JPH02189691A - Device for recognizing object - Google Patents

Abstract

PURPOSE:To improve the recognition accuracy of a high-order object and a low-order object by collating the edge image of a detected object selected by a candidate selecting means with a detailed model segment and specifying the detected object in accordance with the collated result. CONSTITUTION:A constitutional part constituting the contour of a detected object is collated with a main model segment by the candidate selecting means, and when the object has collation more than a fixed value, the detected object is selected as a candidate. An edge image is found out through an edge image computing means by binarizing the differential images of plural density images having respectively different light irradiating directions and a composed edge image is found out by composing the edge images. The edge image has a fixed width and is not disconnected by disturbance like a contour line. The edge images of all candidates are collated with the detailed model segment by an object specifying means and the object is specified in accordance with the collated result. Since the edge of the object is not disconnected, the high-order object and the low-order object can be surely discriminated.

Description

[Detailed description of the invention] [Industrial application field]

TECHNICAL FIELD The present invention relates to an object recognition device.
The present invention relates to an object recognition device that can recognize the topmost object to be gripped at high speed and with high precision when gripping the topmost object.

[Prior art]

Conventionally, this type of object recognition device recognizes an object by correlating a captured grayscale image with a model grayscale image previously captured using a model object.
(Japanese Patent Laid-Open No. 62-269287), the object is recognized by extracting the contour of the object and comparing it with the contour of a pre-registered model.

[Problem to be solved by the invention]

However, since the former takes a correlation in one direction in a grayscale image, the recognized object may be limited to objects with simple shapes such as cylindrical objects, or even if there is an object at the top in the matching part, it may be at the bottom in other parts. There is a problem in that even objects that exist and are not suitable to be grasped as a whole are recognized as the topmost object. In addition, in the latter case, the contour line of the detected object extracted on the image is normally cut by the shadow V of disturbance such as illumination, so it is difficult to distinguish it from the case where the object is located below and the contour line is cut. matching with detailed contours of the model improves An lJk accuracy but reduces recognition speed; matching with some major contours of the model improves recognition speed, but contours that are not matched Even when a line portion is obstructed by a higher-level object, there is a problem in that the object is recognized as the highest-level object. The present invention has been made to solve the above problems, and its purpose is to enable a robot to grasp objects from among objects existing in a complex background or irregularly piled up. The objective is to efficiently recognize an object that is appropriate for the object, that is, the topmost object.

[Means to solve the problem]

The configuration of the invention for solving the above problems is as shown in the outline in FIG. The extracted line segments are compared with a part of the curve or straight line that constitutes the outline of the model to be compared (referred to as a model line segment in the present invention), and based on the degree of matching, the object is identified as RJ.
L'l object recognition device, comprising: main model line segment storage means storing information regarding the positions of main model line segments constituting the main part of the contour line of the model; and substantially the entire contour line of the model. a detailed model line segment storage means that stores information regarding the position of the detailed model line segment, etc., and the constituent line segments of the detected object are compared with the main model line segments stored in the main model line segment recording means. , a candidate selection means for determining the position and orientation of the detected object according to the comparison result and selecting a candidate for the detected object; and obtaining an edge image by binarizing differential images of a plurality of grayscale images with different illumination directions. , edge image calculation means for synthesizing these edge images to obtain a synthetic edge image; and on the synthetic edge image, based on the position and orientation of the detected object selected by the candidate selection means, The object identification means is provided for comparing detailed model line segments stored in the storage means with edge images of the detected object, and specifying the detected object according to the comparison result.

[Effect]

In the first stage, the candidate selection means compares the contour line of the detected object with the main model line segments and the main model line segments, and selects objects that match a certain amount or more as candidates for the detected object. In the second stage, an edge image is obtained by binarizing the differential images of a plurality of a-light side images with different illumination directions by the edge image calculation means, and a composite edge image is created by combining the edge images. This edge image is a composite of images obtained by illuminating from different directions, and is an image obtained by binarizing a differential image using a predetermined threshold. , has a constant width and will not be interrupted by disturbances like a contour line.Next, the object identification means compares the edge images of all candidates for the detected object selected in the first step with the detailed model line segments. The target object is identified according to the matching results.In this way, in the first step, the constituent line segments of the object are matched with the characteristic main model line segments to determine the position and orientation of the detected object. , the calculation time is shortened.In addition, in the second stage, detailed model line segments are projected onto the edge image based on the position and orientation of the detection object candidate selected in the first stage. The final detected object is identified based on the degree of coincidence at the time.This second stage matching is performed only with the candidates already selected in the first stage, or the candidate's position and orientation have already been determined. is required, so the calculation time is short because it is a simple comparison of how many detailed model line segments are recognized on the edge image.Also, in the edge image, the edges of the object are not interrupted due to disturbances. Therefore, it is possible to reliably distinguish between higher-ranking objects and lower-ranking objects.

