Vehicle paint surface recognition system based on AI and data analysis

Abstract

This invention discloses a vehicle paint surface recognition system based on AI and data analysis, belonging to the field of vehicle paint surface recognition technology. It includes a data debugging module, a data acquisition module, and a paint surface recognition module. The data debugging module is used for data debugging of vehicle paint surface recognition, identifying camera information of the target vehicle based on vehicle information, and performing data acquisition and debugging based on the camera information. It generates a paint surface reference model of the target vehicle based on vehicle information, camera information, and paint surface information. The data acquisition module is used for real-time data acquisition, obtaining monitoring images of the target vehicle's paint surface, adding corresponding image timestamps to the monitoring images, and sending the monitoring images to the paint surface recognition module. The paint surface recognition module analyzes the received monitoring images to obtain paint surface recognition results and confidence values ​​for various parts of the target vehicle's paint surface. It then supplements the paint surface recognition results and confidence values ​​into the paint surface reference model and displays the paint surface reference model to the user in real time.

Description

Technical Field

[0001] This invention belongs to the field of vehicle paint surface recognition technology, specifically a vehicle paint surface recognition system based on AI and data analysis. Background Technology

[0002] With the rapid development of the new energy vehicle industry, new energy electric vehicles have gradually become the mainstream in the market. These vehicles not only have advantages such as environmental protection and energy saving, but are also generally equipped with advanced multi-camera equipment for various functions such as driver assistance, automatic parking, and driving recording. These camera devices not only improve the safety and convenience of the vehicles, but also provide new technical means for vehicle paint surface recognition.

[0003] Traditional methods for identifying vehicle paint surfaces primarily rely on manual visual inspection or the use of specialized testing equipment. Manual inspection suffers from issues such as subjectivity, low efficiency, and high costs, and cannot provide real-time detection of the vehicle's paint surface. While specialized testing equipment can provide more accurate results, it is difficult to monitor a user's vehicle in real time; professional inspections are generally only possible at designated locations. Summary of the Invention

[0004] To address the problems of the aforementioned solutions, this invention provides a vehicle paint surface recognition system based on AI and data analysis to solve the problem that existing vehicle paint surfaces cannot be monitored in real time.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A vehicle paint surface recognition system based on AI and data analysis, including a data debugging module, a data acquisition module, and a paint surface recognition module;

[0007] The data debugging module is used to debug the data for vehicle paint surface recognition, mark the user's vehicle as the target vehicle, obtain the vehicle information of the target vehicle, identify the camera information of the target vehicle based on the vehicle information, and perform data collection and debugging based on the camera information.

[0008] The paint surface information of the target vehicle is obtained, and a paint surface reference model of the target vehicle is generated based on the vehicle information, camera information and paint surface information. The paint surface reference model is a three-dimensional data model of the paint surface of the target vehicle.

[0009] Furthermore, a paint reference model of the target vehicle is generated based on vehicle information, camera information, and paint surface information, including:

[0010] An initial paint surface model of the target vehicle is generated based on vehicle information and paint surface information. The initial paint surface model is a three-dimensional data model. Camera information is then added to the corresponding areas of the initial paint surface model.

[0011] Based on the initial paint surface model, an identification and analysis is performed to obtain the target identification method for the paint surface in the corresponding location area of ​​the initial paint surface model; based on the target identification method, the image quality of the corresponding location in the initial paint surface model is numerically quantified to obtain the quality influence value corresponding to the location;

[0012] Associate the quality impact value and target identification method with the corresponding location area in the initial paint surface model; mark the current initial paint surface model as the paint surface reference model.

