Intelligent detection method and system for image recognition of hole positions of automotive parts
By identifying the geometric features and the cluster of positioning points of components, combining self-positioning modules and limit detection solutions, using multi-eye camera systems and intelligent detection modules for parallel detection, the poor accuracy and low efficiency of porous position detection in the existing technology are solved. The problem of this method achieves high accuracy and high efficiency hole position detection.
Patent Information
- Application Number
- CN202410781075.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-18
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-06-18
AI Technical Summary
The existing online pore position detection methods have problems of poor accuracy and low efficiency in the detection of porous position components, which cannot meet the needs of large-scale production.
Using intelligent detection method for hole position image recognition of automotive parts, we can identify the geometric features of target parts, identify the cluster of positioning points, combine the self-positioning module and limit detection scheme to carry out image acquisition and positioning point cloud recognition, and use multi-purpose The camera system and the intelligent detection module perform parallel detection to fit the detection results to improve detection accuracy and efficiency.
It realizes high accuracy and high efficiency detection of porous parts, improves the accuracy and efficiency of detection, and meets the needs of large-scale production.
Smart Images

Figure CN118552514B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of detection equipment, and in particular to an intelligent detection method and system for image recognition of hole positions of automotive parts. Background Art
[0002] During the manufacturing and assembly of automotive parts, the hole position accuracy of parts has an important impact on the performance and safety of the car. Existing online hole position detection methods are prone to misjudgment or omission when facing multi-hole parts, and cannot meet the needs of large-scale production. There are technical problems such as poor multi-hole position detection accuracy and low efficiency. Summary of the invention
[0003] The purpose of this application is to provide an intelligent detection method and system for hole position image recognition of automotive parts, so as to solve the technical problems of poor accuracy and low efficiency of multi-hole position detection in the prior art.
[0004] In view of the above technical problems, the present application provides an intelligent detection method and system for image recognition of hole positions of automotive parts.
[0005] In a first aspect, the present application provides an intelligent detection method for image recognition of holes in automotive parts, wherein the method comprises:
[0006] Identify the geometric features of the target parts and mark the positioning point cloud set, which includes a distribution point cloud and an edge point cloud, and has point cloud trajectories and class attribution marks under spatial angle conversion;
[0007] In combination with the positioning point cloud set, a self-positioning module is determined, wherein the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module;
[0008] Determine a limit detection scheme based on the hole detection fixture, and obtain a pre-collection strategy of the spatial perspective of the scheme in combination with the front locator;
[0009] The pre-capture strategy performs image capture control of the camera group in response to the multi-camera system to obtain a captured image set;
[0010] Combined with the post-locator, the positioning point cloud recognition mark of the collected image set is executed, and in the intelligent detection module, hole mark recognition and class segmentation are performed, and the matching target detection branch is used for parallel detection to determine the image detection result. The intelligent detection module includes multiple parallel detection branches mapped to the hole type;
[0011] The limit detection result and the image detection result are fitted as the hole position detection result.
[0012] In a second aspect, the present application also provides an intelligent detection method system for image recognition of hole positions of automotive parts, wherein the system comprises:
[0013] A geometric feature recognition component, which is used to recognize the geometric features of the target parts and identify the positioning point cloud set, which includes a distribution point cloud and an edge point cloud, and has point cloud trajectories and class attribution marks under spatial angle conversion;
[0014] A self-positioning component, the self-positioning component is used to determine a self-positioning module in combination with the positioning point cloud set, the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module;
[0015] A limit detection scheme determination component, the limit detection scheme determination component is used to determine the limit detection scheme based on the hole detection tooling, and in combination with the front locator, obtain a pre-collection strategy of the scheme space perspective;
[0016] An image acquisition control component, wherein the image acquisition control component is used for the pre-acquisition strategy to respond to the multi-camera system, perform image acquisition control on the camera group, and obtain an acquired image set;
[0017] A positioning point cloud recognition component, which is used to perform positioning point cloud recognition marking of the collected image set in combination with a post-locator, perform hole mark recognition and class segmentation in an intelligent detection module, match the target detection branch for parallel detection, and determine the image detection result. The intelligent detection module includes a plurality of parallel detection branches mapped to the hole type;
[0018] The limit detection result fitting component is used to fit the limit detection result and the image detection result as the hole position detection result.