【Example】

The present invention will be described below based on specific examples. In FIG. 2, many workpieces W are placed on a pallet gutter 5, and a camera 6 is provided to take an image of the workpieces W from above the pallet 5. In addition, as illumination devices for the workpiece W, a first illumination device @7 uniformly illuminates the workpiece W from the upper center, a second illumination device 8 illuminates the workpiece W from the upper left, and a second illumination device @7 illuminates the workpiece W from the upper right. A third lighting device 9 that emits light from both directions is provided. These lighting bags! ? , 8.9 are configured to illuminate the workpiece W when commanded by the illumination control circuit 2. The object recognition device A includes a central processing unit 1 that performs data processing such as matching and determination, and a configuration that processes a video signal obtained by a camera 6 to detect an outline of a detected object and configure the outline. An image processing device 3 performs data processing such as extracting line segments and obtaining a composite edge image;
It is composed of a device 4 and a lighting control circuit 2. Furthermore, the image processing device 3 includes an image input device 31 that samples the video signal output from the camera 6 and generates grayscale image data in which the grayscale level is digitized, and calculates a brightness gradient from the grayscale image data by differential calculation. , an edge detection device 32 that generates edge image data representing edges of an object image, traces a contour line from the edge image data, extracts constituent line segments that constitute the contour line, and detects the position of the constituent line segments. Line segment extraction device 33 that generates data
It is made up of. The storage device 4 is composed of a RAM, etc., and includes a main model line segment memory 41 that stores information regarding the positions of main model line segments that constitute the main part of the contour of the model, and Candidate data memory 42 that stores data related to the position and orientation of a detected object candidate obtained from a set of constituent line segments having a degree of matching equal to or higher than a predetermined value, and a detailed model line segment that constitutes substantially the entire contour of the model. a detailed model line segment memory 43 that stores information regarding the position, etc. of the
It is comprised of a recognition result memory 44 that compares the detailed model line segment with an edge image on the memory in the central processing unit corresponding to the selected detection object candidate and stores the result. Next, the operation of the object recognition apparatus A will be explained according to the flowcharts of FIGS. 3 and 4 showing the processing procedure of the object recognition apparatus A, and the explanatory diagram of FIG. 5. The first lighting device 7 is normally turned on by the lighting control circuit 2, and a video signal obtained by the camera 6 is input to the image input device 31. Then, in the image input device 31,
A grayscale image is generated by sampling a video signal and converting it into a digital signal. The grayscale image data is input to an edge detection device 32 and differentiated to generate an edge image. The edge image data is input to the line segment extraction device 33, and by tracing the edges, the outline of the object is extracted, and the outline is approximated by broken lines or circles, as shown in FIG. 5(a). A line segment image can be obtained. Then, in step 100, the central processing unit 1
Constituent line segments consisting of circles whose radius is approximately equal to the radius of the circle corresponding to the main model line segment and straight lines whose length is greater than a certain value are extracted from the line segment image, and in step 102, the center position of the circle, the radius, and the straight line are extracted. The data at both end positions are stored in area A and area B of the constituent line segment memory in the central processing unit, respectively. On the other hand, the main model line segment memory 41 stores position data of the center position of a circle, which is an element of the main model line segment, and the positions of both ends of the line segment, as shown in FIG. 5(b). In this example, the main model line segments are a circle C1 that shows the outline of a hole formed in the model, and straight lines Ll and L2 that show a part of the main twill lines.
．． It is composed of L3. Next, proceeding to step 104, the constituent line segment memory 34
By determining whether or not area A is empty, it is determined whether or not the constituent line segment to be verified remains. In the first execution cycle, as long as the component line segment extraction is successful, area A is naturally not empty, so step 106
Move to central processing bag ¥! One circle, which is an arbitrary component line segment, is selected from area A of the component line segment memory. Next, in step 108, area B of the constituent line segment memory
Three straight lines, which are arbitrary constituent line segments, are selected from . Then, the process moves to step 110, and the degree of coincidence is calculated for all combinations of the selected constituent line segments and the main model line segments stored in the main model line segment memory 41. The numeral ρ is a parameter that indicates how much the extracted constituent line segments and the main model line segments match, including their relative positional relationship, and in this example, it is calculated using the following formula. . ρ=Eω, ・(Δα, ・Δβl) However, ω is the weight given to the main model line segment,
The more the main model line segment has a large weight, the greater the degree of matching as a whole. ω1 is the weight of a circle, and ω2 to ω4 are the weights of straight lines. Further, Δα1 indicates the degree of matching between the i-th model line segment and the constituent line segments to be compared. In the case of a circle, a circle that is approximately equal to the radius of the circle of the main model line segment has already been extracted as a constituent line segment, so Δα+=1