[0013] Furthermore, identification and analysis are performed based on the initial paint surface model, including:

[0014] A number of candidate recognition methods for identifying vehicle paint surfaces are obtained; paint surface image materials corresponding to each position area in the initial paint surface model are collected based on camera information; the paint surface image materials are simulated for recognition according to the candidate recognition methods to obtain the corresponding simulated recognition results; the simulated recognition results are compared and analyzed with the standard recognition results of the paint surface image materials to obtain the recognition accuracy of the candidate recognition methods for each position area in the initial paint surface model;

[0015] The priority value of the candidate recognition method is calculated based on the recognition accuracy of each location area, and the target recognition method of the paint surface in the corresponding location area in the initial paint surface model is determined based on the priority value.

[0016] Furthermore, the calculation of priority values ​​includes:

[0017] Determine the priority analysis region, identify the area of ​​the priority analysis region, and mark the area of ​​the priority analysis region as the analysis area; identify the recognition accuracy of each position within the priority analysis region, and merge adjacent positions within the priority analysis region with the same recognition accuracy into one region, which is marked as a unit region;

[0018] Label the unit region as i, i = 1, 2, ..., n, where n is the number of unit regions in the priority analysis area; identify the area of ​​the unit region and label it as the unit area;

[0019] The priority value of the corresponding candidate recognition method is calculated according to the priority formula, which is as follows:

[0020]

[0021] In the formula: YA is the priority value; DA i This represents the cell area of ​​the corresponding cell region, where AF is the area under analysis; BL i This indicates the recognition accuracy of the corresponding unit region.

[0022] Furthermore, the calculation of priority values ​​includes:

[0023] Obtain the probability of anomalies on the paint surface of the target vehicle at various locations, determine the priority analysis area, identify the area of ​​the priority analysis area, and mark it as the analysis area;

[0024] Within the priority analysis area, adjacent locations with the same recognition accuracy and anomaly probability are merged into one area and marked as an adjustment area. The adjustment area is labeled j, j = 1, 2, ..., m, where m is the number of adjustment areas within the priority analysis area; the area of ​​the adjustment area is identified and marked as the adjustment area.

[0025] The priority value of the corresponding candidate recognition method is calculated according to the priority formula, which is as follows:

[0026]

[0027] In the formula: YB is the priority value; DB j This indicates the adjusted area of ​​the corresponding adjustment region, where AF is the analysis area; BT j This indicates the recognition accuracy of the corresponding adjusted area.

[0028] The acquisition module is used to acquire real-time data, obtain monitoring images of the target vehicle's paint surface, add corresponding image timestamps to the monitoring images, and send the monitoring images to the paint surface recognition module.

[0029] The paint surface recognition module is used to analyze the received monitoring images to obtain the paint surface recognition results and confidence values ​​of various parts of the target vehicle; the paint surface recognition results and confidence values ​​are added to the paint surface reference model, and the paint surface reference model is displayed to the user in real time.

[0030] Furthermore, the monitoring images are analyzed, including:

[0031] The paint surface and the paint surface reference model in the monitoring image are located and calibrated. The target recognition method corresponding to the paint surface reference model is identified. The paint surface area corresponding to the target recognition method is summarized. The monitoring image is segmented according to the paint surface area to obtain the monitoring and analysis image of the target recognition method.

[0032] The monitoring and analysis image is identified and analyzed using the target recognition method to obtain the paint surface recognition results for each paint surface.

[0033] The paint surface reference model is used to match the corresponding quality influence value for the paint surface recognition result, and the trust value of the paint surface recognition result is calculated based on the quality influence value.

[0034] Furthermore, the formula for calculating the trustworthiness value is as follows:

[0035] KR = (1-LZ) × 100;

[0036] In the formula: KR is the trust value; LZ is the quality impact value.

[0037] Furthermore, based on the quality impact value, a confidence value for the paint surface identification result is calculated, including:

[0038] A dynamic correction model is established, and the paint surface recognition results, quality impact values, and paint surface change images are analyzed in real time based on the dynamic correction model to obtain the trust correction value at the corresponding time.

[0039] The trust correction value is matched to the paint surface recognition result based on the image time corresponding to the monitored image.

