio that is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image with respect to the region detected as the region where the object exists by the object region detector; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using, the first matching ratio or the second matching ratio is less than a preset second ratio. It is characterized in that the region is detected as an object of the second object group.

Preferably, the image converter comprises: a reflection image of N partial regions corresponding to a partial region of the total reflection image; and an actual image of N partial regions corresponding to the reflection images of the N partial regions, respectively, among the total real images of the internal space in which the object is stored; a real image imitation of each of the reflection images of the N partial regions and converts each of the actual images of the N partial regions into a imitation reflection image, wherein N is an integer of 2 or more.

In addition, in the object recognition system of the present invention, the position information of the object of the first object group and the object of the second object group detected for each of the reflection images of the N partial areas and the actual images of the N partial areas respectively By unifying the duplicated detected objects among the objects detected for all of the reflection images of the N partial areas and the actual images of the N partial areas by using the location information, the entire object actually stored in the internal space It is preferable to further include; an object determining unit to be determined.

In addition, the object recognition system of the present invention further comprises; an image acquisition unit for acquiring an image; and a camera for acquiring images of the interior space and the plurality of mirrors. In addition, the front glass of the plurality of mirrors may be in a transparent or opaque state by application of an electrical signal.

In the object recognition method of the present invention, the reflected image separated from the image acquired from the camera is converted into a imitation real image that imitates the real image, and the real image separated from the image acquired from the camera is imitation reflection that imitates the reflection image An image conversion step of converting into an image; including, wherein the reflection image before conversion used by the image conversion step corresponds to at least a part of the total reflection image, and the actual image before conversion used by the image conversion step is, It is characterized in that it corresponds to at least a part of the entire real image.

In addition, the object recognition method of the present invention includes an object region detection step of detecting an area in which an object exists from each of the reflected image, the real image, the imitation real image, and the imitation reflection image; a feature point extraction step of extracting feature points of each of the reflection image, the real image, the imitation real image, and the imitation reflection image; and an object detection step of detecting an object existing in the image obtained from the camera using information on a region in which the object detected in the object region detection step exists and the feature point extracted in the feature point extraction step; it is preferable

Specifically, in the feature point extraction step, the feature point of the reflective image and the feature point of the imitation reflection image are matched with each other, and the feature point of the real image and the feature point of the imitation real image are matched with each other.

In addition, in the object detection step, a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image, with respect to the region detected as the region in which the object exists by the object region detection step; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image. It is preferable to detect the region as an object of the first object group.

In addition, in the object detection step, a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image, with respect to the region detected as the region in which the object exists by the object region detection step; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using, the first matching ratio or the second matching ratio is less than a preset second ratio. It is characterized in that the region is detected as an object of the second object group.

Preferably, in the image conversion step, a reflection image of N partial regions corresponding to a partial region among the total reflection images; and an actual image of N partial regions corresponding to the reflection images of the N partial regions, respectively, among the total real images of the internal space in which the object is stored; a real image imitation of each of the reflection images of the N partial regions and converts each of the actual images of the N partial regions into a imitation reflection image, wherein N is an integer of 2 or more.

In addition, in the object recognition method of the present invention, the position information of the object of the first object group and the object of the second object group detected for each of the reflected images of the N partial regions and the actual images of the N partial regions respectively By unifying the duplicated detected objects among the objects detected for all of the reflection images of the N partial areas and the actual images of the N partial areas by using the location information, the entire object actually stored in the internal space It is preferable to further include; an object determination step of confirming.

In addition, the object recognition method of the present invention, an image acquisition step of acquiring images of a plurality of mirrors installed on a plurality of sidewalls of an internal space that can store an object and the internal space; further comprising, the front surface of the plurality of mirrors The glass can be made transparent or opaque by application of an electrical signal.

According to the deep learning-based image object recognition system and the recognition method of the present invention, when a product is stored in an internal space such as a refrigerator or a vending machine, an image of the product obscured by a reflection image using a mirror is also secured and the corresponding image is obtained. product can be recognized.