[0019] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0020] By identifying the geometric features of the target parts, the positioning point cloud set is identified, including the distribution point cloud and the edge point cloud, as well as the point cloud trajectory and class attribution mark under the spatial angle conversion; combined with the positioning point cloud set, the self-positioning module is determined, including the connection between the front positioner and the multi-camera system, and the connection between the rear positioner and the intelligent detection module; the limit detection scheme based on the hole detection tooling is designed, and the pre-collection strategy of the spatial perspective of the scheme is implemented in combination with the front positioner; the pre-collection strategy responds to the multi-camera system, controls the image acquisition of the camera group, and obtains the collected image set; combined with the rear positioner, the positioning point cloud recognition mark of the collected image set is executed, the hole mark recognition and class segmentation are performed in the intelligent detection module, and the target detection branch is matched for parallel detection to determine the image detection result. The intelligent detection module includes multiple parallel detection branches mapped to the hole type; the limit detection result and the image detection result are combined to form the hole position detection result. Then the technical effect of improving the accuracy and efficiency of multi-hole position detection is achieved.
[0021] The above description is only an overview of the technical solution of the present application. In order to more clearly illustrate the technical means of the present application and to implement it in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The embodiments of the present invention and the following brief description are illustrated in conjunction with the accompanying drawings, which are described as follows:
[0023] Figure 1 A schematic diagram of the process of an intelligent detection method for image recognition of holes in automotive parts in this application;
[0024] Figure 2 This is a structural schematic diagram of an intelligent detection method system for image recognition of hole positions of automotive parts in this application.
[0025] Explanation of the reference numerals: geometric feature recognition component 11, self-positioning component 12, limit detection scheme determination component 13, image acquisition control component 14, positioning point cloud recognition component 15, limit detection result fitting component 16. DETAILED DESCRIPTION
[0026] The present application solves the technical problems of poor accuracy and low efficiency in multi-hole detection faced by the prior art by providing an intelligent detection method and system for hole image recognition of automotive parts.
[0027] The solution in the embodiment of the present technology, in order to solve the above problems, adopts the following overall ideas:
[0028] First, the geometric features of the target parts are identified, and the positioning point cloud set is marked, including the distribution point cloud and the edge point cloud, as well as the point cloud trajectory and class attribution mark under the spatial angle conversion. Then, combined with the positioning point cloud set, the self-positioning module is determined, including the front locator connected to the multi-eye camera system and the rear locator connected to the intelligent detection module. Then, the limit detection scheme based on the hole detection tooling is determined, and combined with the front locator, the pre-acquisition strategy of the spatial perspective of the scheme is formulated. Next, the pre-acquisition strategy responds to the multi-eye camera system, controls the camera group to perform image acquisition, and obtains the acquired image set. Finally, combined with the rear locator, the acquisition image set is marked with positioning point cloud recognition, and the hole mark recognition and class segmentation are performed in the intelligent detection module, matched with the target detection branch, and the image detection result is determined. The intelligent detection module includes multiple parallel detection branches mapped to the hole type.
[0029] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the drawings and specific implementation methods of the specification. It should be noted that the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that for the sake of ease of description, only the parts related to the present invention are shown in the drawings, rather than all of them.
[0030] Embodiment 1
[0031] like Figure 1 As shown, the present application provides an intelligent detection method for image recognition of hole positions of automotive parts, the method comprising:
[0032] Identify the geometric features of the target parts and mark the positioning point cloud set, which includes a distribution point cloud and an edge point cloud, and has point cloud trajectories and class attribution marks under spatial angle conversion;
[0033] Optionally, by interacting with the design files, drawings, 3D models and other materials of the target parts, analyze the shape, size, position and other information of the target parts, extract their key geometric features, obtain the distribution point cloud and edge point cloud, and mark them as positioning point cloud sets. Among them, the distribution point cloud in the positioning point cloud set reflects the distribution of the hole positions on the parts, and the edge point cloud reflects the boundary characteristics of the hole positions.
[0034] The point cloud trajectory under spatial angle conversion refers to the change of point cloud data of the positioning point cloud set under different spatial angles. The class attribution tag refers to a category tag assigned to each positioning point cloud set to distinguish different parts.
[0035] In some embodiments, identifying a set of positioning point clouds includes:
[0036] Traversing the geometric features, taking the hole center as a positioning reference, and combining the relative position of the hole space, determining the distribution point cloud;
[0037] Traversing the distributed point cloud, determining the edge positioning points of the hole position point by point cloud, and obtaining the edge point cloud, wherein the number of edge positioning points is a fixed value;
[0038] The distributed point cloud and the edge point cloud are fused to determine the positioning point cloud set.
[0039] Optionally, in the geometric features of the target parts, the hole center is used as the reference point, combined with the spatial relative position of the hole, to determine the distribution point cloud of the hole. First, the target part coordinate system is used as the reference coordinate system, and the geometric feature data of the target part is traversed to find all the hole center points; then, based on the relative position of the hole center and the hole space, that is, the relative position relationship between the hole center and the reference point of the target part coordinate system, the distribution point cloud is determined, and the distribution point cloud reflects the distribution of the holes on the parts.
[0040] Optionally, in the distributed point cloud of the hole position, edge positioning points are determined one by one to obtain the edge point cloud of the hole position. The number of edge positioning points is a fixed value to ensure that the amount of data of the acquired point cloud data is controllable and the representation of the hole position is accurate. Specifically, the neighborhood range of the distributed point cloud of each hole position is traversed to determine the edge positioning points of each hole position, which reflect the boundary characteristics and spatial distribution of the hole position.
[0041] Optionally, the distributed point cloud is fused with the edge point cloud to determine the final positioning point cloud set. The fusion process may involve registering and overlaying the two sets of point cloud data to ensure that the positioning point cloud set contains all necessary positioning information.
[0042] In some implementations, fusing the distribution point cloud with the edge point cloud to determine the positioning point cloud set includes:
[0043] In combination with the multivariate hole position types, the distribution point cloud is clustered, and the hole position classification is marked in combination with the cluster clusters;
[0044] Based on the fused distributed point cloud and the edge point cloud, an initial angle point cloud is determined, point cloud dynamic analysis is performed at the full spatial angle of the target component, and the position space trajectory of each point cloud is determined as the point cloud trajectory;
[0045] The class attribution mark, the point cloud trajectory and the initial angle point cloud are fused to determine the positioning point cloud set.
[0046] Optionally, first, the distributed point cloud is clustered in combination with the multivariate hole type. Exemplarily, based on a clustering algorithm, such as the K-means algorithm, the point cloud is grouped and similar point cloud data is classified into one category. For subsequent attribution marking, the multivariate hole type refers to different types of holes, and the classification basis of the multivariate hole type includes: blind holes, through holes, countersunk holes and other hole forms; hole specification standards (such as ISO standards, ANSI standards, etc.); functions in parts (such as positioning holes, connecting holes, through holes, etc.).
[0047] Optionally, the hole positions are marked according to the clusters, and a class mark is assigned to each cluster to distinguish different hole position types and provide a reference for subsequent positioning.
[0048] Optionally, the initial angle point cloud is determined based on the fused distribution point cloud and edge point cloud. The initial angle point cloud refers to the point cloud data determined after the fusion of the distribution point cloud and the edge point cloud. Specifically, according to the distribution of the point cloud and the edge features, the initial angle of the target component is estimated and the initial angle point cloud is generated. The initial angle point cloud includes the distribution and boundary information of the hole positions.
[0049] Optionally, the point cloud is dynamically analyzed based on the full spatial angle of the target component to help determine the positional spatial trajectory of each point cloud, i.e., the point cloud trajectory. The full spatial angle refers to multiple observation angles of the target component in three-dimensional space. Point cloud dynamic analysis refers to analyzing the law of point cloud data changing with angle.
[0050] Furthermore, the data of the class attribution mark, point cloud trajectory and initial angle point cloud are fused to determine the final positioning point cloud set, so as to ensure that the positioning point cloud set contains the location information and trajectory information of the hole position, and provide accurate positioning reference for subsequent operations.
[0051] Through such steps, the positioning point cloud can be effectively determined to provide accurate data support for the positioning and assembly of parts.
[0052] In combination with the positioning point cloud set, a self-positioning module is determined, wherein the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module;
[0053] Specifically, the self-positioning module is a module used to determine the position of parts, including a front locator and a rear locator. The front locator is connected to the multi-camera system to capture and process image data from the multi-camera system, while the rear locator is connected to the intelligent detection module to process and analyze image data from the front locator and further process and confirm position information. The entire self-positioning module achieves accurate positioning and detection of parts through the cooperation of the multi-camera system and the intelligent detection module.
[0054] Determine a limit detection scheme based on the hole detection fixture, and obtain a pre-collection strategy of the spatial perspective of the scheme in combination with the front locator;
[0055] Exemplarily, the limit detection scheme based on the hole detection tooling is combined with the front locator to obtain the pre-acquisition strategy of the solution space perspective. First, determine the range of parts or workpieces that need to be limit detected. Including determining the position and number of holes that need to be detected. Then, install the hole detection tooling to ensure that it is accurately positioned on the hole positions that need to be detected. The hole detection tooling includes adapters, sensors and other equipment for detecting the position and state of the hole positions. Next, connect the front locator to ensure that it can obtain the solution space perspective. The front locator obtains the overall position and posture information of the parts through the connected multi-eye camera system. Then, through the acquisition parameters of the front locator, configure the pre-acquisition strategy for obtaining the solution space. Including determining the camera's shooting angle, exposure time, light source brightness and other parameters to ensure that the image of the hole position can be accurately captured.