．． It is O. Here, length(eJ) is the length of the extracted j-th constituent line segment eJ. Further, the Δβ amount indicates the degree of agreement between the relative positional relationship between the constituent line segments and the relative positional relationship between the main model line segments. Similarly, in the case of a circle, Δβ,=i,o. Also, regarding straight lines (1≠1), when the component line segment e exists in the region S, Δβi = 1.0 When the component line segment el is partially included in the region S, Δβi = 0
．． 5 If the constituent line segment e is not included in the area S at all, Δβ1−
0 However, the area S is defined by the following equation. This is the range in which the main model line segment - can exist, which is determined by the relative positional relationship between the main model lines when the segment e3 is matched one by one by matching the end points. Therefore, the region S is a common region of possible regions determined from the correspondence between the three main model line segments and the three component line segments. In this way, the numeric value ρ is calculated in step 110. In addition, in this example, the combination of the main model line segment and the selected component line segment corresponds to a circle, and the other three
There are six possible combinations depending on the correspondence between the two straight lines, and for each combination, a -number position is calculated, and among them, the largest set of -number positions is selected as a possible combination that corresponds to the main model line segment. be done. Then, in the next step 112, it is determined whether or not the numeral ρ is larger than the threshold Th. If the - numeral ρ is larger than the threshold Th, in step 118, the extracted constituent line The position of the detected object is calculated from the center position of the circle, which is one of the minutes, and the attitude of the detected object is calculated from the slope of the straight line of the component line segment corresponding to a specific straight line among the main model line segments. The candidate data memory 42 stores the candidate data. Then, in step 120, it is determined whether the number of selected detection object candidates is sufficient. If not, the process moves to step 122, and the extracted constituent line segments are It is deleted from the constituent line segment memory, and the process returns to step 104, where the next constituent line segment is extracted. On the other hand, in step 112, the -numeric value ρ is the threshold value T
If it is not larger than h, the correspondence set of the selected component line segment and the main model line segment is returned to region B with a label of evaluated. Then, proceeding to step 114,
It is determined whether or not there are other combinations to be extracted from among the constituent line segments stored in area B of the constituent line segment memory in the central processing unit 1, and if there are other combinations,
Proceeding to step 108, three other straight lines are extracted from the constituent line segments, and the single digit ρ is calculated for all corresponding sets thereof in the same manner as described above. If there are no other combinations of component line segments to be extracted, in step 116, the circle of the extracted component line segments is determined not to correspond to the model and is stored in the area of the component memory in the central processing unit 1. A
, the process returns to step 104, the next other circle is extracted, the other three straight lines are selected, a corresponding set is generated, and the degree of coincidence is calculated in the same way. Further, if the number of detected object candidates selected in step 120 is sufficient, selection of candidates is discontinued and the processes from step 124 onwards are executed. At this time, detection object candidates are selected as shown in FIG. 5(C). In step 124, the second illumination device 8 is turned on, the workpiece W is illuminated from the upper left, a video signal is input from the camera 6, and in step 126, the edge detection bag Vfi
32, the grayscale image is differentiated and binarized into an edge image, which is then stored in an edge image memory within the central processing unit. Next, in step 128, the third illumination device 9 is turned on, the workpiece W is illuminated from the upper right, a video signal is input from the camera 6, and in step 130, the edge detection device 3
By differentiating the grayscale image by 2, a binarized edge image at the time of upper right illumination is obtained. Next, in step 132, the central processing unit l calculates the logical sum of the edge image when the upper right side is illuminated and the edge image when the upper left side is illuminated, which is already stored in the edge image memory. A composite edge image as shown in Figure (d) is stored in the edge image memory. Next, in step 134, the candidate data memory 42
The position and orientation data of one detected object candidate stored in is selected. Next, in step 136, on the composite edge image stored in the edge image memory in the central processing bag ul, data is stored in the detailed model line segment memory 43 based on the position and orientation data of the detected object candidate. When a detailed model line segment as shown in FIG. 5(e) is projected, the number of pixels of overlap between the detailed model line segment and the selected edge of the detected object candidate in the composite edge image is 13. Calculated. However, B is the number of overlapping pixels in detailed model line segment e, and m is the number of detailed model line segments. Then, in the next step 138, the candidate data memory 42
If it is determined that the calculation of the number of overlapping pixels has not been completed for all detected object candidates stored in the
Returning to step 134, the number of overlapping pixels is calculated for other detection object candidates. Then, when the number of overlapping pixels 7b is calculated for all the selected detected object candidates, in step 140, the detected object with the maximum number of overlapping pixels γ3 is recognized as the uppermost object by SR. In this way, the topmost object is selected from a large number of irregularly stacked objects with the same shape as shown in Figure 5(f).
As shown in the figure, it can be recognized with high accuracy. Also, instead of 15 overlapping pixels, detailed model line segment el
When the number of pixels is V, the object with the maximum overlap ratio among the detection object candidates that satisfy Iil/vl≧T for all detailed model line segments may be set to the topmost object. However, T is the allowable value of the overlapping ratio of each detailed model line segment.