1 is an exemplary view of a camera image obtained for a narrow interior space.
2 is a block diagram of an image object recognition system based on deep learning according to a preferred embodiment of the present invention.
Figure 3 is an exemplary view of a plurality of mirrors installed on the side wall of the inner space of the storage box.
4 is an exemplary diagram of an image area acquired by a camera;
5 is an exemplary diagram of a timing diagram of an electrical signal for controlling each transparent state of the windshield of a plurality of mirrors;
6 is an exemplary view of an image area separated by an image separator;
Fig. 7 is an explanatory diagram of image conversion by an image conversion unit;
Fig. 8 is an explanatory diagram of an object detection method by an object detection unit;
Fig. 9 is an explanatory diagram of a method for determining an object by an object determining unit;

Hereinafter, a system for recognizing an image based on deep learning and a method for recognizing the same according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. Of course, the following examples of the present invention are not intended to limit or limit the scope of the present invention only to embody the present invention. What can be easily inferred by an expert in the technical field to which the present invention pertains from the detailed description and examples of the present invention is construed as belonging to the scope of the present invention.

First, FIG. 2 shows a configuration diagram of an image object recognition system 100 based on deep learning according to a preferred embodiment of the present invention.

As can be seen from FIG. 2, the object recognition system 100 of an image based on deep learning according to a preferred embodiment of the present invention includes an image acquisition unit 10, an image separation unit 20, an image conversion unit ( 30 ), an object region detection unit 40 , a feature point extraction unit 50 , an object detection unit 60 , and an object determining unit 70 .

The image separation unit 20, the image conversion unit 30, the object region detection unit 40, the feature point extraction unit 50, the object detection unit 60 and the object determination unit 70 except for the image acquisition unit 10, It may be implemented as part of a computing device.

The image acquisition unit 10 serves to acquire an image. Specifically, the image acquisition unit 10 may include a plurality of mirrors installed on a plurality of sidewalls inside a storage box (S) capable of storing objects, such as a refrigerator or a vending machine; And it is preferable to include a; and a camera for acquiring images of the inside and a plurality of mirrors in the storage box (S). Here, as an example of an object, various goods may be mentioned as an example.

3 is an exemplary view of a plurality of mirrors installed on the side wall of the inner space of the storage box (S) such as a refrigerator or a vending machine. In addition, FIG. 4 is an exemplary diagram of an image area acquired by a camera.

By application of an electrical signal, the front glass of the plurality of mirrors can be controlled to be transparent or opaque. The front glass may enable transition of a transparent or opaque state by attaching a PDLC film or the like. For example, when an electric signal is applied to the film of the corresponding windshield, an electric current may flow from the opaque state to the transparent state. However, depending on the film, when an electric signal is applied to the film, the film may be changed from a transparent state to an opaque state.

When a plurality of mirrors operate at the same time, the image inside the storage box (S) is not only reflected on each mirror, but also the image reflected on other mirrors is reflected, making it difficult to specify the actual reflected image inside the storage box (S) There is this.

In order to solve this problem, it is preferable that the sections in the transparent state of each of the front glass of the plurality of mirrors are different from each other.

5 is an exemplary diagram of a timing diagram of an electrical signal for controlling each transparent state of the windshield of a plurality of mirrors; As can be seen from FIG. 5, it is preferable that each mirror sequentially becomes transparent by the application of an electric signal.

As shown in FIG. 5 , by temporarily setting a section in a transparent state of each of the front glasses of a plurality of mirrors, mirrors other than one mirror in a transparent state become opaque and the image is not reflected, so the opaque mirror There is also no reflection of the image of the transparent mirror.

6 is an exemplary view of an image region separated by the image separator 20 .

The image separation unit 20 serves to separate the image acquired from the camera into a reflection image reflected by a mirror and an actual image of the internal space in which the real object is stored.

Specifically, it is preferable that the image separation unit 20 selects and separates a plurality of images obtained when a plurality of mirrors are sequentially and temporally operated, by selecting images corresponding to each other as follows.

(1) a first reflection area, which is a partial area of the area of the first mirror, and a first inner area of the inner space of the storage box (S) adjacent to the side wall where the first mirror is installed.

(2) a second reflection area, which is a partial area of the area of the second mirror, and a second inner area of the inner space of the storage box (S) adjacent to the side wall where the second mirror is installed.

(3) a third reflection area, which is a part of the area of the third mirror, and a third inner area of the inner space of the storage box (S) adjacent to the side wall where the third mirror is installed.

(4) a fourth reflection area, which is a partial area of the area of the fourth mirror, and a fourth inner area of the inner space of the storage box (S) adjacent to the side wall where the fourth mirror is installed.

