BR102014009460A2 - device for dimensional measurement of parts; parts analysis and inspection system; automated dimensional control method; and uses - Google Patents

Abstract

device for dimensional measurement of parts; parts analysis and inspection system; automated dimensional control method; and uses the present invention is a device for dimensional measurement of parts; parts analysis and inspection system; automated dimensional control method; I just. The present invention has application in the control of the geometric characteristics of parts that demand high precision in their measurements.

Description

DEVICE FOR DIMENSIONAL MEASUREMENT OF PARTS; PARTS ANALYSIS AND INSPECTION SYSTEM; DIMENSIONAL CONTROL METHOD

AUTOMATED; AND USES

FIELD OF INVENTION

The present invention is a device for dimensional measurement of parts; parts analysis and inspection system; automated dimensional control method; I just.

[002] The present invention has application in the control of the geometric characteristics of parts that demand high precision in their measurements.

TECHNICAL STATE

[003] The production of automotive parts requires a great effort in obtaining parts with high precision dimensions, geometry and alignment. In this context, hot-forged cylindrical automotive parts, such as axles, bolts, and crankshafts, are those that require the most time to be measured and gauged and present the greatest risk to the physical integrity of the worker while performing this service.

Much research has been carried out to develop computer systems that can assist human operators in risky tasks that can be automated. In this scenario, computer vision systems have contributed significantly to various phases of the production cycle, such as object measurement, assembly, three-dimensional reconstruction and many other repetitive tasks. Computerized visual inspection allows automatic and exhaustive examination of products, which is not the case with human inspection, which is usually based on statistical approximations.

[005] With increasing globalization and competition industry has invested in process automation and many mathematical and computational tools have been proposed. Image processing and analysis create a science that lets you manipulate, analyze, and modify digital images. Thus, vision systems include a set of methods that support the development of reliable products or efficient processes that can be employed in practical situations in a manufacturing environment.

[006] For the development of vision systems used in industry, C.CD (Charge Coupled Device) cameras are commonly used as vision sensors to obtain the position and orientation of objects in a scene.

Vision systems can operate with one or more cameras, which can be mobile or fixed. Although it is possible to get an idea of the 3D world using a single image, the most robust procedures can be obtained from two images, which can be obtained from the same camera that changes position.

[008] The goal of a vision system is to create a real-world model from images. Due to the particularities and set of constraints of each type of application, technologies are used to solve a specific problem, adapting the mathematical toolset to solve each type of problem. It is not possible to have a generic vision system, but the application of various techniques in a staggered way allows to solve the most complex problems.

[009] Determining 3D data (three dimensions) from 2D images (two dimensions) is a very important area in vision systems and has been studied extensively. A vision system retrieves information about the scene from its 2D projections. Because images are 2D projections of the real world in 3D, any information about the scene is not obtained directly, or by triangulation, however, there are some well-defined problems related to obtaining 3D information from a scene.

[010] The first problem is called matching location, which is how to establish a correlation between two images of the same object taken from different points of view. A point projected on the first image must match a point on the second image, which is the projection of the same point on the actual scene. Numerous methods for the automatic determination of this correspondence have been proposed. As known to those skilled in the art, certain algebraic and geometric procedures may correlate the points of the images to be studied.

[011] The second problem is related to three-dimensional reconstruction, more specifically camera calibration. Camera calibration is required for any visual inspection that seeks to determine the position of objects or the location of the camera in the environment. Usually the only information available is the images of the objects. The basic idea of calibration is to establish a relationship between the image points and the spatial coordinates given in metric units. This relationship depends on the setting of the camera's optical parameters as well as the estimation of the relative position between the cameras. The relative position is given by a rotation matrix and a translation vector, which form a matrix expression called the essential matrix.

[012] Determining the appearance of the object in terms of volume and position is described in several studies dealing with the extraction of 3D information from images. When it is necessary to know details of the faces of the object, it is common to use point cloud to reconstruct certain surface. As the computational effort is high, the reconstruction optimization was proposed with the objective of minimizing the computational effort, mainly by the disparity map techniques and the Delaunay triangulation to create triangular facets.

[013] There are currently dozens of manufacturers offering solutions for dimensional control of parts using cameras.

[014] However, the main industrial vision systems are applied in the areas of electro-electronic component inspection, printed inspection in the pharmaceutical and food industries and dimensional control in the automotive industry, and the solutions offered are tailored to each client's application.

[015] In patent literature, we see that several technologies related to surface measurements have already been developed. US7869061 relates to a technology for measuring surface distortion using a camera on the surface to be measured, and a mirror positioned behind the surface.

[016] Since JP2004150810 is known a form of measurement that promotes the combination between laser and image for the reconstruction of three-dimensional bodies using the technique of triangulation and sensory fusion, which compares the images obtained with a CAD model of the object analyzed.

[017] DE102010015353 relates to a technology using a CCD sensor that compares the images obtained with a CAD model of the object analyzed.

[018] CN102538672 refers to a technology that uses a CMOS sensor which does not allow object reconstruction and uses a technique of image binarization.

[019] As can be seen, the use of cameras for dimensional control in automotive components is gaining importance in the current context as it has great advantage over conventional metering systems, its main disadvantage being the development of software and hardware for each specific application, and there is no system developed for dimensional control of components of different geometries. BRIEF DESCRIPTION OF THE INVENTION The present invention is a device for dimensionally measuring parts; a computer vision system for dimensional inspection of parts on the production line; the measurement method; and the use of the device and the system to control the geometrical characteristics of the parts that demand high precision in their measurements; Because it is remotely operated, it reduces operator safety problems.

Accordingly, an object device for dimensional measurement of parts is object of the present invention; does not require the aid of measuring instruments or measuring machines; comprises a set of cameras for detecting part images for further computational analysis of the images; This therefore reduces safety problems for the operator who will not be at risk from facing a part with a high temperature.

Said device (1) comprises a base (2) for holding the part; by the image acquisition means (3) for dimensional inspection of the part on the production line, wherein: —the base (2) comprises a table (2.1); a cradle (2.2) and a shock absorber (2.3); wherein the table (2.1) serves as support for the imaging system (3) and for the part to be inspected; the cradle (2.2) is used for laying the piece and keeping it immobilized; and the damper (2.3) serves to mitigate vibration. - the image capture means (3) comprises a set of 4 to 8 cameras (3.1) which have a translation and rotation with each other; and a set of 4 to 8 lights (3.2); that enable the piece to be fully visualized by having adjacent images with overlapping points between them. - the image capture means (3) comprise a set of 6 cameras (3.1); and the set of 6 lights (3.2).

A second object of the present invention is a part analysis and inspection system comprising a device for dimensional measurement of parts; means of integration and means of security.

Said system (4) comprises a device (1) for dimensional measurement of parts; means of integration (MSI); and safety means (IS) that integrated with the factory production control system enable data collection and processing for dimensional, geometric analysis and inspection and alignment of parts, as follows: - the device (1) ) for dimensional measurement of parts be as described above. - the integration means (MSI) promotes communication between a production line robot (1R1), the device (1), the security means (IS), and a power supply (Al) being comprised of a circuit board. data acquisition (5.1); and a computer (5.2) connected to a database (5.2). - the data acquisition board (5.1) is an NI USB card; and an Ethenet card connecting the card (5.1) to the computer (5.2). - the means for integrating (MSI) with the device (1) takes place via the control board (Cl) which connects to interfaces 1/0 which switch on and off the imaging means (3) comprised of the set of 4. 8 cameras (3.1) and the set of 4 to 8 lights (3.2).

[025] A third object of the invention is a method of automated dimensional control through computational and geometric imaging of parts on the production line; which is remotely operated, reducing safety problems for the operator who will not be at risk from facing a high temperature part.

[026] This method comprises the following main steps: (a) initialization; (b) image acquisition and processing; (c) Construction of a 3D image of the part; (d) analysis of the part; (e) removal of the analyzed part; and, (f) Rest.

[027] Where: - The initialization step comprises the verification of the safety system items that are monitored by the safety relay, as well as setting all the necessary variables for the system startup. - an assembly line robot picks up the automotive part and places it on the cradle (2.2) of the table (2.1); and then returns to a rest position where it will wait until the process is complete; The system then clears the variables required for acquisition, processing and analysis, and sends the signal to the imaging means (3) to initiate step (b). - in step (b) the image capture means (3) are calibrated and then capture between 12 to 20 images per camera from the set of 4 to 8 cameras (3.1) so that the part is fully viewed and which images adjacent areas have overlapping points with each other. - the set of 4 to 8 lights (3.2) emitting a spotlight corrects for problems with incident lighting on the part by enabling proper image capture; and the cameras acquire the images and send them to the computer. - the processing occurs after the images are loaded on the computer to obtain homologous points of the image pairs and, by the execution of a calibration algorithm and processing software. - in step (c) a 3D piece image is constructed using state of the art software. - at the analysis stage, the 3D image be inspected and the data relating to the accuracy of the dimensions of the part as a whole as well as from each of its sections are obtained; the data is compared to the standard part data stored in the system. - in step (e) the robot removes from the cradle (2.2) the analyzed part and carries it to an air carrier according to the result of the analysis.

[028] The last object of the invention is the use of an image analysis system for dimensionally measuring parts in the quality control of a production line.

BRIEF DESCRIPTION OF DRAWINGS

[029] In order to obtain a full view of the objects of the present invention, it is necessary to read this document and analyze the accompanying drawings to which reference is made as follows.

[030] Figure 1 - is the perspective view of the device for dimensional measurement with an automotive part represented.

[031] FIGURE 2 - is the measurement system layout model.

[032] FIGURE 3 - Representation of a camera's coordinate system.

[033] FIGURE 4 - Image of a part from camera image plane 1.

[034] FIGURE 5 - is an example of a homologous point location vector.

[035] FIGURE 6 - is an example of the location vector f (x, y) of homologous points in pixels.

[036] FIGURE 7 - illustrates the determination of alignment cylinders.

[037] FIGURE 8 - Graph with ellipse parameters.

[038] FIGURE 9 - Frequency histogram.

[039] Figure 10 - Analysis and inspection system architecture.

DETAILED DESCRIPTION OF THE INVENTION

[040] The present invention deals with a device (1) for dimensional measurement of parts comprising a base (2) for supporting the part; image acquisition means (3) for dimensional inspection of the part on the production line.

[041] For purposes of this invention, hot forged automotive parts are cylindrical parts, for example, pilot shaft, and crankshaft. Preferably, the automotive part is a crankshaft.

[042] This object of the invention is best illustrated by Figure 1, where it can be seen that the base (2) comprises a table (2.1); a cradle (2.2) and a shock absorber (2.3). The table (2.1) serves as support for the image acquisition system (3) and for the part to be inspected; The cradle (2.2) is used for laying the part and keeping it immobilized, thus allowing the dimensional analysis to be performed with greater precision.

[043] The damper (2.3) serves to mitigate vibration; comprises a damper assembly fitted to each of the table supports (2.1.1) (2.1); shock absorbers are state of the art, and may be hydraulic, spring, pneumatic or a combination thereof.

[044] Table (2.1) and cradle (2.2) are made of a metal belonging to the group comprising steel, stainless steel, hot work stainless steel, carbon steel, hot work steel, structural aluminum and cast iron . As the cradle (2.2) directly receives the still hot cast on it, it is made of a more heat resistant material, such as preferably hot work steel and hot work stainless steel. Already the table (2.1) must be light and resistant and therefore preferably produced in structural aluminum, cast iron, or carbon steel.

The imaging means (3) comprises a set of 4 to 8 cameras (3.1) which have a translation and rotation with each other; and a set of 4 to 8 lights (3.2); which enable the piece to be fully visualized and to have adjacent images with overlapping points in one another. In the preferred embodiment of the invention, the imaging means (3) comprise a set of 6 cameras (3.1) which have a translation and rotation between each other. itself; and the set of 6 lights (3.2) assists in image capture by correcting the poor lighting problems of the part.

[046] In the present invention, the cameras in the camera set belong to the group comprising: digital camera, smart camera, webcam and analog camera. Preferably, the cameras belong to the group comprising: digital camera and smart camera.

[047] The dimensional measuring device (1) enables dimensional analysis, inspection and alignment of parts on the production line without endangering the health of the operator.

[048] The invention also relates to a part analysis and inspection system (4) comprising a device (1) for dimensional measurement of parts; means of integration (MSI); and safety features (IS) that integrated with the plant's production control system enable data collection and processing for dimensional, geometric analysis and inspection and part alignment.

[049] In system (4) the device (1) for dimensional measurement of parts comprising a base (2) for supporting the part; image acquisition means (3) for dimensional inspection of the part on the production line.

The base (2) comprises a table (2.1); a cradle (2.2) and a shock absorber (2.3). The table (2.1) serves as support for the image acquisition system (3) and for the part to be inspected; The cradle (2.2) is used for laying the piece and keeping it immobilized, thus allowing the dimensional analysis to be performed with greater precision.

[051] The damper (2.3) serves to mitigate vibration; comprising an assembly of dampers fitted to each of the table supports (2.1); the dampers are state of the art.

[052] Table (2.1) and Cradle (2.2) are made of a metal belonging to the group comprising steel, stainless steel, hot work stainless steel, carbon steel, hot work steel, structural aluminum and cast iron . As the cradle (2.2) directly receives the still hot cast on it, it is made of a more heat resistant material, such as preferably hot work steel and hot work stainless steel. Already the table (2.1) must be light and resistant and therefore preferably produced in structural aluminum, cast iron, or carbon steel.

[053] Imaging means (3) comprise a set of 4 to 8 cameras (3.1) which have translation and rotation with one another; and a set of 4 to 8 lights (3.2); that enable the piece to be fully viewed and adjacent images to have overlapping points with each other. In the preferred embodiment of the invention, the imaging means (3) comprise a set of 6 cameras (3.1) which have translation and rotation with one another; and the set of 6 lights (3.2) assists in image capture by correcting the poor lighting problems of the part.

[054] In the present invention, the cameras in the camera set belong to the group comprising: digital camera, smart camera, webcam and analog camera. Preferably, the cameras belong to the group comprising: digital camera and smart camera.

[055] The device (1) for dimensional part measurement enables dimensional, geometric analysis and inspection of parts on the production line without endangering operator health.

[056] The integration means (MSI) promotes communication between a production line robot (1R1), the device (1), the security means (IS), and a power supply (Al) comprising a data acquisition board (5.1); and a computer (5.2) connected to a database (5.2).

[057] The data acquisition board (5.1) is a state of the art board such as an NI USB board, among others. A network connection such as an Ethenet card known to those skilled in the art connects the card (5.1) to the computer (5.2), which can be any prior art personal computer that has an open or closed operating system that can enable the execution of the software inherent to the realization of the invention, such as Blender, Labview, and Matlab.

[058] The integration means (MSI) are connected to the device (1) by means of the control board (Cl) which connects to the 1/0 interfaces that turn the imaging means on and off (3). The imaging means (3) are comprised of the set of 4 to 8 cameras (3.1) and the set of 4 to 8 lights (3.2).

[059] The invention further relates to a method of automated dimensional control through computational and geometric imaging of parts of the present invention comprising the steps of: (a) initialization; (Image acquisition and processing; (c) Construction of a 3D image of the part; (d) Analysis of the part; (e) Removal of the analyzed part; and (f) Rest.

[060] The startup step comprises checking the safety system items that are monitored by the safety relay, and setting all the variables required for system startup.

[061] In this step; an assembly line robot picks up the automotive part and places it on the cradle (2.2) of the table (2.1); and then returns to a rest position where it will wait until the process is complete. The system then clears the variables required for acquisition, processing and analysis, and sends the signal to the imaging means (3) to initiate step (b).

[062] In step (b) the imaging means (3) are calibrated and then capture between 12 to 20 images per camera from the set of 4 to 8 cameras (3.1) so that the part is fully viewed and adjacent images have overlapping points with each other. The 4-8 set of lights (3.2) emitting a spotlight corrects for problems with incident lighting on the part by enabling proper image capture. The cameras acquire the images and send them to the computer.

Preferably, the imaging means (3) captures 15 images per camera from the 6-camera array (3.1), and the 6-array of lights (3.2) emits a continuous light focus.

[064] Figure 2 shows the focal length data, position and orientation of each camera of the 6 camera set of the preferred embodiment of the invention. Where: PW [XW, YW, ZW] = global coordinate system zero; (.Οχχ, Ογΐ ') = zero of camera 1 coordinate system; Χx2 Oyz) = zero of camera 2 coordinate system; (0x3.0yz) = camera coordinate system zero; (O 4 Oy4) = camera coordinate system zero; (0-5 Oys) = camera coordinate system zero 0; (O 6 Oyb) = camera coordinate system zero;

[065] The focal length of each camera shown in Figure 2 is presented as: pc1 = focal length of camera 1; FC2 = focal length of camera 2; FC3 = camera focal length 3; FC4 = camera 4 focal length; FC5 = camera focal length 5; FC6 = camera focal length 6 [066] The position and orientation of each camera is given as follows: 1RT2 - rotation and translation matrix between cameras 1 and 2; 2RT3 - rotation and translation matrix between cameras 2 and 3; 3RT4 = rotation and translation matrix between cameras 3 and 4; 4RT5 - rotation and translation matrix between cameras 4 and 5; 5RT6 = rotation and translation matrix between cameras 5 and 6;

[067] With this data, processing takes place after the images have been uploaded to the computer to obtain homologous points of the image pairs and, by the execution of a calibration algorithm and processing software, pertaining to the state of the art, such as like the SIFT Algorithm, and the software is, for example, Blender, Labview, and Matlab.

[068] Step (c) a 3D piece-by-piece image is constructed using state-of-the-art software such as CAD and an image is displayed on the computer screen as shown in the figure. 3

[069] In the analysis step, the 3D image is inspected and data on the accuracy of the dimensions of the part as a whole, as well as from each of its sections are obtained. This data is compared to the standard part data stored in the system. After comparison, and according to the tolerance set for the part, the decision-making process is performed whether or not the part is approved.

[070] These data are obtained from a calibration procedure performed by the software employed, where, πΐ is the image plane of camera 1; π2 is the image plane of camera 2; π3 is the image plane of camera 3; π 4 is the image plane of camera 4; π5 is the image plane of camera 5; πβ is the image plane of camera 6.

[071] With the data of the intrinsic and extrinsic parameters obtained, it is possible to control the illumination and background color variables so that there is the best enhancement of the edges of the part in order to find the largest number of correlated points between the cameras of the set. 4 to 8 cameras (3.1), providing the highest possible accuracy in the measurement process.

[072] In step (e) the robot removes the analyzed part from the cradle (2.2) and carries it to an air carrier according to the result of the analysis. If approved in the analysis, the part is transported to another sector of the production line; If it fails the analysis, the part is discarded.

[073] Inspections of hot forged parts currently take up to 30 minutes to perform, and expose the operator to hazards such as burns and hand damage. The automated dimensional control method through computational and geometrical imaging of parts on the production line of the present invention enables a lightweight part between 1 and 20 seconds to be remotely inspected, which eliminates the risk of operator contact with the part in question. analyze. Preferably, each piece is analyzed between 5 and 15 seconds; more preferably, each piece is inspected within 10 seconds.

[074] In step (f), the robot returns to a rest position, where it waits for a new part to be measured, and re-establishes the starting conditions for a new cycle.

[075] Moreover, the invention deals with the use of the image analysis system (4) for dimensional measurement of parts in the quality control of a production line, according to the automated dimensional control method through computational imaging and geometry of parts in the production line of the present invention comprising the steps of: (a) initialization; (b) image acquisition and processing; (c) Construction of a 3D image of the part; (d) analysis of the part; (e) removal of the analyzed part; and, (f) Rest.

[076] The objects of this invention described above have their embodiments illustrated by the examples below. However, it is noteworthy that the examples given are merely illustrative and should not be used to limit the rights of inventors, which are restricted solely to that contained in the appended claims.

Finally, while the invention has been broadly described, it is obvious to those skilled in the art that various changes and modifications may be made without such changes being covered by the scope of the invention.

Examples Example 1 - Measurement prototype: [078] The development of the measurement prototype is done with the aid of the BLENDER platform (http://wiki.blender.org, 2011) to build 3-D virtual models. Such a platform allows to obtain images of the projections of the object in the cameras, in a virtual environment. The MATLAB platform (Inc., 2011) is also used as a mathematical tool for processing and analyzing information from images.

[079] Blender is open source virtual environment simulation software optimized for the production of 3D graphics. Blender offers possibilities for modeling, texturing, lighting, animation, rendering and video processing. This software allows the construction of controlled virtual environments, all coordinates (absolute or relative) being known.

[080] To obtain images, Blender allows you to simulate ideal cameras, according to the model described, with the option of adjusting focus, orientation, number of pixels and their respective position, among others.

[081] Figure 3 shows the model adopted for the arrangement of the X, Y and Z coordinate system in each camera inserted in the measurement system, and shows the image plane where the points are projected.

[082] From the coordinate system of each camera, the layout of the metering system is shown, shown in figure 2. The system consists of six cameras, each with its reference system, and establishes a nomenclature. appropriate for each system parameter as described above.

[083] Intrinsic and extrinsic parameters are obtained from the camera calibration procedure.

[084] From the system layout and its nomenclature, the virtual layout was created and this way it is possible to control, the illumination and background color variables were adjusted so that there is the best edge enhancement in order to find the highest number of correlated points between the 6 cameras, providing the highest possible accuracy in the measurement process.

[085] Besides controlling these variables, in the virtual layout it is also possible to control the intrinsic parameters of the system, table 1.1 shows the values established in the construction of the virtual model.

[086] The virtual layout was developed to simulate the working conditions of the measurement system and from this virtual model will be simulated the images and other important aspects for the validation of the system.

Example 2 - Description of the prototype operation sequence: [087] From the virtual layout developed in Blender, it is necessary to establish a sequence of work operations that makes it possible to reconstruct the object, in this case the crankshaft. necessary steps to enable crankshaft reconstruction.

[088] The first step to be detailed will be the extraction of intrinsic and extrinsic parameters obtained from the virtual model, the obtaining of images, the processing of images with the SIFT algorithm to obtain homologous points in pairs of images and to finalize. the mathematical model of the 3D crankshaft reconstruction.

Example 3 - Measurement System 3D Calibration: [089] To obtain the calibration of the measurement system it is necessary to estimate the intrinsic and extrinsic parameters of the system, so the chessboard calibration method will be used. , known to those skilled in the art. The same 6-camera virtual model will be used, with the chessboard inserted and the crankshaft removed.

[090] The chessboard inserted in the virtual model has dimensions of 800 x 800 x 1 (mm), which is ideal for viewing cameras.

[091] For the correct estimation of parameters it is recommended to get 15 images on each camera for each pair of cameras to be calibrated, ie to estimate the 1RT2 distance 15 images are required for each camera and proceed with the calibration algorithm. .

[092] From the choice of the planar calibration method with chessboard, it is necessary to choose a calibration tool that has the necessary tools to estimate the parameters according to the method of (ZHANG, 2000).

[093] Thus, the tool chosen for camera calibration was the camera calibration toolbox for Matlab developed by (BOUGUET, 1999), which is available on the web at http://www.vision.caltech.edu/bouguetj / calib_jdoc /, this app allows you to read calibration images, extract calibration points, the vertices of the squares present in the image; calibrate the camera, display results, control accuracy, add and delete images, correct distorted images, change the camera's intrinsic model by choosing which parameter to optimize, calibrate a stereo system, rectify stereo images and determine 3D points using the principle of Stereo triangulation. Moreover, it allows adapting the data used by the calibration methods proposed by (ZHANG, 2000), (BAKSTEIN, 1999), (HEIKKILA, 1997) to be used by the calibration method implemented in the toolbox.

[094] According to the theory indicated in item 3.2, the camera model used by the implemented calibration method has the following parameters to estimate: —Extrinsic parameters: used to transform the 3D coordinates of the world system into 3D coordinates in the system of each position and orientation of the calibration plan. Thus, there are six extrinsic parameters: the 3D rotation matrix Rc = (3x3) and the three components of the translation vector Tc = (3xl). —Intrinsic Parameters: Used for transforming 3D coordinates in the camera system to 2D coordinates.

[095] There are four intrinsic parameters in the model: —Focal distance: represents the distance, in pixels, between the projection center and the image plane (on the i and j axes) and is placed on the fc vector (2x1). —Optical Image Center: Represents the coordinates, in pixels, of the optical center of the image and is placed in the cc vector (2x1). —Lift coefficient: represents the angle between the x and y axes of the image and is placed in the alpha_c variable. —Distances: represents the distortion coefficients of the image (radial and tangential) and are placed in the vector kc.

[096] For the extraction of the parameters described above, for each camera, the calibration procedure was performed, requiring the following steps: reading the calibration images, extracting the vertices of the squares in the images, camera calibration and presentation of the results.

[097] After reading the images, the vertices of the calibration tray were extracted and for this step it was necessary to provide the parameters in Matlab, such as: Number of images to process - This parameter indicates the number of images to be processed for extraction of calibration points; Wintx and Winty - This parameter estimates the default window size when searching for vertices in X and Y directions respectively; Number of Squares - This parameter allows you to parameterize the amount of squares present in the calibration image.

[098] After inputting the input parameters, the images are shown in the numbered sequence in which four vertices are chosen at the ends of the rectangular pattern by following this procedure: first, the corresponding vertex location is pressed over the image. the origin point of the tray reference window; The other three points are then chosen on the chessboard in a circular order regardless of the direction chosen.

[099] Following the procedure of locating the four vertices in the fifteen photos the algorithm finds the conforming corners of the standard board.

[100] Retrieving 3D information by estimating camera parameters consists of associating some arbitrary values for intrinsic parameters and then estimating extrinsic ones. The process of estimating intrinsic parameters can be obtained by the method described by (BOUGUET, 1999). The results obtained by this method are shown in table 2 which presents an example of the values found for the calibration matrix.

[101] After calibrating each camera, stereo calibration should be performed on camera pairs to estimate extrinsic parameters. From these parameters it is possible to develop the mathematical model for product measurement and determine the 3D coordinates.

[102] The first step in estimating the parameters is to read the intrinsic calibration results of f 'each camera, this function allows you to read the pairs of images obtained by the two cameras (each pair corresponds to the same calibration plane observed by the cameras in different positions / orientations), being by convention, the two cameras which perform the stereo calibration are called: left and right camera.

[103] The input and output parameters shown below: —Name of the Left Calibration File: Name of the file containing the left camera calibration parameters. —Name of the Right Calibration File: Name of the file that contains the right camera calibration parameters. —Stereo Calibration Parameters: Intrinsic parameters, estimated from the individual calibrations of each camera, and extrinsic parameters, estimated from stereo calibration, describing the position of the right camera relative to the left camera are presented.