【Effect of the invention】

The present invention provides a candidate selection means for comparing constituent line segments of a detection object with main model line segments and selecting detection object candidates according to the comparison result, and differentiation of a plurality of grayscale images with different illumination directions. An edge image calculation means for obtaining a composite edge image obtained by binarizing and synthesizing the images, and comparing the edge image of the detected object selected by the candidate selection means with the detailed model line segment on the composite edge image. Since it has object identification means for identifying the detected object according to the result,
The candidate selection time is shortened, the matching method with detailed model line segments is simple and the matching time is shortened, and since matching is performed using edge images that are not interrupted by disturbances, it is possible to distinguish between higher-ranking objects and lower-ranking objects. WI has significant effects such as increased recognition accuracy.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an outline of the present invention;
FIG. 2 is a block diagram showing the configuration of an object recognition device according to a specific embodiment of the present invention, FIGS. 3 and 4 are flowcharts showing the operation of the device, and FIG. 5 is a data processing diagram. FIG. 1 Central processing unit 2 Lighting control circuit 3/image processing device 4 tα device 6-camera 7 First lighting device 8 Second lighting device 9 Third! ! @Mei device patent applicant Representative of Nippondenso Co., Ltd. Patent attorney Shudai Fujitani

Claims (1)

[Claims] A contour line of a detected object is obtained from a two-dimensional image, a plurality of constituent line segments constituting the contour line are extracted from the contour line, and a contour line of a model to be compared with the constituent line segments is extracted. In an object recognition device that recognizes an object based on the degree of matching by comparing model line segments that make up the model, the main model stores information such as the position of the main model line segments that make up the main part of the contour of the model. Line segment storage means; Detailed model line segment storage means that stores information regarding the positions, etc. of detailed model line segments that constitute substantially the entire contour of the model; Constituent line segments of the detected object and the main model line segments. candidate selection means for comparing the main model line segments stored in the storage means, determining the position and orientation of the detection object according to the comparison result, and selecting detection object candidates; an edge image calculation means for obtaining an edge image obtained by binarizing a differential image of a grayscale image and synthesizing these edge images to obtain a composite edge image; and a detection selected by the candidate selection means on the composite edge image. Object specifying means for comparing the detailed model line segment stored in the detailed model line segment storage means with an edge image of the detected object based on the position and orientation of the object, and specifying the detected object according to the comparison result. An object recognition device comprising: