JP2019174287A - Object recognition device, method, program, and object removal system - Google Patents

Abstract

To provide an object recognition device, a method, a program, and a rock removal system which can detect a surface of a massive object having a convex polyhedron-like contour and can easily detect the massive object.SOLUTION: The object recognition device 100 for recognizing a massive object includes a three-dimensional data generating unit 111 which generates three-dimensional data expressing positional information for photography based on a plurality of pictures photographed at a different angle; a surface extraction unit 113 which extracts three-dimensional data of a predetermined division as the surface of the massive object (for example, a rock 300) when judged in the distance from a datum plane of three-dimensional data within a predetermined limit about a predetermined division; and a recognition unit 115 which recognizes the massive object using the extracted surface. The surface of a massive object which has a convex polyhedron-like contour can be extracted by setting a datum plane in this way, and the object can be recognized easily. As a result, a large mass rock blocked in a crushing process of the rock 300, for example in a mesh, can be recognized efficiently.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an object recognition apparatus, method, program, and object removal system for recognizing a massive object from a captured image.

  At the mining site of the mine, the raw material rock is crushed to an appropriate size and transported. In this step, since the crushed rock is passed through a sieve (grid river), it is necessary to detect and crush or remove large rocks that have accumulated on the sieve without passing through the sieve. For the crushing work, first, large rocks are identified. As a result, it is possible to reduce the burden of removing large rocks in the mining chamber.

  Conventionally, a method for recognizing a rock using an image is known. The system described in Patent Document 1 discriminates the flow state of a rock from an image captured by a camera and a sound signal detected by a microphone, and the large amount left on the sieve when the rock crushed at the mine is put into a crusher. Although a method for recognizing massive rocks and automatically removing large massive rocks has been proposed, details of the method for recognizing massive rocks are not disclosed.

JP 2009-235781 A

  In addition, even if large rocks are automatically recognized by monitoring the flow state as described above, if the accuracy is not sufficient, it is necessary to confirm the worker. In that case, the worker himself / herself must confirm the large block rock and perform the removal operation, and the situation where the worker needs to be arranged does not change. In addition to such an example, it is difficult to detect a massive object having a convex polyhedron-like outer shape such as a rock with high accuracy.

  The present invention has been made in view of such circumstances, and by setting a reference surface, it is possible to detect the surface of a massive object having a convex polyhedron-like outer shape, and to easily detect the massive object An object is to provide an apparatus, a method and a program, and an object removal system.

  (1) In order to achieve the above object, an object recognition apparatus according to the present invention is an object recognition apparatus for recognizing a block-shaped object, and the position information of a shooting object based on a plurality of images shot at different angles. A three-dimensional data generating unit for generating three-dimensional data representing the three-dimensional data of the predetermined section when it is determined that the distance from the reference plane of the three-dimensional data is within a predetermined range for the predetermined section And a recognition unit for recognizing a blocky object using the extracted surface. Thereby, the surface of the block-shaped object which has a convex polyhedron-like external shape can be extracted, and an object can be recognized easily. As a result, for example, large rocks clogged with a sieve in a rock crushing process can be recognized efficiently.

  (2) Further, the object recognition apparatus of the present invention is characterized in that the surface extraction unit uses a plurality of planes having different angular arrangements as the reference plane. Thereby, the surface of the block-shaped object which has a convex polyhedron-like external shape can be extracted easily.

  (3) Moreover, the object recognition apparatus of this invention is further equipped with the representative point calculation part which calculates the representative point of the said three-dimensional data for every said predetermined division about the said extracted surface, The said recognition part is the said extraction. It is characterized by recognizing a block-like object using the representative point as a surface. Thereby, since the surface can be handled with representative points instead of a large number of three-dimensional data for each predetermined section, the amount of data to be processed can be reduced and the processing efficiency can be increased.

  (4) Moreover, the object recognition apparatus of the present invention is characterized in that the representative point calculation unit calculates an average position of the three-dimensional data for each of the predetermined sections as the representative point. By setting the average position, it is possible to appropriately represent the surface of the block-shaped target in a predetermined section.

  (5) Moreover, the object recognition apparatus of this invention is further equipped with the outline extraction part which extracts the outline of a block-shaped object using the brightness | luminance difference in the said three-dimensional data, The said recognition part is the said extracted surface It is also characterized by recognizing a massive object using the contour. As a result, not only the surface extracted from the position information but also the contour extracted from the luminance information can be used together, so that the position, size, shape, etc. of the massive object can be recognized with high accuracy.

  (6) Further, in the object recognition device of the present invention, the contour extraction unit may detect a relative difference in luminance between conditions or positions where the luminance of the three-dimensional data in the predetermined section is smaller than an absolute reference value. The predetermined section is extracted as a contour when at least one of the conditions larger than the reference value is satisfied. Thereby, the difference between the surface of the massive object and the gap can be captured, and the outline of the massive object can be extracted.

  (7) Moreover, the object removal system of the present invention includes a camera that captures the plurality of images, the object recognition device according to any one of (1) to (6), and the recognized massive object. Are determined at the same position for a predetermined time, the determination unit that determines the block target that remains in the same position as a removal target, and a crusher that crushes or moves the block target to be removed It is characterized by that. Thereby, the large block rock clogged with the sieve in the rock crushing process can be removed. As a result, the rock splitting work can be automated, reducing the burden on the operator.

  (8) Further, the object recognition method of the present invention is a method for recognizing a block-like object, and generates three-dimensional data representing position information of a shooting target based on a plurality of images shot at different angles. Extracting the three-dimensional data of the predetermined section as the surface of a block object when it is determined that the distance from the reference plane of the three-dimensional data is within a predetermined range for the predetermined section; Recognizing a massive object using the extracted surface. Thereby, the surface of the block-shaped object which has a convex polyhedron-like external shape can be extracted by setting a reference plane, and an object can be recognized easily.

  (9) Further, the program of the present invention is a program for recognizing a block-shaped object, and a process for generating three-dimensional data representing position information of a photographing object based on a plurality of images photographed at different angles. A process of extracting the three-dimensional data of the predetermined section as a lump target surface when it is determined that the distance from the reference plane of the three-dimensional data is within a predetermined range for the predetermined section; And a process of recognizing a block-like object using the formed surface. Thereby, the surface of the block-shaped object which has a convex polyhedron-like external shape can be extracted by setting a reference plane, and an object can be recognized easily.

  According to the present invention, the surface of a massive object having a convex polyhedron-like outer shape can be extracted, and the position, size, and shape of the massive object can be easily and accurately recognized.

It is a figure which shows the functional structure of the object removal system of this invention. It is a flowchart which shows operation | movement of the object removal system of this invention. It is a flowchart which shows operation | movement of the object recognition apparatus of this invention. It is a schematic diagram which shows each area | region of the representative point extraction of three-dimensional data. It is a schematic diagram which shows the process of representative point calculation. (A), (b) is a schematic diagram which shows the relative positional relationship of a camera and a representative point, respectively, and a schematic diagram which shows the surface extraction process using a reference plane. It is a figure which shows the image data after the surface extraction by a single reference plane. (A), (b) is a schematic diagram showing processing when a reference plane perpendicular to the optical axis is used and when a reference plane tilted from a plane perpendicular to the optical axis is used. It is a figure which shows the image data after the surface extraction by a some reference surface. It is a figure which shows the image after a process at the time of performing contour extraction in addition to surface extraction.

  Hereinafter, embodiments of the present invention will be described. Note that the described value in the description is an example, and a numerical value other than the described value may be taken depending on the size and shape of the object to be recognized, the performance of the camera and the processing device, and the like.

[First Embodiment]
(Configuration of target removal system)
The object removal system 100 is a system that automatically recognizes whether a block-like object is a removal object, and can remove the object when the object is a removal object. For example, it is possible to automatically identify a large block rock in the crushed rock deposit and remove the large block rock at a mine mining site or the like. In addition, the rocks mined from the mine are used not only as a raw material for cement but also as a raw material for concrete aggregate and inorganic material.