[104] The extrinsic parameters om and T resulting from stereo calibration are determined by matching the calibration points of each pair of calibration plane images observed by the two cameras in different position / orientation. Thus, these parameters are defined such that considering a point P in 3D space, its two coordinate vectors Xi and Xr in the left and right camera frame, respectively, are related by the rigid body transformation: Xr = R x Xi x T where R is the rotation matrix (3 x 3) corresponding to the rotation vector om (3 x 1).

[105] After the calibration procedure, the extrinsic parameters om and T are obtained for each pair of cameras, as shown in table 3.

[106] For the proposed measurement model it is necessary to perform the intrinsic calibration procedure for each camera and as already mentioned it is necessary to perform the extrinsic calibration procedure for the five camera pairs. According to figure 2.

Example 4 - Positioning the cameras from the global coordinate system: [107] From the declaration of the nomenclature of the model items of the calibration procedure, the next step is to calculate the position of the cameras from the global coordinate system PW. [XW, YW, ZW], being the rotation and translation matrices obtained from camera calibration it is possible to obtain the camera position from equation (1): pc = RPw + T (1) Where: pc = position from the camera from the global coordinate system; R = rotation vector between cameras;

Pw = global coordinate system; T = translation vector between cameras;

[108] According to the rotation matrices the calculation is made for the position of camera 2, taking as an example the data from table 3 we have the following rotation matrix: [109] Similarly for the vector of [110] Being the global coordinate system Pw with the values shown below: [111] Performing camera 2 calculates: [112] Reversing the result gives the value of camera position 2 from Pw. pc2-1 = [1209.576380 0.009532 0.94070] [113] This way you can calculate the positioning of the rest of the cameras by adding the translation matrices, adding the rotation angles and performing the calculations as described above.

Example 5 - Obtaining images of the part (crankshaft) for reconstruction: [114] This step consists of obtaining the images of the crankshaft using the virtual model. As mentioned, with Blender it is possible to obtain images with background control, brightness and other scene characteristics, besides these properties it is necessary to apply a texture to the crankshaft, this procedure is necessary to facilitate the identification of homologous points when the images are processed with the SIFT algorithm.

[115] Thus, to obtain the crankshaft images it is necessary to position the Blender's plane of view on the camera's plane of view.

[116] Selecting the camera takes the image on the selected camera, Figure 4 shows the crankshaft image obtained from the camera image plane 1.

Example 6 - Image Processing: [117] With crankshaft images, it is necessary to process the images with the SITF Algorithm to obtain homologous points in the image pairs. With the aid of the Matlab computational tool it is possible to process image pairs.

[118] This correspondence is very important because with these points it will be possible to calculate the coordinates of the points and reconstruct the crankshaft.

[119] The algorithm returns the result of the corresponding points in a vector with the corresponding numbering in both figures, as shown in figure 5. Finally the algorithm also returns a vector with the f (x, y) pixel coordinates of the location. each point in the corresponding photo, as shown in Figure 6.

[120] With the coordinate vector, as shown in figure 6, it is possible to plot these points on an image plane and from the following topic, perform the 3D crankshaft reconstruction.

[121] A point plotted on image plane 1, where the vector column indicates the location of the point from the vector in figure 5, line one indicates the pixel coordinate of the x-axis coordinate, and line two indicates the coordinate of the pixel on the y axis.

[122] Thus, from the mathematical model that will be explained in the following topic, with a point seen in the image plane of two different cameras, for example, in cameras 1 and 2 it is possible to calculate the coordinates in the space of the system. global coordinates PW [XW, Yw, Zwl Example 7 - Mathematical model for the 3D reconstruction of the. With all the procedures shown above, it is necessary to reconstruct the crankshaft, so it is necessary to use a mathematical model that makes it possible to access the coordinates of the homologous points from the image processing with the SIFT Algorithm, to perform the calculation of the coordinates of it in space from the global coordinate system Pw \ xw, Yw zw \ and plot it.

[124] For this task, a mathematical model will be developed that will be implemented in Matlab and as a final result will return the coordinates of each point in space, in addition to showing the reconstructed crankshaft with the calculated points.

[125] In this model, the first task to be implemented is to locate cameras in space from the PW global coordinate system [XW, YW, ZW]. Each camera should be positioned from the rotation and translation matrices obtained from the camera calibration procedure. From this need, we have the following equations for camera location.

[126] For camera 1 we have equation 3: pcl - Pw (3) [127] That is, the center of camera 1 coincides with the global coordinate system Pw [Xw. Yw, Zw].

[128] For camera 2 we have equation 4: pc2 = R12Pw + 7 * i2 (4) [129] Where: pc2 = position of camera 2 from the global coordinate system; R12 = rotation vector between camera 1 and 2; P_w = global coordinate system; T12 = translation vector between camera 1 and 2;

[130] For camera 3 we have equation 5: pc3 = R23pc2 + T23 (5) [131] Where: pc3 == position of camera 3 from the global coordinate system; R23 = rotation vector between camera 2 and 3; pc2 = camera position 2 from the global coordinate system; T23 = translation vector between camera 2 and 3;

[132] For camera 4 we have equation 6: pc4 = R34pc3 + 7 * 34 (6) [133] Where: pc4 = camera 4 position from the global coordinate system; R34 - rotation vector between camera 3 and 4; pc3 = camera position 3 from the global coordinate system; T34 = translation vector between camera 2 and 3;

[134] For camera 5 we have equation 7: pc5 = /? 45pc4 + Τ45Π) [135] Where: pc5 = position of camera 5 from the global coordinate system; R4 5 = rotation vector between camera 4 and 5; pc4 = camera position 4 from the global coordinate system; T45 = translation vector between camera 4 and 5;

[136] For camera 6 we have equation 8: pc6 = R56pc5 + T56 (8) [137] Where: pc6 = camera 6 position from the global coordinate system; R56 = rotation vector between camera 5 and 6; pc5 = camera position 5 from the global coordinate system; T56 = translation vector between camera 5 and 6;

[138] From the positioning of cameras in space, as shown in equations 3, 4, 5, 6, 7 and 8, it is necessary to read the vectors generated by the SIFT algorithm, and plot the coordinates of each point processed in its image plane, this way you can calculate its respective coordinate in space.

[139] An example of obtaining a point in space from the theory shown in 2.8.5, that is, from the access to pixel coordinates f (x, y) of a homologous point, one should construct two lines that depart from their respective focal lengths until they intersect in space, the distance between PW [XW, YW, ZW] to the point P [X, Y, Z] indicates the coordinates of this point in space.

[14Ο] The following equations are mathematically used to plot the points on their respective image planes.