[0056] In some embodiments, determining a limit detection scheme based on a hole detection tooling includes:
[0057] In combination with the front positioner, the positioning point cloud is traversed, and a single limit detection scheme is determined in combination with the operating mechanism of the detection rod and the spatial positioning structure;
[0058] Combined with the relative position of the detection rod, the single limit detection scheme is collaboratively analyzed to determine the limit detection scheme;
[0059] In combination with the limit detection scheme, the hole detection tooling is controlled to perform limit detection on the target component to determine the limit detection result, wherein the detection dimensions include hole diameter, hole depth trend and position.
[0060] Optionally, the positioning point cloud set is traversed in combination with the front locator. The position information of the component obtained by the front locator is combined with the position and posture information of the hole detection fixture to determine the hole position that needs to be limited.
[0061] Optionally, a single limit detection scheme is determined by combining the operating mechanism of the detection rod and the spatial positioning structure. This includes determining how to use the detection rod to detect the hole position, and how to determine the limit scheme based on the role of the spatial positioning structure. The limit scheme includes the position of the detection rod, the movement trajectory of the detection rod, the operation sequence of the detection rod, etc. Among them, the detection rod and the spatial positioning structure are used to realize the limit detection of the target parts.
[0062] Furthermore, combined with the relative position of the detection rod, a collaborative analysis is performed on the single limit detection scheme to determine how to combine the limit detection schemes of multiple hole positions to ensure the overall limit accuracy. Exemplarily, it includes performing interference analysis on multiple single limit detection schemes, and performing avoidance adjustment based on the interference analysis results to determine the limit detection scheme.
[0063] Optionally, in combination with a determined limit detection scheme, the control hole detection tooling detects the limit position of the target component, and the limit detection result is obtained by controlling the movement of the tooling and the use of sensors.
[0064] The limit detection result includes multi-dimensional information about the limit position of the target component obtained through the limit detection. The detection dimensions of the multi-dimensional information include hole diameter, hole depth trend and position.
[0065] Through this process, combined with the spatial position information provided by the front locator, the positioning point cloud is traversed, and the single limit detection scheme is determined in combination with the operating mechanism of the detection rod and the spatial positioning structure. Then, combined with the relative position of the detection rod, the single limit detection scheme is collaboratively analyzed to determine the final limit detection scheme. Finally, according to the limit detection scheme, the control hole detection fixture performs limit detection on the target parts and determines the limit detection results. Accurate limit detection of the target parts is achieved.
[0066] The pre-capture strategy performs image capture control of the camera group in response to the multi-camera system to obtain a captured image set;
[0067] Optionally, based on the pre-acquisition strategy, the multi-camera system is activated to perform image acquisition control of the camera group, including determining parameters such as the working mode, exposure time, aperture size, etc. of each camera in the multi-camera system to ensure that the quality of the acquired images meets the requirements.
[0068] Optionally, during the image acquisition process, the image quality and camera parameters are monitored in real time, and the acquisition strategy is adjusted in a timely manner to ensure the validity of the acquired image information.
[0069] Combined with the post-locator, the positioning point cloud recognition mark of the collected image set is executed, and in the intelligent detection module, hole mark recognition and class segmentation are performed, and the matching target detection branch is used for parallel detection to determine the image detection result. The intelligent detection module includes multiple parallel detection branches mapped to the hole type;
[0070] Optionally, the positioning point cloud recognition marking of the acquired image set is performed in combination with a post-locator, firstly, by mapping the positioning point cloud information onto the image and marking the corresponding position information.
[0071] Optionally, the image set with the positioning point cloud recognition mark is input into the intelligent detection module. It includes multiple parallel detection branches for specifically processing the detection tasks of different types of holes. Among them, multiple parallel detection branches refer to multiple independent detection models established for different hole types, which can run simultaneously to improve the detection efficiency. The intelligent detection module set up in this way not only improves the accuracy of image detection and analysis, but also provides the possibility of parallel processing, which helps to improve the detection efficiency.
[0072] Optionally, in the intelligent detection module, hole mark recognition and class segmentation are performed on the image set through deep learning and other technologies to ensure effective recognition and classification of marks of different hole types.
[0073] Among them, the image detection results are determined. This includes determining the location, type and other relevant information of the hole, providing accurate data support for subsequent processing. Through the above steps, the post-locator can be effectively used to perform the positioning point cloud recognition mark of the collected image set, and combined with the intelligent detection module to perform hole mark recognition and class segmentation, thereby achieving accurate detection and analysis of the image, obtaining the image-based hole position detection results, and generating image detection results.