In the present invention, the first to fourth reflective regions are not all regions of the corresponding mirror, but are limited to a part of the reflected region of the product, which is an object, to optimize the image region for comparison. Similarly, the comparison area was optimized only for the area adjacent to the mirror for the inner space of the storage box (S).

In addition, a first image for comparing the first reflective region and the first internal region, a second image for comparing the second reflective region and the second internal region, and the third reflective region and the third internal region are compared The third image and the fourth reflective region and the fourth inner region for comparison are obtained from a camera visually with each other.

That is, only when the mirror becomes transparent and reflects the image of the internal space of the storage box (S), the comparison of the reflective area with the internal space is meaningful.

7 is an explanatory diagram of image conversion by the image conversion unit 30 .

The image conversion unit 30, by artificial intelligence, converts the reflected image separated by the image separation unit 20 into a imitation real image that imitates the real image, and the real image separated by the image separation unit 20 plays a role in converting the reflected image into a mimicking reflective image.

In the case of an image obtained due to the reflection of a mirror, it is not easy to determine that it is the same object because it is taken completely different from the one taken from above when the tilt or reflection angle is taken into account. If the mirror recognizes the image as the same object, complex calculations such as calculating the coordinates in the reflected image by taking into account the angle of reflection and camera position, and warping the image accordingly are required.

In the image conversion unit 30, an image conversion model is used based on artificial intelligence, specifically, deep learning. In addition, as the image conversion unit 30 , a conditional GAN (cGAN) may be used as a deep learning model.

That is, by the learning model by the image conversion unit 30, the imitation real image assumes that the reflective image is located in the actual internal space and converts the composition of the reflective image to imitate the image when located in the actual internal space. it paid off Similarly, the imitation reflection image simulates the image reflected by the mirror by changing the composition of the real image on the assumption that the real image is reflected by the real mirror.

As can be seen from the description of the image separation unit 20 , the pre-conversion reflection image used by the image conversion unit 30 corresponds to at least a part of the total reflection image, and is used by the image conversion unit 30 . The actual image before conversion to be converted is characterized in that it corresponds to at least a part of the entire actual image.

That is, the image conversion unit 30 may include: reflection images of N partial regions corresponding to partial regions among the total reflection images separated by the image separation unit 20 ; and N real images of the N partial regions, respectively, corresponding to the reflection images of the N partial regions among the total real images of the internal space in which the object is stored. Accordingly, the image conversion unit 30 converts each of the reflection images of the N partial regions into a imitation real image, and converts each of the real images of the N partial regions into a imitation reflection image. Here, N is an integer of 2 or more. For example, when four mirrors are used in the image acquisition unit 10, N becomes 4.

When there is a reflection image of four mirrors, the image conversion unit 30 converts the image as shown in Table 1 below.

number before conversion
reflection image after conversion
imitation real image before conversion
real image after conversion
imitation reflection image One first reflective area
image first inner region
imitation image first inner region
image first reflective area
imitation image 2 second reflective area
image second inner area
imitation image second inner area
image second reflective area
imitation image 3 third reflective area
image 3rd inner area
imitation image 3rd inner area
image third reflective area
imitation image 4 fourth reflective area
image 4th inner area
imitation image 4th inner area
image fourth reflective area
imitation image

In [Table 1], an image imitating the first internal region image using the first reflection region image is the first internal region imitation image. Similarly, by using the first inner region image, an image simulating the first reflective region image becomes the first reflective region imitation image.

The object region detection unit 40 serves to detect a region in which an object exists from each of the reflective image, the real image, the imitation real image, and the imitation reflection image.

That is, the object region detection unit 40 detects a region in which an object exists from each of the N reflected images, N real images, N imitation real images, and N imitation reflection images.

Specifically, the object region detection unit 40 may detect a region in which an object exists by a deep learning learning model including a convolutional neural network and an anchor.

The convolutional neural network is composed of multiple layers and outputs a feature map. The anchor serves to detect the area of an object using the feature map that is the output of the convolutional neural network.

The feature point extraction unit 50 serves to extract feature points of each of the reflection image, the real image, the imitation real image, and the imitation reflection image. That is, the feature point extraction unit 50 extracts feature points from each of the N reflection images, N real images, N imitation real images, and N imitation reflection images.

