JPH0738223B2 - Image recognition device for mobile robots - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention (Industrial field of application) The present invention is based on information related to both basic figures extracted from image information of a recognition object and basic figures of a reference model. The present invention relates to an image recognition device for a mobile robot that recognizes an object to be recognized corresponding to a reference model.

(Prior Art) When a mobile robot autonomously travels, an imaging device such as a camera is provided on the robot as eyes of the mobile robot,
It is necessary to travel while detecting the target object from the image captured by this imaging device.

As a method of detecting a target object in this way, conventionally, the features relating to the target object are registered in advance as a reference model,
There is a method of detecting a target object in an image using the characteristic information of the reference model. Specifically, as one of the methods, an image of a target object or a typical two-dimensional shape is registered as a reference model, and matching between the figure of the object to be recognized in the captured image and the figure of the reference model is detected. By doing so, the reference model corresponding to the recognition object is detected, and the recognition object is recognized as the target object from the reference model.

(Problems to be Solved by the Invention) The target is usually a three-dimensional object, but the shape of the object obtained by imaging such a three-dimensional object as an image changes and differs depending on the direction in which the target is viewed. As described above, the conventional method for detecting a target object using a simple two-dimensional shape has a problem that the target object cannot be detected accurately and the detection accuracy is poor.

Another method is to generate various two-dimensional projected figures by using a three-dimensional model in which the three-dimensional structure of the target object is described in CAD, etc., and detect the matching on the image by these two-dimensional figures. However, there is a problem that it takes time and effort to input a three-dimensional model, and analysis and verification of the model become complicated.

The present invention has been made in view of the above, and an object of the present invention is to provide an image recognition device for a mobile robot that performs recognition processing using a two-dimensional figure in a relatively simple and accurate manner.

[Configuration of the Invention] (Means for Solving Problems) In order to achieve the above object, an image recognition device for a mobile robot according to the present invention is a mobile robot for recognizing an object to be recognized corresponding to a reference model from image information. An image recognition apparatus, as shown in FIG. 1, a basic figure storing means 2 for storing a basic figure forming a reference model together with parameters of the figure.
1. Positional relationship storage means 23 for storing the positional relationship between the basic figures of the reference model, basic figure extraction means 25 for extracting the basic figure together with the parameters of the figure from the image information of the object to be recognized, The positional relationship detecting means 27 for detecting the positional relationship between the extracted basic figures, and the recognized object corresponding to the reference model based on the positional relationship between the basic figure of both the reference model and the object to be recognized and the basic figure. The gist is to have an identification means 29 for identifying the body from the image information.

(Operation) The image recognition apparatus for a mobile robot according to the present invention stores in advance the basic figure forming the reference model, the parameters of the basic figure, and the positional relationship between the basic figures, and the basic figure is calculated from the image information of the object to be recognized. The figure is detected together with the positional relationship between the parameter and the basic figure, and the recognized object corresponding to the reference model is detected from the image information based on the positional relationship between the basic figure, the parameter and the basic figure of both the reference model and the recognized object. It has recognized.

(Examples) Examples of the present invention will be described below with reference to the drawings.

FIG. 2 is a block diagram showing the configuration of an image recognition device for a mobile robot according to an embodiment of the present invention. The image recognition device for a mobile robot stores an image of a target object, which is an object to be recognized, captured by the mobile robot in an image memory 5 via an image input unit 4. The image information stored in the image memory 5 is supplied to the basic figure extracting unit 2, where the basic figure and various information related to the basic figure are extracted from the image information.

On the other hand, the image recognition apparatus has a target object model creation unit 1 for creating and inputting a target object model corresponding to the target object, that is, a reference model, and supports the target object to be detected from the target object model creation unit 1. While decomposing the information related to the target model into the basic figures forming the target model,
Information related to the basic figure and a method of extracting the basic figure
It is analyzed as information regarding the order and the like, and these pieces of information are stored in the target object model storage unit 6.

More specifically, the target object model is generally composed of some two-dimensional basic figures, but in the present invention, the target object model is previously decomposed into basic figures and subdivided, and the basic figures are described above. The comparison is made not only with the comparison of the basic figures but also with the parameters such as the size of the basic figures, the basic figures, and the like. The recognition operation is performed by comparing the positional relationship between figures and identifying whether or not the object in the captured image is equal to the target object model.

Therefore, the target object model storage unit 6 stores the following first to fourth information from the target object model for the target object to be detected by the input from the target object model creating unit 1.