[0074] In some embodiments, hole mark recognition and class segmentation are performed, including:
[0075] Acquire branch training data, where the branch training data corresponds one-to-one to the hole type;
[0076] Extracting common features and individual features based on the hole type, setting a learning rate for the branch training data, wherein the learning rate is positively correlated with the individual degree, and the learning rate can be iteratively adjusted;
[0077] Based on the branch training data after the learning rate labeling, supervised learning and convergence condition verification are performed to generate the multiple parallel detection branches;
[0078] The multiple parallel detection branches are pre-connected to an image segmentation layer to generate the intelligent detection module.
[0079] Optionally, to implement hole mark recognition and class segmentation, first, it is necessary to obtain branch training data corresponding to hole types one by one. The branch training data includes an image dataset with hole marks and corresponding annotation information.
[0080] Then, for each hole type, its common features and individual features are extracted. Among them, the common features refer to the features that all hole types have, while the individual features refer to the unique features of each hole type.
[0081] Next, different learning rates are set according to the individuality of the hole type. The individuality reflects the discrimination of a feature for a specific category. Features with high discrimination have higher learning rates so that the model can better learn its features. Optionally, the learning rate is iteratively adjusted, and the learning rate is dynamically adjusted according to the performance of the model during the training process to obtain the best training effect. For example, the learning rate is higher in the early stage of learning, and the learning rate decreases accordingly with iterative learning.
[0082] Furthermore, supervised learning is performed by setting branch training data with a learning rate, and the convergence condition of the model is verified. Specifically, this is achieved by training the model and observing its performance on the validation set, and exemplary includes checking whether the loss function of the model reaches a predetermined threshold to ensure that the model has converged to a sufficiently good state.
[0083] Optionally, multiple parallel detection branches are generated based on the results of supervised learning. Each branch corresponds to a hole type and can identify the mark of the hole type. The multiple parallel detection branches are connected to the image segmentation layer to form an intelligent detection module. The image segmentation layer is used to segment the input image into multiple regions based on the distributed point cloud, each region corresponding to an object or part in the image, so that each hole can be processed separately.
[0084] In some implementations, the convergence verification method includes inter-class verification and cross-class verification;
[0085] If the inter-class check result is greater than or equal to the first preset threshold, and the class cross-check result is less than the second preset threshold, the convergence condition is met;
[0086] Taking the hole area as an image recognition constraint, incremental learning is performed on the multiple parallel detection branches.
[0087] Among them, inter-class verification is a method for evaluating model performance in supervised learning. By checking the recognition accuracy of the model for samples of the same category, it is ensured that the model can well distinguish the same type of holes. Class cross-verification is a method for evaluating the generalization ability of the model in supervised learning. By checking the model's ability to distinguish samples of different categories, it is ensured that the model can effectively distinguish different types of holes and avoid misidentification.
[0088] Optionally, a first preset threshold and a second preset threshold are set to determine whether the convergence condition is met. The first preset threshold is used to determine the upper limit of the accuracy of the same type of analysis, and the second preset threshold is used to determine the lower limit of the non-same type of analysis. If the inter-class verification result is greater than or equal to the first preset threshold, and the class cross-verification result is less than the second preset threshold, the convergence condition is met. At this time, multiple parallel detection branches can meet the accuracy requirements of different categories of detection tasks, ensuring the pertinence and personalization of multiple parallel detection branches.
[0089] Optionally, in order to improve recognition efficiency and reduce unnecessary consumption of computing resources, the hole area is used as a constraint for image recognition. In other words, only the area related to the hole in the image is focused on, rather than the entire image. Only the image within the hole coverage area will be used for incremental learning, avoiding recognition of the entire image, thereby reducing redundant processing, improving processing efficiency, and improving recognition accuracy.
[0090] In some embodiments, after the matching target detection branch performs parallel detection, the following steps are included:
[0091] Traversing the positioning point cloud set, and determining a detection multi-tuple in combination with a quality detection standard;
[0092] Taking the detection dimension as the matrix row and the indicator feature as the matrix column, the detection multi-element group is subjected to matrix transformation to determine the detection matrix, wherein the detection matrix includes N, where N is a positive integer less than or equal to the total number of wells;
[0093] Execute mapping between the detection matrix and the positioning point cloud set, and store the mapping in a temporary database;
[0094] A connection is established between the temporary database and the plurality of parallel detection branches.
[0095] Optionally, the positioning point cloud set is traversed, and the multi-tuple that needs to be inspected is determined in combination with the quality inspection standard to determine the inspection multi-tuple. The inspection multi-tuple includes multiple inspection dimensions of the hole position, such as the hole size, edge clarity, hole integrity, etc.