[141] To mathematically locate the set of points ρ'1 [χ ', γ'] in the image plane nl with respect to PW [XW, YW, ZW \, use equation 9: PttI - [(960 - Pulx) - 480 (270 - Pitly) FC1] {9) [142] Where: Ρπΐ = Location of the point on the image plane nl with respect to PW {XW, YW, Zw];

Ptíxx = pixel coordinate of the point on the X axis, whose coordinate is accessed in the location vector vector f (x, y), as shown in figure 6;

Pitly = coordinate in pixels of the point on the Y axis, whose coordinate is accessed from the location vector vector / (x, y), as shown in figure 6; PCI = camera focal length 1, result of camera calibration;

[143] To mathematically locate the set of points p'2 [x ', yr] in the image plane π2 with respect to Pw [Xw, yw, zw], use equation 10: Ρπ2 = [((960 - Ρπ2χ) - 48θ) + (lRT2x) (270 - Pn2y) + (lRT2y) FC2 (10) Where: Ρπ2 = Location of the point on image plane n2 with respect to PW [XW, Yw, Zw]; Ρπ2χ = coordinate in pixels of the point on the X axis, whose coordinate is accessed in the location vector vector f (x, y), as shown in figure 6;

Ptí2y = coordinate in pixels of the point on the Y axis, whose coordinate is accessed in the location vector vector / (;; x:, y), as shown in figure 6; lRT2x and lRT2y = XeY components of the translation vector between cameras 1 and 2; FC 2 = camera focal length 2, result of camera calibration;

[144] To mathematically locate the set of points p'3 [x ', y'] in the image plane π3 with respect to P \ V [XW, YW'ZW], use equation 11: Pn 3 = [ ((960 - Pn3x) - 480) + (lRT2x) + (2RT3x) (270 - Ρπ2γ) + (lRT2y) + (2P73y) FC3] {11) [145] Where: Ρπ3 = Location of the point in the image plane n3 with respect to Pw l.Xw Yw> Zwl 'Ρίι3χ = coordinate in pixels of the point on the X axis, whose coordinate is accessed in the location vector vector fix.y'), as shown in figure 6; Ρπ3γ = pixel coordinate of the point on the Y axis, whose coordinate is accessed in the location vector vector / (^, y) r as shown in figure 6; lRT2x and lRT2y = XeY components of the translation vector between cameras 1 and 2; 2RT3x and 2RT3y = XeY components of the translation vector between cameras 2 and 3; FC3 = camera focal length 3, result of camera calibration;

[146] To mathematically locate the set of points P'4 [* ', y'] in image plane n4 with respect to PW \ XW, YW, ZW], use equation 12: Ρπ4 = [((Ρπ4χ - 480)) + (-4RT5x) + (-5RT6x) (270 - Pm4y) + 4RT5y FC4] (12) Where: Ρπ4 = Image point location n4 with respect to 30 Pw [Xw, Yw, Zw]; Ρπ4χ = coordinate in pixels of the point on the X axis, whose coordinate is accessed in the location vector vector as shown in figure 6;

Pix4y = coordinate in pixels of the point on the Y axis, whose coordinate is accessed in the location vector vector f (x, y), as shown in figure 6; 4RT5x and 4RT5y = X and Y components of the translation vector between cameras 4 and 5; 5RT6x = X component of the translation vector between cameras 5 and 6; FC4 = camera 4 focal length, result of camera calibration;

[147] To mathematically locate the set of points p'5 \ x ', y'] in the n5 image plane with respect to pw [xw, Yw, zw], use equation 13: Ρπ5 = [((Ρπ5χ - 480)) + (-5RT6x) (270 - Pm5y) + (4RT5y) FC5] (13) [148] Where: Ρπ5 = Location of the point on image plane n5 with respect to Pw [Xw.Yw, Zw];

Ptt 5x = coordinate in pixels of the point on the X axis, whose coordinate is accessed in the location vector vector f (.x.yh as shown in figure 6;

Pm5y = coordinate in pixels of the point on the Y axis, whose coordinate is accessed in the location vector vector / (x, y), as shown in figure 6; 4RT5y = Y component of the translation vector between cameras 4 and 5; 5RT6x = X component of the translation vector between cameras 5 and 6; FC5 = camera focal length 5, result of camera calibration [149] Finally, to mathematically locate the set of points p'6 [x ', y'] in the image plane πβ with respect to PW \ XW, YW, ZW], equation 14 is used: Ρηβ = [((Ρπ6χ - 480)) (270 - Pmóy) FCô] (14) [150] Where: Ptt6 = Location of the point on the image plane n6 with respect to Pv l ^ w> Yw>% wl t ππχ = coordinate in pixels of the point on the X axis, whose coordinate is accessed in the location vector vector f (pc, y), as shown in figure 6;

Ptx 6y = coordinate in pixels of the point on the Y axis, whose coordinate is accessed in the location vector vector f (jx, y), as shown in figure 6; FC6 = camera focal length 6 as a result of camera calibration;

[151] From the pixel coordinates access of the homologous points processed with the SIFT Algorithm and its positioning in their respective image planes, it is necessary at this moment to calculate it in space, so it will be implemented the processing of the coordinate calculation.

[152] With pixel coordinates relative to pAXW. Yw> · The first procedure for calculating the point in space is to find the matrices A and B that will be used to find the intersection of the lines formed by the point projection on the two cameras.

[153] The 2x1 matrix A indicates the position of the point in X and Y from the zero of the PW world coordinate system [XW, Yw, Zw]. Already the matrix B of dimensions 2x1 relates the focal length of the cameras whose points are shown in their respective image plane.

[154] With the matrices A and B solved, we must solve equation 2.79 which will result in a matrix of dimensions 2x1, where we will find the term t [155] E to finish and find P [X, Y, Z] From the global coordinate system Pw [Xw, Yw, Zw] it is necessary to apply the rotation matrices, this rotation is necessary because the cameras are positioned with a rotation angle.

[156] The entire procedure should be repeated for each pair of homologous points.

Example 8 - Mathematical Model for Statistical Comparison between the CAD Model and the Reconstruction Method: [157] From the mathematical model for crankshaft reconstruction, with the application of the model shown above, it is necessary to verify the accuracy of this reconstruction a demonstrate the effectiveness of the measurement system application.

[158] In this topic, mathematical models and the methodology for studying the error caused by the variables of the camera measurement process will be proposed.

[159] In the first study it is necessary to create key points for comparing the reconstructed crankshaft alignment, so the information shown in chapter 1 on the manufacturing process will be used. From these points it will be possible to establish the crankshaft alignment.

[160] From the alignment of the points, it is necessary to establish a mathematical model for error checking, this model will be based on statistical procedures for determining the error caused by the measurement process variables, such as the estimation of intrinsic parameters and extrinsic.

[161] For the statistical study of error, two studies are proposed: [162] The first study will analyze only the error obtained by comparing the coordinates of each CAD model point with their corresponding coordinates obtained from the Matlab reconstruction. ;

[163] The second study will use the error results obtained in the first study to analyze the reliability of the reconstruction using a dispersion analysis of this error.

Example 9 - Mathematical model for checking the alignment of a crankshaft: [164] In the previous items of this chapter we developed the mathematical models for crankshaft reconstruction, in this topic it is necessary to develop a mathematical model for checking the characteristics of the crankshaft. In the case of the crankshaft, some characteristics will be studied, such as the location of the center of gravity and its respective angles of key points to verify its alignment in space.

[165] From the need cited above, Figure 7 shows the three cylinders that were established as key points for crankshaft alignment, these cylinders coincide with the analysis of the manufacturing process.

[166] To perform alignment, perform the following tasks: [167] From the determination of the cylinders as shown in figure 7, it is necessary to locate their center of gravity from the coordinates. of the point cloud. In this way the points should be searched from the established search criteria, that is, to establish the limits for the search of the points in the cloud in order to locate the coordinates that are part of the cylinders for later calculation of the center of gravity.

[168] From the location of the points belonging to each cylinder, there is a set of two-dimensionally located pixels that represents the cylinder's outline in the YZ plane. Knowledge of these points has one purpose, three-dimensional reconstruction. Obtaining the center of gravity of each of the cylinders means knowing the equation that represents it. In this way the general equation of an ellipse can be used, in a two-dimensional plane, as (COXETER, 1969) this equation is written as shown in the equation below. ax2 + by2 + cx + dy + exy + f = 0 (15) [169] From the coordinates, the set of points representing the cylinder contour is obtained. These points on the two-dimensional plane have their Y and Z coordinates shown from the zero of the PW [XW, YW, ZW} world coordinate system.

[170] Using the general ellipse equation and the coordinates of all cylinder contour points, we can write a system of linear homogeneous equations, as shown in equation 16, where the number of minimum contour points of the circle that should be known is six, but the more points, the better the solution found. Points range from 1 to i.

[171] Decomposing into singular values (svd) of the matrix ix 6 gives the result shown in equation 17.

[172] The elliptic coefficients a through / from equation 4.16 are found in the eigenvector of the decomposed matrix V associated with the smallest eigenvalues of the diagonal matrix D.

[173] Thus, knowing the equation of the ellipse in its general form, shown in equation 15, it is possible to extract the parameters corresponding to the major and minor arm sizes, slope, and center coordinates. Another way to write the equation of an ellipse with this new parameter is shown in equation 18.

[174] Where: aei = Larger arm of the ellipse; bel = Smaller arm of the ellipse;

Zc = Z axis center coordinate;

Yc = Y axis center coordinate; Θ = Angle of inclination of the ellipse;

Finally, the relationships between the coefficients of the equation of ellipse 4.15 and their parameters of equation 4.18 are shown in equations 19 through 23.

[176] Because the coordinates of the ellipse contour points are used, as already mentioned and shown in equation 16, the parameters found for the ellipse ZC, YC, aei, and bel are dimensionless.

[177] Applying the procedure exposed to the coordinates of each of the reference cylinders gives the coordinates of the centers of gravity of cylinders 1, cylinder 2 and cylinder 3, with the coordinates of the center of the cylinders shown below. from the global reference system Pw CGX = [335.81 -4594.61 1431.52] CG2 = [2614.61 -4594.61 1431.52] CG3 = [4858.7 -4594.61 1431.52] [178] From the coordinates of the center of gravity of the cylinders, an alignment vector must be found across the centers of gravity to compare crankshaft alignment with reference axes. VA = (CG3 - CGX) (24) [179] From equation 24 and the coordinates of the centers of gravity of crankshaft cylinders 1 and 3, we have the following vector: VA = (4858.7 - 335.81 T) + ¢ -4594.61 - (-4594.61) /) + (1431.52 - 1431.52 k) [180] Thus, we have the following alignment vector: VA = (4522.89 T ) + (Oj) + (0 k) Π 81] From the alignment vector, equation 25 must be used.

[182] Where: uVA = unit vector; VA = Cartesian vector module; VAx ΐ Λ-VAy j + VAZ k = Cartesian vector;

[183] From the application of equation 25, it is possible to calculate coordinate directional angles from equation 26, 27 and 28.

[184] A being the angle formed by the vector with the X axis.

[185] Where β is the angle formed by the vector with the Y axis.

[186] Being γο angle formed by the vector with the Z axis.

[187] From the application of equations 25, 26, 27 and 28 we have the following rotation matrix for the crankshaft.

Example 10 - Mathematical Model for Statistical Error Analysis: [188] The procedures for error checking between the standard crankshaft (CAD) and the reconstructed crankshaft, and the error checking shall be as follows: comparison between a standard crankshaft point cloud (CAD) and the rebuilt crankshaft point cloud.

[189] A point cloud I is a set of three-dimensional points pi, with their respective coordinate in space, such that i = N and pi € R3, where N is the size of the cloud and only has geometric information about these points, without knowledge of any topological relationship obtained from a CAD model.

[190] From the coordinates obtained from the standard crankshaft point cloud (CAD) and the reconstructed crankshaft point cloud, you should find the nearest corresponding points in each of the clouds to check the proximity between these points. To perform this task the following nomenclature will be used. di - coordinates of cloud point distances in CAD; dj = coordinates of cloud point distances in Matlab;

[191] From the CAD model and its respective reconstruction, two point clouds are obtained, and for one the vertex has the respective coordinates in the CAD model.

Pdi = [1093,87 -1440,69 732] [192] Similarly, the coordinates of the same point are obtained from the point cloud obtained from the Matlab reconstruction.

Pdj = [1093,4040 -1423,6970 722,833] [193] From the coordinates of the points obtained in the CAD model and the coordinates of the points obtained in Matlab, shown above, it is necessary to calculate the distance between the point and the system origin Pw \ XW, YW, ZW], both in the CAD model and the Matlab model.

[194] To calculate the modulus of the model position vector in CAD, equation 21 will be used.

[195] Similarly to calculate the modulus of the position vector of the model in Matlab, equation 4.22 will be used.

[196] In order to perform statistical error analysis between the CAD model and Matlab model coordinates, the distance between the points dt and dj in space must be obtained, ie the vectorial deviation of the location in space must be calculated. . Thus, to calculate the vector deviation, equation 23 will be used: [197] Where: dij = distance between the point of the CAD model and the Matlab model;

Xi = X coordinate of the point in the CAD model;

Xj = X coordinate of the point in the Matlab model;

Yi = Y coordinate of the point in the CAD model;

Yj = Y coordinate of the point in the Matlab model; = Z coordinate of the point in the CAD model;

Zj = Z coordinate of the point in the Matlab model;

[198] Applying the equation for the example gives the following result. = (1093.87 - 1093.40) 2 + (-1440.69 -142.69) 2 + (732 - 722.833) 2 [199] Performing the calculation gives the following result: d ^ = 19, 31 (mm) [200] The result obtained from equation 23 is the error, so all points in common between the CAD model and the Matlab model can have an error that can be calculated. In this way the average measurement system error can be obtained.

[201] Where: d = mean error; dij = error; n = number of common points calculated;

[202] For the analysis of the error between the position of the CAD model points (d, ·) and the Matlab model (dj) with respect to the PAXw.Y «, z„] reference system, the calculation must be performed. relative error, which is given from equation 25: [203] Relative error can be given as a percentage, so equation 4.26 shows the error as a percentage: [204] Applying equation 26 for error calculation Relative to the given point as an example we have: [205] When calculating, the relative error value for the given point with an example is: 5rel = 0.99% [206] From the error in each point you can calculate the mean error using equation 27: [207] Where: ô_rel = mean relative error; rel = relative error; n = number of points;

[208] With the relative error data at all points compared, it is possible to construct graphs with the following characteristics: relative error at each point, maximum error, minimum error and mean error.

[209] In addition to the graphs with the aforementioned characteristics, you can also establish the correlation index between CAD (real) coordinates and calculated coordinates from Matlab, this correlation index is calculated by equation 28: [210] From the correlation coefficient, shown in equation 28, plot the correlation graph with the coordinates dt and dj.

[211] From the correlation (CAD) and reconstructed (Matlab) correlation graph, an important feature that should be used for analysis is the slope of the straight line resulting from the correlation graph, equation 29 shows the function to estimate dj from dL. dj = α + bdt (29) [212] Since the parameter α is given by equation 30: [213] Since α is the angular coefficient of the straight line calculated from the above equation 29, the value of b is calculated from equation 30, b = dj - adl (31) Example 10 - Mathematical Model for Error Scattering Analysis: [214] With the result of the error and the relative error obtained it is necessary to present the data in such a way that it is possible to analyze the tendency of the data to remain around the central value, ie around the mean error.

[215] From this need, this topic will present how the data will be grouped for trend analysis from the construction of a normal curve and its subsequent analysis.

[216] To analyze the frequency from the normal distribution, you need to use the mean error value as shown in equation 27: [217] After averaging you need to calculate the standard deviation so you can rewrite the same equation applied to the error as shown by equation 432.

Where: S = standard deviation of error; 5_rcl = mean relative error; õrei = relative error; n = number of points;

[218] With standard deviation, frequency distribution should be performed in such a way as to account for the number of error occurrences in each class, taking into account the mean error value as a basis and using standard deviation as the interval. class.

[219] For better organization this data should be arranged in a table and so it is possible to plot the relative error frequency histogram, Figure 9 shows an example of the frequency histogram.

Claims (20)

Device (1) for dimensional measurement of parts comprising a base (2) for supporting the part; by the image acquisition means (3) for dimensional inspection of the part. Device according to Claim 1, characterized in that the base (2) comprises a table (2.1); a cradle (2.2) and a shock absorber (2.3); the table (2.1) serves as support for the image acquisition system (3) and for the part to be inspected; the cradle (2.2) serves to settle the piece and to keep it immobilized; and the damper (2.3) serves to mitigate vibration. Device according to claim 1, characterized in that the image capture means (3) comprise a set of 4 to 8 cameras (3.1) which have a translation and rotation with each other; and a set of 4 to 8 lights (3.2); that enable the piece to be fully visualized by having adjacent images with overlapping points between them. Device according to claim 3, characterized in that the image capture means (3) comprise a set of 6 cameras (3.1); and the set of 6 lights (3.2). 5 - Part analysis and inspection system (4), characterized in that it comprises a device (1) for dimensional measurement of parts; means of integration (MSI); and safety features (IS) that integrated with the plant's production control system enable data collection and processing for dimensional, geometric analysis and inspection and part alignment. System according to Claim 5, characterized in that the device (1) for dimensionally measuring parts is as claimed in claims 1 to 4. System according to claim 5, characterized in that the integration means (MSI) provides communication between a production line robot (1R1), the device (1), the security means (IS), and a power supply (Al) comprising a data acquisition board (5.1); and a computer (5.2) connected to a database (5.2). System according to claim 5, characterized in that the data acquisition board (5.1) is a USB NI card; and an Ethenet card connecting the card (5.1) to the computer (5.2). System according to Claim 5, characterized in that the means for integrating (MSI) with the device (1) occurs via the control board (Cl) which connects to the interfaces 1/0 which switch the means on and off. (3) comprised of the set of 4 to 8 cameras (3.1) and the set of 4 to 8 lights (3.2). 10 - Method of dimensional control through computational and geometric image of parts characterized by comprising the steps of: (a) Initialization; (b) image acquisition and processing; (c) Construction of a 3D image of the part; (d) analysis of the part; (e) removal of the analyzed part; and, (f) Rest. Method according to claim 10, characterized in that the initialization step comprises verification of the safety system items that are monitored by the safety relay, in addition to establishing all the variables necessary for the system initialization. Method according to claim 10, characterized in that an assembly line robot picks up the automotive part and places it on the cradle (2.2) of the table (2.1); and then return to a rest position where you will be waiting until the process is completed; The system then clears the variables necessary for acquisition, processing and analysis, and sends the signal to the imaging means (3) to initiate step (b). Method according to claim 10, characterized in that in step (b) the image capture means (3) are calibrated and then capture between 12 to 20 images per camera of the set of 4 to 8 cameras ( 3.1) so that the piece is fully visualized and adjacent images have overlapping points with each other. A method according to claim 13, characterized in that the set of 4 to 8 lights (3.2) emit a spotlight corrects the problems in incident lighting on the part by enabling proper image capture; and the cameras acquire the images and send them to the computer. Method according to claim 10, characterized in that the processing takes place after the images are loaded on the computer to obtain homologous points of the image pairs and, by the execution of a calibration algorithm and processing software. The. Method according to claim 10, characterized in that in step (c) a 3D piece-by-piece image is constructed using state of the art software. A method according to claim 10, characterized in that in the analysis step the 3D image is inspected and the data relating to the accuracy of the dimensions of the part as a whole as well as of each of its sections are obtained; the data is compared to the standard part data stored in the system by statistical comparison. Method according to Claim 10, characterized in that in step (e) the robot removes the analyzed part from the cradle (2.2) and loads it into an air carrier according to the result of the analysis. Use of an image analysis system for dimensional measurement of parts as described in claims 5 to 9 characterized in that it has application in the control of the geometrical characteristics of parts and which demand high precision in their measurements. Use of an image analysis system for dimensional measurement of parts as described in claims 5 to 9 characterized for being for quality control.