For example, when there are 4 mirrors, the feature points from each of the first to fourth reflective region images, the first to fourth internal region images, the first to fourth internal region imitation images, and the first to fourth reflective region imitation images will extract

In addition, the feature point extraction unit 50 matches the feature points of the reflective image with the feature points of the corresponding imitation reflection image, and matches the feature points of the real image with the feature points of the corresponding imitation real image with each other.

For example, the feature point extraction unit 50 may match the feature point of the first reflection region image with the feature point of the first internal region imitation image, and match the feature point of the first internal region image with the first internal region imitation image with each other will do In addition, the similarity between the two (2) images can be known by this matching ratio.

The feature point extraction unit 50 may use Speed Up Robust Features (SURF).

8 is an explanatory diagram of an object detection method by the object detection unit 60 .

The object detection unit 60 is configured to detect an object existing in an image obtained from a camera using information on a region in which the object detected by the object region detection unit 40 exists and the characteristic point extracted by the characteristic point extraction unit 50 . plays a role

Specifically, the object detection unit 60 may include a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflection image, with respect to the region detected by the object region detector 40 as the region where the object exists; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using a first matching ratio or a second matching ratio, an area having a matched feature point equal to or greater than a preset first ratio It is detected as an object of a group of objects. That is, when the similarity is equal to or greater than the first ratio, it is detected as an object of the first object group appearing in both the reflected image and the actual image.

In addition, the object detection unit 60 may include a first matching ratio, which is a matching ratio between the characteristic points of the reflective image and the characteristic points of the imitation reflection image, with respect to the region detected as the region in which the object exists by the object region detection unit 40 ; and a matching ratio between the feature point of the real image and the feature point of the imitation real image; using the first matching ratio or the second matching ratio, an area having a matched feature point less than a preset second ratio is defined as an object of the second object group. detected with That is, when the similarity is less than the second ratio, it is detected as an object of the second object group appearing only in one of the reflected image and the actual image.

9 is an explanatory diagram of a method for determining an object by the object determining unit 70 .

However, in FIG. 9 , each image is simplified and exaggerated for explanation.

As shown in FIG. 9, when there is a small object 'D' hidden by one large object 'C', the corresponding object 'D' appears in the first reflection area image and the third reflection area image, which are two reflection images, so that the object detection unit ( 60), the object 'C' and the object 'D' are detected even when the object is detected using the first reflective region image, and the object 'C' and the object 'D' are detected even when the object is detected using the third reflective region image. will be detected. That is, in this case, object 'C' is an object of the first object group, and object 'D' is an object of the second object group.

Accordingly, the object determining unit 70 may include the position information of the object of the first object group and the position information of the object of the second object group detected with respect to each of the reflection images of the N partial regions and each of the actual images of the N partial regions. by unifying the duplicated detected objects among the objects detected for all the reflection images of the N partial regions and the real images of the N partial regions by using to determine the entire object actually stored in the internal space do. That is, the duplicately detected object 'C' and object 'D' are unified into one, respectively, and are detected as object 'C' and object 'D'.

By unifying the duplicated object, the object determining unit 70 may determine the entire object stored in the internal space of one storage box (S) without duplication.

Hereinafter, an image object recognition method based on deep learning according to a preferred embodiment of the present invention will be described.

Since the method for recognizing an object of an image of the present invention uses the above-described image object recognition system 100, all features of the image object recognition system 100 are included even if there is no separate explanation. In addition, at least a part of the image object recognition method of the present invention may be implemented in the form of a computer program executed by a computing device.

The image object recognition method of the present invention includes an image acquisition step (S10), an image separation step (S20), an image conversion step (S30), an object region detection step (S40), a feature point extraction step (S50), an object detection step ( S60) and an object determination step (S70).

Specifically, in the image acquisition step ( S10 ), an image of a plurality of mirrors installed on a plurality of sidewalls of an internal space and an internal space capable of storing an object is acquired.

In addition, in the image separation step (S20), the image obtained from the camera is separated into a reflection image reflected by the mirror and an actual image of the internal space in which the real object is stored.

For example, in the image separation step (S20), a first reflection area, which is a partial area of the area of the first mirror among the plurality of mirrors, is selected, and the sidewall on which the first mirror is installed and the adjacent real object are stored. A first inner region is selected as an region corresponding to the first reflective region, a second reflective region that is a partial region of the second mirror among the plurality of mirrors is selected, and a real object adjacent to the sidewall on which the second mirror is installed A second inner area of the inner space in which is stored is selected as an area corresponding to the second reflection area.