[0096] Optionally, each detection tuple is converted into a matrix form to form a detection matrix, which is a two-dimensional data structure, in which each row represents the detection dimension of a hole position and each column represents an indicator feature. Each detection matrix contains the detection data of N holes, where N is a positive integer less than or equal to the total number of holes. Among them, the indicator feature is a specific parameter used to describe the detection dimension, such as diameter, roundness, coordinates, etc. The tuple data is converted into a matrix form according to certain rules to facilitate the subsequent structured management of the data.
[0097] Optionally, a mapping is performed between the detection matrix and the positioning point cloud set, the data spaces in the detection matrix are matched with the coordinates in the positioning point cloud set, and the results are stored in a temporary database for subsequent processing and analysis.
[0098] Optionally, a connection is established between the temporary database and multiple parallel detection branches. Each detection branch can access the data in the temporary database and further analyze and process it, thereby achieving efficient and accurate detection of the hole position. In this way, not only the accuracy and efficiency of the detection can be improved, but also different types of hole position detection requirements can be met, with high flexibility and practicality.
[0099] The limit detection result and the image detection result are fitted as the hole position detection result.
[0100] Optionally, the limit detection result is combined with the image detection result to obtain the final hole position detection result. By combining the advantages of physical measurement and image processing, the physical accuracy of the hole position can be guaranteed and detailed information of the hole position can be provided. High-precision and high-efficiency hole position detection can be achieved.
[0101] In summary, the intelligent detection method for hole position image recognition of automotive parts provided by the present invention has the following technical effects:
[0102] By identifying the geometric features of the target parts, the positioning point cloud set is identified. The positioning point cloud set includes distributed point clouds and edge point clouds, and there are point cloud trajectories and class attribution marks under spatial angle conversion; combined with the positioning point cloud set, the self-positioning module is determined, and the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module; the limit detection scheme based on the hole detection tooling is determined, and the pre-collection strategy of the spatial perspective of the scheme is obtained in combination with the front positioner; the pre-collection strategy responds to the multi-camera system to perform image acquisition control of the camera group and obtain the acquired image set; combined with the rear positioner, the positioning point cloud recognition mark of the acquired image set is executed, and in the intelligent detection module, hole mark recognition and class segmentation are performed, and the target detection branch is matched for parallel detection to determine the image detection result. The intelligent detection module includes multiple parallel detection branches mapped to the hole type; the limit detection result and the image detection result are fitted as the hole position detection result. Thus, the technical effect of improving the accuracy and efficiency of multi-hole position detection is achieved.
[0103] Embodiment 2
[0104] Based on the same concept as the intelligent detection method for hole position image recognition of a vehicle component in the embodiment, Figure 2 As shown, the present application also provides an intelligent detection method system for image recognition of hole positions of automotive parts, the system comprising:
[0105] A geometric feature recognition component 11, which is used to recognize the geometric features of the target parts and identify the positioning point cloud set, which includes a distribution point cloud and an edge point cloud, and has a point cloud trajectory and a class attribution mark under spatial angle conversion;
[0106] A self-positioning component 12, the self-positioning component is used to determine a self-positioning module in combination with the positioning point cloud set, the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module;
[0107] A limit detection scheme determination component 13, which is used to determine a limit detection scheme based on a hole detection tooling, and obtain a pre-collection strategy of the scheme space perspective in combination with the front locator;
[0108] An image acquisition control component 14, wherein the image acquisition control component is used for the pre-acquisition strategy to respond to the multi-camera system, perform image acquisition control on the camera group, and obtain an acquired image set;
[0109] A positioning point cloud recognition component 15, which is used to perform positioning point cloud recognition marking of the collected image set in combination with a post-locator, perform hole mark recognition and class segmentation in an intelligent detection module, match the target detection branch for parallel detection, and determine the image detection result. The intelligent detection module includes a plurality of parallel detection branches mapped to the hole type;
[0110] The limit detection result fitting component 16 is used to fit the limit detection result and the image detection result as the hole position detection result.
[0111] Among them, the geometric feature recognition component 11 includes:
[0112] A distribution point cloud determination unit, used to traverse the geometric features, take the hole center as a positioning reference, and determine the distribution point cloud in combination with the relative position of the hole space;
[0113] An edge point cloud acquisition unit is used to traverse the distributed point cloud, determine the edge positioning points of the hole position point by point cloud, and acquire the edge point cloud, wherein the number of edge positioning points is a fixed value;
[0114] The positioning point cloud set determining unit is used to fuse the distribution point cloud and the edge point cloud to determine the positioning point cloud set.
[0115] Furthermore, the positioning point cloud determination unit in the geometric feature recognition component 11 further includes:
[0116] A cluster processing unit, used to perform cluster processing on the distribution point cloud in combination with multiple hole position types, and to perform hole position classification marking in combination with cluster clusters;
[0117] A point cloud dynamic analysis unit is used to determine an initial angle point cloud based on the fused distributed point cloud and the edge point cloud, perform point cloud dynamic analysis at the full spatial angle of the target component, and determine the position space trajectory of each point cloud as the point cloud trajectory;
[0118] The positioning point cloud set determination unit is used to fuse the class attribution mark, the point cloud trajectory and the initial angle point cloud to determine the positioning point cloud set.
[0119] Furthermore, the limit detection scheme determination component 13 also includes:
[0120] A limit detection scheme determination unit, used to determine a single limit detection scheme in combination with the front positioner, traverse the positioning point cloud, and combine the working mechanism of the detection rod and the spatial positioning structure;
[0121] A collaborative analysis unit, used to perform collaborative analysis on the single position limit detection scheme in combination with the relative position of the detection rod, and determine the position limit detection scheme;
[0122] The hole detection tool control unit is used to control the hole detection tool in combination with the limit detection scheme, perform limit detection on the target component, and determine the limit detection result, wherein the detection dimensions include hole diameter, hole depth trend and position.
[0123] Furthermore, the positioning point cloud recognition component 15 also includes:
[0124] A branch training data acquisition unit, used to acquire branch training data, wherein the branch training data corresponds to the hole type one by one;
[0125] A feature extraction and learning rate setting unit, used to extract common features and individual features based on the hole type, and set the learning rate for the branch training data, wherein the learning rate is positively correlated with the individual degree, and the learning rate can be iteratively adjusted;
[0126] A supervised learning and convergence condition verification unit, used to perform supervised learning and convergence condition verification based on the branch training data after the learning rate mark, and generate the multiple parallel detection branches;
[0127] The intelligent detection module construction unit is used to pre-connect the multiple parallel detection branches to the image segmentation layer to generate the intelligent detection module.
[0128] Furthermore, the positioning point cloud recognition component 15 also includes:
[0129] A detection element determination unit, used for traversing the positioning point cloud set and determining a detection multi-element group in combination with a quality detection standard;
[0130] A matrix conversion unit, used to perform matrix conversion on the detection multi-tuple with detection dimensions as matrix rows and indicator features as matrix columns to determine a detection matrix, wherein the detection matrix includes N, where N is a positive integer less than or equal to the total number of wells;
[0131] A positioning point cloud mapping unit, used to perform mapping between the detection matrix and the positioning point cloud set, and store the mapping in a temporary database;
[0132] A branch connection unit is used to establish a connection between the temporary database and the multiple parallel detection branches.
[0133] Furthermore, the positioning point cloud recognition component 15 also includes:
[0134] A dual threshold determination unit, configured to satisfy the convergence condition if the inter-class check result is greater than or equal to a first preset threshold and the class cross-check result is less than a second preset threshold;
[0135] The incremental learning unit is used to perform incremental learning on the multiple parallel detection branches by taking the hole position area as an image recognition constraint.
[0136] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the intelligent detection system for image recognition of vehicle component holes described in embodiment two. For the sake of brevity of the specification, they will not be further elaborated here.
[0137] It should be understood that the embodiments disclosed in this application and the above description can enable those skilled in the art to use this application to implement this application. At the same time, this application is not limited to the above-mentioned embodiments, and obvious modifications, combinations and substitutions of the embodiments mentioned in this application also fall within the scope of protection of this application.
Claims
1. An intelligent detection method for image recognition of hole positions of automotive parts, characterized in that: The method comprises: Identify the geometric features of the target parts and mark the positioning point cloud set, which includes a distribution point cloud and an edge point cloud, and has point cloud trajectories and class attribution marks under spatial angle conversion; In combination with the positioning point cloud set, a self-positioning module is determined, wherein the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module; Determine a limit detection scheme based on the hole detection fixture, and obtain a pre-collection strategy of the spatial perspective of the scheme in combination with the front locator; The pre-capture strategy performs image capture control of the camera group in response to the multi-camera system to obtain a captured image set; Combined with the post-locator, the positioning point cloud recognition mark of the collected image set is executed, and in the intelligent detection module, hole mark recognition and class segmentation are performed, and the matching target detection branch is used for parallel detection to determine the image detection result. The intelligent detection module includes multiple parallel detection branches mapped to the hole type; Fitting the limit detection result and the image detection result as the hole position detection result; The determination of the limit detection scheme based on the hole detection tooling includes: In combination with the front positioner, the positioning point cloud is traversed, and a single limit detection scheme is determined in combination with the operating mechanism of the detection rod and the spatial positioning structure; Combined with the relative position of the detection rod, the single limit detection scheme is collaboratively analyzed to determine the limit detection scheme; In combination with the limit detection scheme, the hole detection tooling is controlled to perform limit detection on the target component to determine the limit detection result, wherein the detection dimensions include hole diameter, hole depth trend and position.
2. The method according to claim 1, characterized in that The identification positioning point cloud set includes: Traversing the geometric features, taking the hole center as a positioning reference, and combining the relative position of the hole space, determining the distribution point cloud; Traversing the distributed point cloud, determining the edge positioning points of the hole position point by point cloud, and obtaining the edge point cloud, wherein the number of edge positioning points is a fixed value; The distributed point cloud and the edge point cloud are fused to determine the positioning point cloud set.
3. The method according to claim 2, characterized in that The fusing the distribution point cloud and the edge point cloud to determine the positioning point cloud set includes: In combination with the multivariate hole position types, the distribution point cloud is clustered, and the hole position classification is marked in combination with the cluster clusters; Based on the fused distributed point cloud and the edge point cloud, an initial angle point cloud is determined, point cloud dynamic analysis is performed at the full spatial angle of the target component, and the position space trajectory of each point cloud is determined as the point cloud trajectory; The class attribution mark, the point cloud trajectory and the initial angle point cloud are fused to determine the positioning point cloud set.
4. The method according to claim 1, characterized in that The hole mark recognition and class segmentation include: Acquire branch training data, where the branch training data corresponds one-to-one to the hole type; Extracting common features and individual features based on the hole type, setting a learning rate for the branch training data, wherein the learning rate is positively correlated with the individual degree, and the learning rate can be iteratively adjusted; Based on the branch training data after the learning rate labeling, supervised learning and convergence condition verification are performed to generate the multiple parallel detection branches; The multiple parallel detection branches are pre-connected to an image segmentation layer to generate the intelligent detection module.
5. The method according to claim 4, characterized in that Convergence verification methods include inter-class verification and cross-class verification; If the inter-class check result is greater than or equal to the first preset threshold, and the class cross-check result is less than the second preset threshold, the convergence condition is met; Taking the hole area as an image recognition constraint, incremental learning is performed on the multiple parallel detection branches.
6. The method according to claim 1, characterized in that After the matching target detection branch performs parallel detection, it includes: Traversing the positioning point cloud set, and determining a detection multi-tuple in combination with a quality detection standard; Taking the detection dimension as the matrix row and the indicator feature as the matrix column, the detection multi-element group is subjected to matrix transformation to determine the detection matrix, wherein the detection matrix includes N, where N is a positive integer less than or equal to the total number of wells; Execute mapping between the detection matrix and the positioning point cloud set, and store the mapping in a temporary database; A connection is established between the temporary database and the plurality of parallel detection branches.
7. An intelligent detection system for image recognition of hole positions of automotive parts, characterized in that: The system comprises: A geometric feature recognition component, which is used to recognize the geometric features of the target parts and identify the positioning point cloud set, which includes a distribution point cloud and an edge point cloud, and has point cloud trajectories and class attribution marks under spatial angle conversion; A self-positioning component, the self-positioning component is used to determine a self-positioning module in combination with the positioning point cloud set, the self-positioning module includes a front positioner connected to the multi-camera system and a rear positioner connected to the intelligent detection module; A limit detection scheme determination component, the limit detection scheme determination component is used to determine the limit detection scheme based on the hole detection tooling, and in combination with the front locator, obtain a pre-collection strategy of the scheme space perspective; An image acquisition control component, wherein the image acquisition control component is used for the pre-acquisition strategy to respond to the multi-camera system, perform image acquisition control on the camera group, and obtain an acquired image set; A positioning point cloud recognition component, which is used to perform positioning point cloud recognition marking of the collected image set in combination with a post-locator, perform hole mark recognition and class segmentation in an intelligent detection module, match the target detection branch for parallel detection, and determine the image detection result. The intelligent detection module includes a plurality of parallel detection branches mapped to the hole type; A limit detection result fitting component, wherein the limit detection result fitting component is used to fit the limit detection result and the image detection result as a hole position detection result; The limit detection scheme determination component also includes: A limit detection scheme determination unit, used to determine a single limit detection scheme in combination with the front positioner, traverse the positioning point cloud, and combine the working mechanism of the detection rod and the spatial positioning structure; A collaborative analysis unit, used to perform collaborative analysis on the single position limit detection scheme in combination with the relative position of the detection rod, and determine the position limit detection scheme; The hole detection tool control unit is used to control the hole detection tool in combination with the limit detection scheme, perform limit detection on the target component, and determine the limit detection result, wherein the detection dimensions include hole diameter, hole depth trend and position.