1. Types of basic figures that compose the target 2. Basic figure extraction processing method 3. Basic figure parameters 4. Relationship between basic figures and extraction sequence A concrete explanation will be given with reference to FIG. 3 to FIG. .

First, the first information regarding the types of basic figures will be explained. As basic figures, there are various basic figures such as ellipses (and circles), parallel lines, rectangles, triangles, etc. It is stored corresponding to the physical model. For example, in FIG. 3, when the meter surrounded by a broken line frame is the target object, as the first information, as shown in FIG. 4, a circle 11, parallel lines 12 and 13 are extracted and the target object model is extracted. It is stored in the storage unit 6.

Also, as the second information regarding the basic figure extraction processing method, there is a procedure for detecting a contour line for extracting a basic figure,
That is, a contour line detection procedure including a procedure in which differential filtering is performed to extract a changed portion having a contour line, this is binarized, noise is removed, and then thinning is performed is stored together with a threshold value.

Furthermore, as the third information regarding the parameters of the basic figure, when the basic figure is an ellipse, its center coordinates, inclination,
The major axis length and the minor axis length are stored, and in the case of parallel lines, the inclination,
The interval, length, etc. of the overlapping portion are stored.

As the fourth information regarding the relationship between the basic figures and the extraction order, the extraction order of the basic figures, the arrangement on the image, the positional relationship between the basic figures, the connection relationship, and the like are stored.

In the extraction order, strong features in the image such as vertical edges and long edges and features unique to the target object such as ellipses and circles are preferentially extracted. For example, in the meter example of FIG. 4, the circle 11 is first extracted, then the parallel line 13 is extracted, and finally the parallel line 12 is extracted.

As for the arrangement on the image, when the relative positional relationship between the basic figures is focused, that is, when one certain basic figure is focused, data describing the positions of all the other basic figures based on this basic figure is used. . For example, in the case of the example shown in FIG. 4, vector data between the centers of the respective basic figures (that is, data such as the distance when the upper left corner of the image is the origin and the angle from the X axis) are stored. . Further, as the connection relationship, which basic figures are connected to each other on the image is described. That is, in the example of FIG.
Parallel line 12 "(circle 11 is connected to parallel line 12)," parallel line 12-circle 11, parallel line 13 "(parallel line 12 is connected to circle 11 and parallel line 13), and A connection list such as "parallel line 13-parallel line 12" (parallel line 13 is connected to parallel line 12) is stored.

The above-described first to fourth information are stored in the target object model storage unit 6 in association with each target object model, and these information are stored in the target object model storage unit 6 into the basic figure extracting unit 2 and the target object. It is supplied to the decision unit 3. As described above, the basic figure extracting unit 2 receives the image information of the target object which is the recognition target input from the image memory 5, and the target object model is described based on this image information. Similarly, the basic graphic and information related to the basic graphic are extracted.

Then, the information related to the extracted basic figure for the object to be recognized is supplied to the target object determining unit 3, where the information related to the basic figure for the target object model input from the target object model storage unit 6 is stored. It is collated and it is recognized whether or not the object to be recognized is the target object model.

Next, the operation will be described with reference to the flowchart of FIG.

In recognizing the target object which is the recognition target by the image recognition device for the mobile robot, the basic figure for the target model as described above and various information related to the basic figure are created. It is previously stored in the target object model storage unit 6 via the unit 1, and the operation will be described from this state.

When image information captured by an imaging device (not shown) provided in the mobile robot is input to the image input unit 4 (step 11
0), this image information is stored in the image memory 5, and is input from the image memory 5 to the basic figure extracting unit 2. The basic figure extraction unit 2 extracts and sets a basic figure (hereinafter, referred to as a key figure) which is a reference of the target object from this image information (step 120). This key figure is selected according to the extraction order of the fourth information. This selected key figure is the second
And extracted using the third information (step 13
0). If the key figure is not extracted, the key figure is selected again according to the extraction order of the fourth information (step 140).

Also, because there is a possibility that multiple key figures will be extracted,
Other basic figures forming the target object are extracted to narrow down the target object candidates. For this reason, an area on the image where other basic figures other than the key figure are expected to exist is set (steps 150 to 170). The data of the fourth information of the target object model storage unit 6 is used for this setting. That is,
Using the relative position information between the key figure and other basic figures, find the amount of displacement on the image from the key figure to the other basic figure, and add this amount of displacement to the position of the key figure on the image. Set the window centered around the position. The size of the window is determined using the third information. For example, if the basic figure of interest is an ellipse, the major axis length is used, and if it is a parallel line, the inclination and the length of the overlapping portion are used. FIGS. 5 (a) and 5 (b) show an example of area setting, but the relative position relation data (γ, θ) is used for the circle which is the key figure to determine the center of the window. And set the window shown by the broken line.