  FIG. 1 is a diagram illustrating an example of a functional configuration of the object removal system 100. The object removal system 100 includes a camera 105, a computer 110 (object recognition apparatus), an operation unit 140, a display unit 150, and a subdivision machine 200, and represents a surface of an imaging region derived from an image captured by the camera 105 3 The rock 300 identified from the dimensional data (original image of stereo vision) can be removed by the subdivision machine 200. Since the data represents the surface, the back side of the rock 300 and the concealed rock 300 cannot be seen, but there is no problem in the recognition of the rock 300.

  The camera 105 captures a plurality of images at different angles. The camera 105 preferably constitutes a stereo vision system in which a plurality of cameras 105 are provided at predetermined positions from the viewpoint of facilitating control and processing. For example, in order to avoid blind spots, one camera 105 can be fixed on the left and right sides, and the right half and the left half can be assigned to each. In addition, you may image | photograph the image of several angles so that the one camera 105 may be moved and a blind spot may not be made. The frame rate of the camera 105 is not particularly limited, but it is preferable that the speed necessary for determination at least once every several seconds is satisfied. It can be set according to the data transfer rate, and can be set to 20 fps, for example.

(Configuration of object recognition device)
The computer 110 includes a memory and a CPU, and for example, a PC can be used. The computer 110 operates by executing a program stored in the memory, and also functions as an object recognition device that recognizes a massive object.

  The lump-like object refers to an object having a convex polyhedron-like outer shape, and in this embodiment refers to the rock 300, but is not limited to this, and includes lumps of various materials such as metals and waste materials. Therefore, the object recognition apparatus can be applied to other than the object removal system 100 by itself. Note that a series of discriminant calculations may take one hundred milliseconds, but the computer 110 only needs to be capable of calculating at a frequency of once every few seconds.

  The computer 110 includes a three-dimensional data generation unit 111, a surface extraction unit 112, a representative point calculation unit 113, a contour extraction unit 114, a recognition unit 115, a determination unit 120, and a subdivision controller control unit 130 as functional configurations. . The computer 110 receives an image captured by the camera 105.

  The three-dimensional data generation unit 111 generates three-dimensional data representing the surface of the shooting area based on a plurality of images shot at different angles. In the generation of three-dimensional data, the same point is searched from two images with different angles, the parallax is obtained, and the three-dimensional position of the same point is calculated based on the parallax. For example, by using a figure that can be clearly distinguished from a pattern existing in the shooting environment as a marker, it becomes easy to search for the same point, and an accurate three-dimensional position can be obtained due to a reduction in error associated therewith. As markers, black and white intersecting graphic signs can be placed in the imaging area.

  For example, a stereo vision system using two CCD cameras can obtain three-dimensional data as a point group composed of several hundred thousand points. From the image, three-dimensional coordinates and luminance are obtained as data of each point. The three-dimensional coordinates are, for example, x, y, when the optical axis of the camera 105 is the z axis, the axis perpendicular to the z axis and parallel to the vertical plane is the y axis, and the axis perpendicular to the z axis and the y axis is the x axis. The z coordinate. In the present embodiment, the computer 110 side generates three-dimensional data. However, the camera 105 side may generate three-dimensional data and transmit it to the computer 110.

  The surface extraction unit 112 determines that the three-dimensional data generated for a predetermined section has a distance from a reference plane within a predetermined range (that is, a difference in distance from the reference plane at each point is within a predetermined range). When determined, the surface of the block target is extracted from the data used for the determination. For example, the degree of flatness is determined from a three-dimensional position with respect to a partition of 5 × 5 = 25 points, and if a certain criterion is satisfied, the partition is determined to be a plane. The preferred size of the predetermined section varies depending on the scale of the equipment, the size, shape, composition, crushing method, etc. of the target object. For example, in the rock splitting work in the mine where the experiment was performed, X5 = 25 points corresponds to 20 mm to 100 mm square of the rock surface to be recognized.

  The reference surface has, for example, an inclination in the range of −20 ° to 40 ° around the x axis with respect to the mounting surface of the sieve 350, and an inclination in the range of −10 ° to 10 ° around the y axis. Can have. Thereby, a reference plane that is nearly parallel to the surface of the massive object can be set, and the surface of the massive object can be reliably extracted. Moreover, the predetermined range of the distance from the reference plane can be, for example, less than 40 mm. The predetermined range of the inclination of the reference plane and the distance from the reference plane varies depending on the scale of the equipment, the size of the openings of the sieve 350, the type and composition of the mass to be crushed, the crushing method, etc. It may take a range other than numerical values. For reference, exemplary numerical values for the size of the predetermined section, the inclination of the reference plane, and the distance from the reference plane were tested in a rock crushing facility with a sieve opening of 800 mm in a limestone mine. Is the case.

  The surface extraction unit 112 preferably uses a plurality of planes having different angular arrangements as the reference plane. By using a plurality of reference planes having different angular arrangements, it is possible to accurately extract the inclined outer surface of the massive object.

  The representative point calculation unit 113 calculates a representative point of the generated three-dimensional data for each predetermined section. As a result, the surface can be handled at a representative point for each predetermined section, so that the amount of data to be processed can be reduced and the processing efficiency can be increased. The representative point calculation unit 113 preferably calculates the average position of the three-dimensional data generated for each predetermined section as the representative point. Thereby, the surface of the block-shaped object of a predetermined division can be appropriately represented by setting it as an average position. Note that the median value and the mode value may be used instead of the average value.

  The contour extracting unit 114 extracts the contour of the block target using the luminance difference in the three-dimensional data. The contour extraction unit 114 is configured such that the luminance of the three-dimensional data in a predetermined section is smaller than the absolute reference value when the interval between the maximum value and the minimum value of the luminance of the three-dimensional data is expressed by 256 gradations or between positions. It is preferable to extract a predetermined section (representative point) as an outline when at least one of the conditions in which the difference in luminance is larger than the relative reference value is satisfied. Thereby, the difference between the surface of the massive object and the gap can be captured, and the outline of the massive object can be extracted.

  The recognition unit 115 recognizes a block-like object using the extracted surface. Specifically, a continuous part is extracted from the processed image by a labeling process (a technique for extracting a continuous area by image processing), and is determined as a block target. Thereby, by setting the reference plane, the surface of a massive object having a convex polyhedron-like outer shape can be extracted, and the massive object can be easily detected. As a result, for example, when applied to a rock crushing process, large rocks clogged in the sieve 350 can be recognized efficiently. It is preferable that the recognition unit 115 recognizes the block target using not only the surface of the extracted block target but also the contour. By superimposing not only the surface position information but also the luminance difference, the edges become clear, and a massive object can be reliably recognized.

  The determination unit 120 determines, as the removal target, the massive object that remains at the same position when the recognized massive object remains at the same position for a predetermined time. In the step of sieving the mined rock, the rock staying at the same position may be regarded as a large block rock that does not pass through the sieve 350.

  The small breaker control unit 130 controls an operation for the small breaker 200 to remove large rocks. That is, with the target recognized as a large block rock staying on the sieve 350 by the determination unit 120, the tip of the splitting machine 200 is moved to a position for crushing the rock, and the tip crushes the large block rock. Or it can be automatically operated to move.

  The operation unit 140 is, for example, a keyboard or a mouse, and accepts an operation by an operator. The operator can adjust, stop, restart, etc. a series of operation parameters by operating the operation unit 140. The operation of the subdivision machine 200 is preferably automatic, but the operation by the operation unit 140 may be possible. The display unit 150 displays an image captured by the camera 105, an image after processing (an image from which a surface and an outline are extracted), and a user interface necessary for an operation.

  Although not specified, the subdivision machine 200 has a serial link mechanism composed of, for example, a boom and an arm, and crushes or moves a block target to be removed by a breaker attached to the tip. A sensor that detects a rotation angle may be attached to each joint of the serial link mechanism, and a hydraulic sensor that detects a generated force or a reaction force may be attached to each hydraulic actuator. The serial link mechanism may be operated by a hydraulic actuator based on signals detected by these sensors and control data output from the splitter control unit 130.

  In this way, large rocks clogged in the sieve 350 in the rock crushing process can be removed. As a result, for example, it is possible to automate the work of dividing large rocks in the counter-compartment room of the limestone mine, which is a process of sieving the mined rock, and the burden on the operator can be reduced.

(Operation of the target removal system)
An example of the operation of the object removal system 100 configured as described above will be described. FIG. 2 is a flowchart showing the operation of the object removal system 100. First, images to be photographed from a plurality of angles photographed by the camera 105 are received by the computer 110 (step S1). Three-dimensional data representing the surface is generated from the received image (step S2). A rock is recognized using the generated three-dimensional data (step S3). Details of the rock recognition process will be described later as the operation of the object recognition apparatus.

  It is determined whether or not there is a rock that continues to stay at the position of the sieve among the recognized rocks (step S4). When it is determined that there is a staying rock, the rock is recognized as a large block rock, and the small block 200 is removed by controlling the small splitting machine 200 (step S5), and the process ends. If it is determined that there is no staying rock, the process ends.

(Operation of the object recognition device)
Next, an example of the operation of the object recognition device will be described. FIG. 3 is a flowchart showing the operation of the object recognition apparatus. First, it is determined whether each predetermined section (for example, one section with 25 pixels) is a plane using three-dimensional data (three-dimensional position information). The determination criteria are all positions in a predetermined section, and when it is determined that the distance from a certain reference surface is within a predetermined range, it is determined to be a flat surface and is extracted as the surface of a block target (step T1). ). For example, if the maximum value (maximum depth distance) in the height direction of 25 points per section with respect to a certain reference plane is less than 40 mm, it can be determined as a plane, and if it is 40 mm or more, it can be determined as an edge (contour).

  For one section, the above calculation and determination are performed for a plurality of reference planes at different angles. For example, if all the reference planes are 60 planes, it is determined that any one of them is a plane. The section is determined to be a plane. For the section determined to be a plane, the representative point of the section is calculated (step T2). Then, the sections determined to be planes are connected and handled as a surface candidate for a block-like target (step T3). That is, the surfaces extracted for a plurality of sections are integrated. Thereby, the exposed surface of the massive object can be specified. In the calculation after step T3, the processing efficiency can be improved by representing the surface with a representative point and processing.

  On the other hand, contour extraction using luminance information of three-dimensional data is also performed. For example, when the maximum value and the minimum value of the luminance of the three-dimensional data are expressed by 256 gradations, whether the maximum value of the luminance in a predetermined section is 35 or less, or the difference between the maximum value and the minimum value is 50 It is determined whether or not the condition is higher than the gradation, whether the maximum value is 128 gradations or less, and the difference between the maximum value and the minimum value is 30 gradations or more. A predetermined section can be extracted as a contour (step T4).

  Next, the obtained surface data and contour data are superimposed (step T5). Then, a massive object is recognized using the superimposed data (step T6). In the data superimposed in this way, the surface and the contour are clear, so that it is easy to recognize the massive object. In the above example, the surface and the contour are integrated to recognize the massive object, but the massive object may be recognized only by the surface extraction process.

(Calculation of representative points)
In order to increase the speed of the point cloud processing, it is preferable to average the point cloud corresponding to the pixel section of the original image and use the value as the representative value of the section rather than using the pixel of the original image as it is. The three-dimensional position of each point includes a minute error due to the influence of noise or the like, but this error can be removed by this averaging.

  FIG. 4 is a schematic diagram showing each region of representative point extraction of three-dimensional data. As shown in FIG. 4, three-dimensional data P0 representing the surface of an object in the area is generated for the imaging area F0. For example, predetermined sections Ryx, Ry + 1x, Ryx + 1, and Ry + 1x + 1 are set for the three-dimensional data P0, and representative points of the respective sections are calculated. For example, it is possible to collect hundreds of thousands of origin points into about 10,000 representative points.

  FIG. 5 is a schematic diagram illustrating processing for calculating representative points. The origin groups Pyx, Py + 1x, Pyx + 1, and Py + 1x + 1 shown in FIG. 5 are origin groups that indicate three-dimensional data in each predetermined section. Then, it is possible to calculate representative points Ayx, Ay + 1x, Ayx + 1, and Ay + 1x + 1 indicating the average with respect to the origin group in the predetermined section. Averaging can be performed by calculating an average value of each axis component and an average value of luminance.

(Surface extraction by a single reference plane)
FIGS. 6A and 6B are a schematic diagram showing a relative positional relationship between the camera 105 and the representative point, and a schematic diagram showing surface extraction processing using a reference plane, respectively. As shown in FIG. 6A, when the rock 300 is placed on the surface (mounting surface) of the sieve 350 inclined by θ from the horizontal plane, each origin group Pyx is virtually provided near the rock surface. . On the other hand, the optical axis z of the camera 105 is represented. In addition, although θ is 12 °, for example, it is a numerical value corresponding to the facility.

  Further, as shown in FIG. 6B, a plane Q0 perpendicular to the optical axis z of the camera 105 is set as a reference plane, and the farthest depth plane Qmax and the latest depth plane Qmin are provided. If the origin group Pyx enters the range of these planes, it is determined that the origin group Pyx is on a continuous plane, and is extracted as the surface of a massive object.

  FIG. 7 is a diagram showing image data after surface extraction by a single reference plane. In the example shown in FIG. 7, the representative points of the three-dimensional data are displayed in a distinction between bright and dark colors in the shooting area F0. A position showing a bright color represents a point extracted as a surface, and a position showing a dark color represents a point not extracted as a surface. As shown in FIG. 7, the shape of the rock generally appears by this process.

(Surface extraction using multiple reference surfaces)
Since crushed rocks by blasting, especially large lumps, often exhibit a polyhedron shape composed of crushed surfaces, it is preferable to extract the “surface” from the three-dimensional shape of the rocks for discrimination of large lumps. In order to extract a surface from a three-dimensional shape, there is a method in which a region whose distance from a reference surface is equal to or less than a threshold = a surface. Since extraction using one reference plane extracts only a plane parallel to the reference plane, it is preferable to perform extraction using a plurality of reference planes. Thereby, a large block having a convex polyhedron-like outer shape can be detected.

  FIGS. 8A and 8B are schematic diagrams showing processing when a reference plane perpendicular to the optical axis z is used and when a reference plane tilted from a plane perpendicular to the optical axis z is used. When a plane Q0 perpendicular to the optical axis z is used as a reference plane as shown in FIG. 8A with respect to a predetermined rock surface 301 inclined with respect to the optical axis z of the camera, the surface 301 When the distance between the nearest depth plane Qmin and the farthest depth plane Qmax of the representative point Ayx becomes large, the plane may not be recognized as a plane.

  On the other hand, as shown in FIG. 8B, when the surface W0 parallel to the mounting surface inclined with respect to the surface Q0 is used as the reference surface, the farthest depth surface Wmin of the representative point Ayx of the surface 301 and the farthest distance. The distance between the depth plane Wmax is reduced and the probability of being recognized as a plane increases. Therefore, the surface 301 can be extracted by using a plurality of reference surfaces and setting an appropriate distance from the reference surface. The surface can be extracted by providing a plurality of reference planes with a slight difference in inclination with a plane perpendicular to the optical axis z as the center reference plane. For example, when taking axes x and y perpendicular to the optical axis z, twelve angles from −10 ° to 45 ° with x-axis rotation and five angles from −10 ° to 10 ° with y-axis rotation The surface can be extracted by determining with 60 kinds of reference planes by combining (each in increments of 5 °). 12 kinds of reference planes by combining 4 steps of angle from -20 ° to 40 ° (in increments of 20 °) with x-axis rotation and 3 steps of angle from -10 ° to 10 ° (in steps of 10 °) with y-axis rotation It may be determined and the surface may be extracted.

  FIG. 9 is a diagram illustrating an image after surface extraction using a plurality of reference surfaces. In the example shown in FIG. 9, in the shooting area F <b> 0, a position indicating a bright color is extracted as a surface, and a position indicating a dark color is not extracted as a surface (that is, a point treated as an outline). Represents. As shown in FIG. 9, by extracting a surface with a plurality of reference planes, it is possible to capture a wider rock surface than when extracting a surface with a single reference plane.

(Outline extraction by luminance)
In the image of fractured rock accumulation, a clear outline is often observed around the rock. Since the brightness of the image is reduced at the contour portion of the rock in the image, the large block rock can be detected by setting the region having the brightness equal to or lower than the threshold value as the contour portion.

  However, when trying to discriminate only by the contour, it may be recognized as another object due to the luminance change due to its pattern on the block object, or it may not be possible to separate the overlapping rocks of the same brightness, It is preferable to superimpose with extraction.

(Object recognition by surface and brightness)
FIG. 10 is a diagram illustrating an image after processing when contour extraction is performed in addition to surface extraction. In the example illustrated in FIG. 10, in the imaging region F <b> 0, a position indicating a bright color represents a point extracted as a surface, and a position indicating a dark color represents a point extracted as a surface or a point extracted as an outline. ing. As shown in FIG. 10, by superimposing the contour shape on the extracted surface, the shape of the rock clearly appears compared to the case of only the surface extraction. In this way, by extracting the surface based on the three-dimensional shape and superimposing the contour portion based on the luminance of the image, mutual defects are compensated, and it is possible to detect a large rock with high probability.

DESCRIPTION OF SYMBOLS 100 Object removal system 105 Camera 110 Computer 111 Three-dimensional data generation part 112 Surface extraction part 113 Representative point calculation part 114 Outline extraction part 115 Recognition part 120 Determination part 130 Small machine control part 140 Operation part 150 Display part 200 Small machine 300 Rock 301 Surface 350 Sieve F0 Imaging region P0 Three-dimensional data Pyx to Py + 1x + 1 Origin group Ayx to Ay + 1x + 1 Representative point Ryx to Ry + 1x + 1 Section z Optical axis Q0 Surface perpendicular to optical axis Qmax Farmost depth surface Qmin To optical axis Vertical nearest depth plane W0 Surface tilted from the plane perpendicular to the optical axis Wmax Farmost depth plane tilted from the plane perpendicular to the optical axis Wmin Recent depth plane tilted from the plane perpendicular to the optical axis

Claims (9)

An object recognition device for recognizing a massive object,
A three-dimensional data generation unit that generates three-dimensional data representing position information of a photographing target based on a plurality of images photographed at different angles;
When it is determined that the distance from the reference plane of the three-dimensional data is within a predetermined range for a predetermined section, a surface extraction unit that extracts the three-dimensional data of the predetermined section as a lump target surface;
An object recognition apparatus comprising: a recognition unit that recognizes a massive object using the extracted surface.   The object recognition apparatus according to claim 1, wherein the surface extraction unit uses a plurality of planes having different angular arrangements as the reference plane. A representative point calculation unit for calculating a representative point of the three-dimensional data for each of the predetermined sections for the extracted surface;
The object recognition apparatus according to claim 1, wherein the recognition unit recognizes a massive object using the representative point as the extracted surface.   The object recognition apparatus according to claim 3, wherein the representative point calculation unit calculates an average position of the three-dimensional data for each of the predetermined sections as the representative point. A contour extracting unit that extracts a contour of a block-like object using a difference in luminance in the three-dimensional data;
The object recognition apparatus according to any one of claims 1 to 4, wherein the recognition unit recognizes a massive object using the extracted surface and contour.   The contour extraction unit satisfies the condition in which at least one of a condition in which the luminance of the three-dimensional data in the predetermined section is smaller than an absolute reference value or a condition in which a luminance difference between positions is larger than a relative reference value is satisfied. 6. The object recognition apparatus according to claim 5, wherein a predetermined section is extracted as an outline. A camera that captures the plurality of images;
The object recognition device according to any one of claims 1 to 6,
When the recognized massive object remains at the same position for a predetermined time, a determination unit that determines the massive object remaining at the same position as a removal target; and
An object removal system comprising: a crusher that crushes or moves the massive object to be removed. A method for recognizing a massive object,
Generating three-dimensional data representing position information of a photographing object based on a plurality of images photographed at different angles;
When it is determined that the distance from the reference plane of the three-dimensional data for a predetermined section is within a predetermined range, the step of extracting the three-dimensional data of the predetermined section as a surface of a block target;
Recognizing a massive object using the extracted surface. A program for recognizing massive objects,
A process of generating three-dimensional data representing position information of a photographing object based on a plurality of images photographed at different angles;
When it is determined that the distance from the reference plane of the three-dimensional data for a predetermined section is within a predetermined range, a process of extracting the three-dimensional data of the predetermined section as a surface of a massive target;
A program for causing a computer to execute a process of recognizing a massive object using the extracted surface.