JP2626751B2 - Object image recognition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an image recognition method capable of recognizing various kinds of objects.

2. Description of the Related Art When an object piled up at an arbitrary position and in a direction on a belt pallet on a belt conveyor or the like is to be gripped by the handling robot, the handling robot receives the information on the position and direction of the object. It is necessary to recognize an object, and one method of recognizing the object is an image recognition method. As this image recognition method, for example, a method of recognizing a bar-shaped component having a uniform cross section has been proposed (Japanese Patent Application Laid-Open (JP-A) no. No. 59-198307, JP-A-60-48579). This is a method in which an imaging device such as an ITV camera captures a rod-shaped component having a uniform cross section, and compares a light-dark change pattern in the image with a pattern taught in advance to find a match. The part can be image-recognized by extracting a cylindrical shape from inside.

As another example, in the case of a component having a hole, a recognition target object may be used, such as a method of measuring the number of holes and the position of the hole from a brightness change pattern on the screen. Various methods have been proposed for extracting a measurable local feature and performing recognition of the recognition target object.

Problems to be Solved by the Invention However, an object to be actually recognized generally has a complicated shape, and even if the local features of the object can be simply recognized, the entire shape of the object is grasped. Is difficult and
Therefore, in order to be able to recognize the entire shape of a part or the like, it is necessary to add an algorithm for recognizing the entire feature of the part for each individual part.
There is a problem in that it takes a lot of time to develop recognition software for each part, which increases the cost for introduction and increases the lead time. Also, since recognition software must be prepared for each part,
There was a problem that the versatility was poor.

The method of the present invention has been made in view of the above problems, and has as its object to provide a highly versatile image recognition method capable of coping with recognition of various types of objects.

Means for Solving the Problems In order to solve the above-mentioned problems, the present invention divides each object to be recognized into a plurality of portions, and focuses on a specific portion of a screen when each portion is imaged. When a shape feature including a light-dark pattern indicating a brightness distribution and a relative positional relationship including an arrangement order of each of the portions identified by these shape features are stored in advance, and the object is recognized. Then, the object is captured in an image, a first portion having a local feature is extracted from the image, and the shape feature of the extracted portion is compared with the stored shape feature. While identifying which part is divided, at least one other part having a local feature is extracted, and the shape characteristic of the extracted part is compared with the stored shape characteristic, The object Which part is divided by the
Recognizing the object by comparing a relative positional relationship between the portion having the local feature of the one and the portion having the other local feature with a previously stored relative positional relationship. It is.

Further, in the present invention, the arrangement direction of the recognition target object can be recognized based on the position of the center of gravity calculated from the position of each portion having local features extracted from the image.

According to the method of the present invention, first, the geometrical features including the light and dark patterns of the divided portions of the recognition target object and the relative positional relationships including their arrangement order are stored in advance. Then, at the time of object recognition, at least two parts of the object are specified based on the shape characteristics obtained from the image and the shape characteristics stored in advance, and the relative positional relationship between the parts having these shape characteristics is determined. Is compared with the relative positional relationship stored in advance to recognize the object. Therefore, according to the method of the present invention, the position and orientation of the recognition target object can be entirely recognized without necessarily requiring all data on the entire shape of the recognition target object.

Embodiment Hereinafter, a case in which a shaft part is recognized by the object image recognition method according to the present invention will be described.

Generally, shaft parts are shown in Fig. 2 (a) (b) (c)
As shown in (d), shaft parts 1, 2, 3, and 4, each of which has a composite shape of a plurality of types of cylindrical parts. For example, in the case of shaft part 1 in FIG. , Cylindrical part
a, a cylindrical portion b,..., a cylindrical portion g (see FIG. 3).

In order to recognize the shaft component 1 by an image recognition method, an image recognition device (not shown) is required in advance.
First, it is necessary to store the shape of the shaft component 1 which is the object to be recognized.

In order for the image recognition device to store the shape of the shaft part 1, first, as a first step, the shaft part 1 is divided into seven parts of the cylindrical part a, the cylindrical part b,. In FIG. 3, the shape characteristics of the portions enclosed by the broken lines and squares are captured by an imaging device (not shown) such as an ITV camera, and the distribution of brightness with respect to a specific portion of each cylindrical portion is taken. are shown dark pattern (see FIG. 3), the length _{(l a, l b, ......} l g), the diameter _{(d a, d b, ......} d g) to store respective information. Next, as a second step, the cylindrical parts a, b,
...... as the relative positional relationship between the cylindrical portion g, arrangement order (cylindrical left portion a, a cylindrical portion b, the order of the ...... cylindrical portion g) of the cylindrical portion and the interval _{(P 1, P 2, ......} P 6 ) Is stored.

As described above, in the image recognition apparatus, the shape characteristics of each part (for example, the parts a to g in the shaft part 1 in FIG. 3) having local characteristics of all recognition target parts and their relative positional relationships. Is stored in advance. The number of cylindrical portions to be stored as features of the recognition target component is the second
As in the case of the shaft part 2 shown in FIG.
The number is not limited as long as it is two or more as in the case of four shaft parts 3 in FIG. 3C, and a part unnecessary for recognition can be omitted. As in the case of the shaft part 4 (crankshaft), it is possible to recognize even if a shape other than a cylindrical shape is included.

Next, a scanning process for recognizing the case where the shaft component 1 is piled up as an object to be recognized in a pallet at an arbitrary position and direction will be described with reference to a flowchart shown in FIG.

When the image recognition device is switched on, the imaging device captures the shaft part 1 in the pallet, and the image of this shaft part 1 is projected on the screen, scanning starts. First, the geometrical features stored for the recognition target part The longest cylindrical portion (cylindrical portion e in FIG. 3) which is the information of (1) is searched on the screen (step 101, step 102). If the longest cylindrical portion is found, the process proceeds to step 103. In this step 103, there is another cylindrical portion to be searched for by comparing with the previously stored information on the shape characteristics of the cylindrical portion of the shaft component 1. A determination is made as to whether or not there is another cylindrical portion to be searched for, and the flow proceeds to step 104.

In step 104, other cylindrical portions (cylindrical portion d and cylindrical portion f in FIG. 3) which are coaxially adjacent to the longest cylindrical portion are searched for, and the relative positions of the cylindrical portions of the shaft component 1 stored in advance are determined. against the information of positional relationship, the position where the cylindrical portion locates the predetermined distance, i.e., the fourth position apart intervals P ₄ or spacing P ₅ in Figure, a given sequence, i.e. the cylindrical portion d → cylinder It is determined whether the parts are arranged in the order of part e → cylindrical part f or cylindrical part f → cylindrical part e → cylindrical part d (step 105).

If it is confirmed that the adjacent cylindrical portions are arranged at a predetermined interval in a predetermined order, the above-described step is performed.
Returning to step 103, in step 103, the information is compared with the information of the arrangement of the cylindrical parts of the shaft component 1 stored in advance,
It is determined whether or not there is another cylindrical portion to be searched. If there is another cylindrical portion to be searched, the process proceeds to step 104, where the cylindrical portions coaxially adjacent to each other (the cylindrical portion c and the cylindrical portion in FIG. 4). g), and the scanning process of these steps 103, 104, and 105 is repeated. When all the cylindrical portions of the shaft part 1 are found and it is determined in step 103 that there is no other cylindrical portion to be searched, Step 107
Proceed to.

In step 107, the position and orientation of the center of gravity of the shaft component 1 as a whole is calculated by calculating from the arrangement of each cylindrical portion found in the pre-scanning process and the information of each cylindrical portion of the shaft component 1 stored in advance. Then, the obtained result is output (step 108), and the recognition of the shaft component 1 is completed.

On the other hand, if the desired cylindrical portion cannot be found in the step 105, the process returns to the step 101, and a new scan is performed again from the process of searching for a completely different longest cylindrical portion.

If the longest cylindrical portion cannot be found in step 102, it is determined that there is no component (step 106), and the result of no component is output (step 108), and the recognition of the shaft component 1 ends.

Although the shaft part 1 which is the recognition target part in the above embodiment has the cylindrical parts which are the features to be recognized arranged in a straight line, the relative positions are not necessarily arranged in a straight line. It is sufficient that the relationships are arranged so that the relationships can be clearly described. For example, the recognition in the case of the shaft component 4 in FIG. 2 (d) can be performed similarly.

Further, in the above embodiment, the cylindrical portion a, the cylindrical portion b,..., The cylindrical portion g are taken as the local features of the recognition target component, but other items used as the local features include, for example, the recognition target component. Features such as parallel parts, corners, and formed holes, and any other appearance features that can be easily read from the projected screen are adopted as local features for object recognition. You.

As described above, according to the method of the present invention, as data for identifying a recognition target object, at least two portions having a shape characteristic and a relative positional relationship between the portions having the shape characteristic are determined by the object. It is configured to recognize the object to be recognized as a whole by recognizing the position and orientation of the recognition target object by acquiring the data from the image and comparing the data with the corresponding data stored in advance. Even if the object has a complicated shape, by adopting as data the number of geometric feature parts corresponding to the shape and their relative positional relationship as data, matching is performed.
The object can be accurately recognized.

Also, regarding the geometric features and their relative positional relationships,
Since the same recognition algorithm can be applied to recognition of any recognition target object to which a similar description format can be applied, a highly versatile image recognition method can be provided. Further, since it is not necessary to individually develop image recognition software for each object to be recognized, the time and cost required for developing and introducing an image recognition system for an object can be significantly reduced.

Furthermore, since one image recognition method has a wide application range,
It has effects such as being able to easily cope with design changes at the time of production of various types of articles and model changes.

[Brief description of the drawings]

1 to 4 show an embodiment of the method of the present invention. FIG. 1 is a flow chart showing a scanning process of the image recognition method, and FIGS. 2 (a), 2 (b), 2 (c) and 4 (d). FIG. 3 is a front view showing an example of a shaft part, FIG. 3 is an explanatory view showing local features of the shaft part as a part to be recognized, FIG.
The figure is an explanatory view showing a relative positional relationship between portions having local features of the shaft component. 1 ... shaft parts, a, b, c, d, e, f, g ... cylindrical parts.

Continued on the front page (72) Inventor Yoshito Kato 1 Toyota Town, Toyota City Inside Toyota Motor Co., Ltd. (72) Inventor Yuji Watanabe 170 Obishicho, Komatsu City (72) Inventor Shigeyuki Tada Aoji Komatsu City 132-cho, No. 72 (72) Nobuyoshi Yamanaka 132, Aoji-cho, Komatsu-shi, Japan (56) References JP-A-61-46568 (JP, A) JP-A-61-292003 (JP, A) JP-A-60- 218164 (JP, A) JP-A-60-200385 (JP, A) JP-A-60-164877 (JP, A)

Claims (2)

(57) [Claims] 1. An object to be recognized is divided into a plurality of parts, and a shape feature including a light-dark pattern indicating a brightness distribution with respect to a specific part of a screen when each part is imaged. The relative positional relationship including the arrangement order of the respective parts identified by the shape characteristics of is stored in advance, and when performing recognition of an object, the object is captured in an image,
A first portion having a local feature is extracted from an image, and a shape feature of the extracted portion is compared with a stored shape feature to determine which portion of the object is divided. At the same time, at least one other portion having a local feature is extracted, and the shape feature of the extracted portion is compared with the stored shape feature. Identify
Recognizing the object by comparing the relative positional relationship between the specified portion having the first local feature and the portion having the other one local feature with a relative positional relationship stored in advance. A method for recognizing an image of an object, comprising: 2. The image of the object according to claim 1, wherein the arrangement direction of the object to be recognized is calculated based on the position of each part having a local feature extracted from the image and is recognized based on the position of the center of gravity. Recognition method.