Next, in the image conversion step (S30), the reflected image separated by the image separation step (S20) by artificial intelligence is converted into an imitation real image that imitates the real image, and separated by the image separation step (S20) The resulting real image is converted into a imitation reflection image that mimics the reflection image. In addition, the reflection image before conversion used in the image conversion step S30 corresponds to at least a part of the total reflection image. Similarly, the real image before conversion used in the image conversion step S30 corresponds to at least a part of the entire real image.

Specifically, in the image conversion step ( S30 ), among the total reflection images separated by the image separation step ( S20 ), the reflection images of N partial areas corresponding to the partial areas; and N real images of the N partial regions, respectively, corresponding to the reflection images of the N partial regions among the total real images of the internal space in which the object is stored. In addition, in the image conversion step ( S30 ), each of the reflection images of the N partial regions is converted into an imitation real image, and each of the real images of the N partial regions is converted into a imitation reflection image. Here, it is preferable that N is an integer of 2 or more.

In addition, in the object region detection step S40, a region in which an object exists is detected from each of the reflective image, the real image, the imitation real image, and the imitation reflective image.

In the feature point extraction step ( S50 ), each feature point of the reflection image, the real image, the imitation real image, and the imitation reflection image is extracted.

In addition, in the feature point extraction step ( S50 ), the feature points of the reflective image and the feature points of the imitation reflection image are matched with each other, and the feature points of the real image and the feature points of the imitation real image are matched with each other.

In the object detection step S60, the object existing in the image acquired from the camera is identified using the information on the area in which the object detected in the object area detection step S40 exists and the feature point extracted in the feature point extraction step S50. serves to detect.

Specifically, in the object detection step S60, the first matching ratio, which is the matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image, with respect to the region detected as the region where the object exists by the object region detection step S40. ; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using a first matching ratio or a second matching ratio, an area having a matched feature point equal to or greater than a preset first ratio It is detected as an object of a group of objects.

In addition, in the object detection step S60, the first matching ratio, which is the matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image, with respect to the region detected as the region where the object exists by the object region detection step S40; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using the first matching ratio or the second matching ratio to generate a region having a matched feature point that is less than a preset second ratio It is detected as an object of the 2 object group.

In the object determination step (S70), the position information of the object of the first object group and the position information of the object of the second object group detected for each of the reflection images of the N partial regions and each of the actual images of the N partial regions are used. Accordingly, by unifying the duplicated objects among the objects detected for all of the reflection images of the N partial regions and the actual images of the N partial regions, the entire object actually stored in the internal space is determined.

As described above, according to the deep learning-based image object recognition system 100 and the recognition method of the present invention, when a product is stored in an internal space such as a refrigerator or a vending machine, it is hidden by the reflection image using a mirror. It can be seen that the image of the product is also secured so that the product can be recognized.

100: object recognition system
10: image acquisition unit
20: image separator
30: image conversion unit
40: object area detection unit
50: feature point extraction unit
60: object detection unit
70: object confirmation part
S : storage box
M: mirror

Claims (18)

An object recognition system comprising:
an image conversion unit that converts the reflected image separated from the image acquired from the camera into a imitation real image that imitates the real image, and converts the real image separated from the image acquired from the camera into a imitation reflective image that mimics the reflection image;
an object region detection unit configured to detect a region in which an object exists from each of the reflection image, the real image, the imitation real image, and the imitation reflection image;
a feature point extraction unit for extracting feature points of each of the reflection image, the real image, the imitation real image, and the imitation reflection image; and
an object detection unit configured to detect an object present in the image obtained from the camera using information on a region in which the object detected by the object region detection unit exists and the characteristic point extracted by the characteristic point extraction unit;
The reflection image before conversion used by the image conversion unit corresponds to at least a part of the total reflection image,
The real image before conversion used by the image conversion unit corresponds to at least a part of the entire real image. delete According to claim 1,
The feature point extraction unit,
The object recognition system, characterized in that the feature points of the reflected image and the feature points of the imitation reflection image are matched with each other, and the feature points of the real image and the feature points of the imitation real image are matched with each other. 4. The method of claim 3,
The object detection unit,
a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image, with respect to the region detected as the region in which the object exists by the object region detection unit; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image. An object recognition system, characterized in that the region is detected as an object of the first object group. 5. The method of claim 4,
The object detection unit,
a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image with respect to the region detected by the object region detection unit as a region in which an object exists; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using, the first matching ratio or the second matching ratio is less than a preset second ratio. An object recognition system, characterized in that the region is detected as an object of the second object group. 6. The method of claim 5,
The image conversion unit,
reflection images of N partial regions corresponding to partial regions among the total reflection images; and N real images of the N partial regions respectively corresponding to the reflection images of the N partial regions among the total real images of the internal space in which the object is stored.
Converting each of the reflection images of the N partial regions into a imitation real image, and converting each of the real images of the N partial regions into a imitation reflection image,
The object recognition system, characterized in that N is an integer of 2 or more. 7. The method of claim 6,
The object recognition system,
Using the position information of the object of the first object group and the position information of the object of the second object group detected for each of the reflection images of the N partial regions and the actual images of the N partial regions, respectively, the N parts An object determining unit for unifying the duplicated detected objects among all the reflected images of the region and for all of the actual images of the N partial regions to determine the entire object actually stored in the internal space; further comprising Object recognition system, characterized in that. According to claim 1,
The object recognition system,
An image acquisition unit for acquiring an image; further comprising,
The image acquisition unit,
a plurality of mirrors installed on a plurality of side walls of an interior space capable of storing objects; and
and a camera for acquiring images of the interior space and the plurality of mirrors. 9. The method of claim 8,
The front glass of the plurality of mirrors,
An object recognition system, characterized in that it becomes transparent or opaque by application of an electrical signal. In the object recognition method,
an image conversion step of converting the reflected image separated from the image acquired from the camera into a imitation real image that imitates the real image, and converting the real image separated from the image acquired from the camera into a imitation reflective image that mimics the reflection image;
an object region detection step of detecting a region in which an object exists from each of the reflection image, the real image, the imitation real image, and the imitation reflection image;
a feature point extraction step of extracting feature points of each of the reflection image, the real image, the imitation real image, and the imitation reflection image; and
an object detection step of detecting an object existing in the image obtained from the camera using information on a region in which the object detected in the object region detection step exists and the feature point extracted in the feature point extraction step;
The reflection image before conversion used by the image conversion step corresponds to at least a part of the total reflection image,
The real image before conversion used by the image conversion step corresponds to at least a part of the entire real image. delete 11. The method of claim 10,
In the feature point extraction step,
An object recognition method, characterized in that the feature points of the reflected image and the feature points of the imitation reflection image are matched with each other, and the feature points of the real image and the feature points of the imitation real image are matched with each other. 13. The method of claim 12,
In the object detection step,
a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image, with respect to the region detected as a region in which an object exists by the object region detection step; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image. An object recognition method, characterized in that the region is detected as an object of the first object group. 14. The method of claim 13,
In the object detection step,
a first matching ratio, which is a matching ratio between the feature points of the reflective image and the feature points of the imitation reflective image with respect to the region detected as the region in which the object exists by the object region detecting step; and a second matching ratio, which is a matching ratio between the feature points of the real image and the feature points of the imitation real image; using, the first matching ratio or the second matching ratio is less than a preset second ratio. An object recognition method, characterized in that the region is detected as an object of the second object group. 11. The method of claim 10,
In the image conversion step,
reflection images of N partial regions corresponding to partial regions among the total reflection images; and N real images of the N partial regions respectively corresponding to the reflection images of the N partial regions among the total real images of the internal space in which the object is stored.
Converting each of the reflection images of the N partial regions into a imitation real image, and converting each of the real images of the N partial regions into a imitation reflection image,
The N is an object recognition method, characterized in that it is an integer of 2 or more. 16. The method of claim 15,
The object recognition method comprises:
Using the position information of the object of the first object group and the position information of the object of the second object group detected for each of the reflection images of the N partial regions and the actual images of the N partial regions, respectively, the N parts Object determination step of unifying the duplicately detected objects among the objects detected for all of the reflected images of the region and the actual images of the N partial regions to determine all objects actually stored in the internal space; further comprising: Object recognition method, characterized in that. 11. The method of claim 10,
The object recognition method comprises:
An image acquisition step of acquiring images of an internal space capable of storing an object and images of a plurality of mirrors installed on a plurality of sidewalls of the internal space; 18. The method of claim 17,
The front glass of the plurality of mirrors,
An object recognition method, characterized in that it becomes transparent or opaque by application of an electrical signal.

Images