After setting the search area in this way, the basic figure is detected in this area using the second and third information (step 180). In this case, among the basic figures detected according to the processing procedure indicated by the second information, one that is close to the parameter of the target object model of the third information is selected.

By repeating the above processing, all the basic figures other than the key figures that constitute the target object that is the object to be recognized are detected (step 190), and the same processing is performed for all of the key figure candidates, A basic figure is detected for all the figure candidates (step 200).

When the key figure and the basic figure for each key figure candidate are thus detected, the certainty Si of each key figure candidate Ki is calculated by the following equation (step 210).

Here, Pij is the reliability of one basic figure Fij among the basic figures detected around the key figure candidate Ki, and the parameter shown in the third information of the target object model storage unit 6 and this It is obtained from the error between the basic figure Fij and the parameter. For example, in the case of parallel lines, the reliability Pij having a large value is given to the basic figure Fij having a small gap or inclination error. n is the number of basic figures (excluding key figures) that achieve the target object, and reliability Pij = O is given to the basic figure Fij that has not been detected. Qi is the key figure Ki
Is a degree of matching between the connection list L obtained with respect to the connection list Lm of the target object model obtained from the fourth information, and the connection list L obtained with respect to the key figure Ki is the connection list Lm of the target object model. In the case of a list that is a subset, the larger the number of elements, the larger the value. If the connection list L is not a subset of Lm, then Qi = O is given.

From the above definition, the larger the certainty factor Si, the more the target object, that is, the target object, having a basic figure configuration close to that of the target object model. Therefore, the target determination unit 3 selects the key figure Ki having the maximum certainty Si and outputs this key figure Ki and the basic figure Fij (j = 1-n) as the target corresponding to the target model. (Step 220).

[Effects of the Invention] As described above, according to the present invention, the basic figure forming the reference model, the parameters of the basic figure, and the positional relationship between the basic figures are stored in advance, and the image of the recognized object is also stored. The basic figure is detected from the information together with the positional relationship between the parameter and the basic figure, and the recognized object corresponding to the reference model is determined based on the positional relationship between the basic figure, the parameter and the basic figure of both the reference model and the recognized object. Since it is recognized, even a three-dimensional object to be recognized can be recognized accurately, accurately and relatively easily by converting it into a simple two-dimensional basic figure.

[Brief description of drawings]

1 is a diagram showing the basic configuration of the present invention, FIG. 2 is a block diagram showing the configuration of an image recognition device for a mobile robot according to an embodiment of the present invention, and FIGS. 3 to 5 are FIG. FIG. 6 is a diagram for explaining the graphic processing of the image recognition apparatus, and FIG. 6 is a flowchart showing the operation of the image recognition apparatus of FIG. 21 ... Basic figure storage means 23 ... Position relation storage means 25 ... Basic figure extraction means 27 ... Position relation detection means 29 ... Identification means

Claims (1)

[Claims] 1. An image memory for storing an image of a target object which is an object to be recognized captured by a mobile robot, and a target object model creating section for creating a reference target object model corresponding to each target object. A target object model storage unit that stores the basic figures forming each target object for each target object model created by this target object model creation section together with the extraction processing method, parameters, relative positional relationship between basic figures, and connection relationship And, for the image of the target object stored in the image memory, a key figure serving as a reference of the target object for each target object model is respectively stored using the extraction processing method and parameters stored in the target object model storage unit. The target object model is extracted and each window is set for each key figure on the image by using the relative positional relationship between the basic figures of the target object model. All the basic figures connected to each key figure are detected together with the parameters by using the basic figure extraction processing method and parameters, and the parameters of all the detected basic figures for each key figure correspond to the target model. The certainty of each key figure is determined from the error of the basic figure parameters and the degree of matching between the connection relationship of all the basic figures detected for each key figure and the connection relationship between the basic figures of the target model. With respect to the basic figure extraction unit to be calculated and the certainty of each key figure calculated by this basic figure extraction unit, select the key figure with the maximum certainty and all the basic figures connected to this, and select this key figure. An image recognition device for a mobile robot, comprising: a target object determination unit that outputs a target object corresponding to the target object model;