[0040] The corresponding trust value is calculated according to the trust value calculation formula, which is as follows:

[0041] KR = (1 - LZ + DZ) × 100;

[0042] In the formula: KR is the trust value; LZ is the quality impact value; DZ is the trust correction value.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] This invention fully utilizes the vehicle's built-in camera equipment, eliminating the need for additional hardware installations and thus reducing the overall system cost. Furthermore, intelligent analysis algorithms process the collected data, minimizing the need for manual intervention and further reducing labor costs. In addition, the system enables real-time monitoring of the vehicle's paint, reducing resource consumption associated with periodic inspections. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a block diagram illustrating the principle of the present invention. Detailed Implementation

[0047] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] like Figure 1As shown, the vehicle paint surface recognition system based on AI and data analysis includes a data debugging module, a data acquisition module, and a paint surface recognition module.

[0049] The data debugging module is used for data debugging of vehicle paint surface recognition. It acquires the user's vehicle information, such as vehicle brand, model, and configuration, and marks the user's vehicle as the target vehicle. Based on the vehicle information, it identifies the camera information of the target vehicle, that is, the information of the camera equipment configured on the target vehicle that can capture images of the target vehicle's paint surface, such as model, acquisition range, image quality, data storage location, and other related information, including the camera equipment built into the target vehicle and the camera equipment added by the user later. Based on the camera information, it performs data acquisition debugging so that the subsequent acquisition module can acquire the monitoring images of the corresponding camera equipment in real time.

[0050] Acquire the paint information of the target vehicle, including paint color, type, process, performance, and other related information; generate an initial paint model of the target vehicle based on the vehicle information and paint information. The initial paint model is a three-dimensional data model, which is set using 3D visualization technology; supplement the initial paint model with information based on camera information, such as supplementing the camera equipment, data location, image quality, etc. corresponding to the paint area.

[0051] Based on the initial paint surface model, identification and analysis are performed to determine the target recognition method for the paint surface in the corresponding regions of the initial paint surface model. The image quality at the corresponding location in the initial paint surface model is quantified according to the target recognition method to obtain the corresponding quality influence value. The quality influence value represents the error of the image quality at that location in applying the corresponding target recognition method to paint surface recognition. A quality influence value is set based on the corresponding error to adjust the confidence value of the subsequent paint surface recognition results. For example, if the confidence value is 0.9 and the quality influence value is 0.05, the adjusted confidence value range is 0.85-0.9. A confidence value within a certain range can be selected as a representative value, or the confidence value range can be displayed. The quality influence value and the target recognition method are associated with the corresponding regions in the initial paint surface model, so that the corresponding quality influence value and target recognition method can be identified based on the corresponding location in the initial paint surface model. The current initial paint surface model is marked as the paint surface baseline model.

[0052] In one embodiment, identification analysis based on an initial paint surface model includes:

[0053] Acquire existing recognition methods for vehicle paint surface recognition based on images, such as those based on AI, deep learning, and other technologies, and mark them as candidate recognition methods. If the user has developed its own recognition method, it will also be considered a candidate recognition method.

[0054] Historical monitoring images of the target vehicle's paint surface are labeled as paint surface image materials. Each paint surface image material corresponds to a specific paint surface recognition result, labeled as a standard recognition result, such as a scratch, crack, or normal condition. Paint surface image materials corresponding to various locations in the initial paint surface model are collected based on camera information. The paint surface image materials are then simulated for recognition using the candidate recognition method to obtain corresponding simulated recognition results. These simulated recognition results are compared and analyzed with the standard recognition results of the paint surface image materials to obtain the recognition accuracy of the candidate recognition method for each location in the initial paint surface model. The recognition accuracy is calculated by comparing the number of normal results with the number of simulations.

[0055] The priority value of the corresponding candidate recognition method is determined based on the recognition accuracy of each location region. The target recognition method of the paint surface in the corresponding location region in the initial paint surface model is determined based on the priority value. That is, the larger the priority value, the higher the priority.

[0056] In one embodiment, the priority value of the corresponding candidate recognition method is determined based on the recognition accuracy of each location region, with the recognition accuracy of the corresponding location region serving as the priority value.

[0057] In one embodiment, the priority value of the corresponding candidate recognition method is determined based on the recognition accuracy of each location area. If we consider the whole picture, that is, one camera device corresponds to one target recognition method, or a certain area is combined to correspond to one target recognition method, then the priority value is calculated comprehensively based on the recognition accuracy of each location within the corresponding area.

[0058] For example, a priority analysis region is determined, and its area is identified and marked as the analysis area. This area can be the camera area corresponding to the camera device, or it can be a region obtained by merging according to regional requirements and clustering algorithms. The recognition accuracy of each position within the priority analysis region is identified, and adjacent positions with the same recognition accuracy within the priority analysis region are merged into one region and marked as a unit region. That is, the priority analysis region consists of several unit regions, and the unit regions are marked as i, i = 1, 2, ..., n, where n is the number of unit regions within the priority analysis region. The area of ​​the unit region is identified and marked as the unit area.

[0059] The priority value of the corresponding candidate recognition method is calculated according to the priority formula, which is as follows:

[0060]

[0061] In the formula: YA is the priority value; DA i This represents the cell area of ​​the corresponding cell region, where AF is the area under analysis; BL i This indicates the recognition accuracy of the corresponding unit region.

[0062] In one embodiment, because the probability of scratches, cracks, and other abnormalities appearing on the paint surface of different areas of the vehicle varies during use, the priority value is calculated as follows in this embodiment:

[0063] Based on the vehicle information of the target vehicle, the probability of abnormality of the paint surface of each part of the target vehicle is counted to determine the priority analysis area. The area of ​​the priority analysis area is identified and marked as the analysis area. The positions that are adjacent in the priority analysis area and have the same recognition accuracy and abnormality probability are merged into one area and marked as the adjustment area. The adjustment area is marked as j, j = 1, 2, ..., m, where m is the number of adjustment areas in the priority analysis area. The area of ​​the adjustment area is identified and marked as the adjustment area.

[0064] The priority value of the corresponding candidate recognition method is calculated according to the priority formula, which is as follows:

[0065]

[0066] In the formula: YB is the priority value; DB j This indicates the adjusted area of ​​the corresponding adjustment region, where AF is the analysis area; BT j This indicates the recognition accuracy of the corresponding adjusted area.

[0067] In one embodiment, the image quality of the corresponding position in the initial model of the paint surface is quantified according to the target recognition method to obtain the recognition accuracy of the target recognition method in the recognition simulation process, and the quality influence value is determined according to the recognition accuracy, that is, the quality influence value = 1 - recognition accuracy.

[0068] In one embodiment, the image quality at the corresponding location in the initial model of the paint surface is quantified according to the target recognition method, or it can be quantified based on existing methods.

[0069] The acquisition module is used to acquire real-time data, obtain monitoring images of the target vehicle's paint surface, and add a corresponding image time to the monitoring image, which is the time when the monitoring image was generated. The monitoring image is then sent to the paint surface recognition module.

[0070] The paint surface recognition module is used to analyze the received monitoring images to obtain the paint surface recognition results and confidence values ​​of various parts of the target vehicle; the paint surface recognition results and confidence values ​​are added to the paint surface reference model, and the paint surface reference model is displayed to the user in real time.

[0071] In one embodiment, the received monitoring image is analyzed according to the target recognition method corresponding to the paint surface location to obtain the paint surface recognition result, and the confidence value is determined based on the quality influence value.

[0072] In one embodiment, as various situations such as vehicle movement, washing, and rain occur, the target recognition result can be corrected from multiple aspects to increase its confidence value; that is, the time span is extended and the confidence value is dynamically updated, as follows:

[0073] The paint surface in the monitoring image is located and calibrated against the paint surface reference model to establish a one-to-one correspondence. The monitoring image is then segmented according to the target recognition method corresponding to the paint surface reference model to form a monitoring analysis image corresponding to the target recognition method, that is, the paint surface area is segmented according to the target recognition method. The monitoring analysis image is then identified and analyzed through the target recognition method to obtain the paint surface recognition results for each location. The corresponding quality influence value is then matched to the paint surface recognition results based on the paint surface reference model.

[0074] A dynamic correction model is established based on neural networks such as CNN or DNN. This model analyzes paint surface recognition results, quality impact values, and paint surface change images over a continuous time period to determine the corresponding trust correction value. A training set is manually created for training. The training set includes input and output data: the input data consists of paint surface recognition results, quality impact values, and paint surface change images over a continuous time period; the output data is the trust correction value, which is no greater than or equal to the quality impact value. The successfully trained dynamic correction model is then analyzed in real time to obtain the corresponding trust correction value.

[0075] The formula for calculating the trust value is:

[0076] KR = (1 - LZ + DZ) × 100;

[0077] In the formula: KR is the trust value; LZ is the quality impact value; DZ is the trust correction value.

[0078] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.

[0079] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An AI and data analysis based vehicle paint identification system, characterized in that, The data debugging module, the acquisition module, and the paint surface identification module are included. The data debugging module is configured to debug data for vehicle paint surface identification, mark a user vehicle as a target vehicle, acquire vehicle information of the target vehicle, identify camera information of the target vehicle according to the vehicle information, and perform data acquisition debugging according to the camera information. The paint surface information of the target vehicle is acquired, and a paint surface reference model of the target vehicle is generated according to the vehicle information, the camera information, and the paint surface information. The paint surface reference model is a three-dimensional data model of the paint surface of the target vehicle. The acquisition module is configured to perform real-time data acquisition, obtain a monitoring image of the paint surface of the target vehicle, mark the monitoring image with a corresponding image time, and send the monitoring image to the paint surface identification module. The paint surface identification module is configured to analyze the received monitoring image, and obtain a paint surface identification result and a trust value of each part of the paint surface of the target vehicle. The paint surface identification result and the trust value are supplemented into the paint surface reference model, and the paint surface reference model is displayed in real time to the user. The analysis of the monitoring image includes: The paint surface in the monitoring image is positioned and calibrated with the paint surface reference model, the target identification mode corresponding to the paint surface reference model is identified, the paint surface area corresponding to the target identification mode is summarized, the monitoring image is segmented according to the paint surface area, and a monitoring analysis image of the target identification mode is obtained. The monitoring analysis image is identified and analyzed by the target identification mode, and a paint surface identification result of each part of the paint surface is obtained. The corresponding quality influence value is matched for the paint surface identification result according to the paint surface reference model, and the trust value of the paint surface identification result is calculated according to the quality influence value. The calculation of the trust value of the paint surface identification result according to the quality influence value includes: A dynamic correction model is established, the paint surface identification result, the quality influence value, and the paint surface change image are analyzed in real time according to the dynamic correction model, and a trust correction value of the corresponding time is obtained. The corresponding trust correction value is matched for the paint surface identification result according to the image time corresponding to the monitoring image. The corresponding trust value is calculated according to the trust calculation formula, and the trust calculation formula is: KR=(1-LZ+DZ)×100; In the formula, KR is the trust value, LZ is the quality influence value, and DZ is the trust correction value. 2.The AI and data analysis based vehicle paint identification system according to claim 1, wherein, The generation of the paint surface reference model of the target vehicle according to the vehicle information, the camera information, and the paint surface information includes: A paint surface initial model of the target vehicle is generated according to the vehicle information and the paint surface information. The paint surface initial model is a three-dimensional data model. The camera information is supplemented into the corresponding position area of the paint surface initial model. The target identification mode of the paint surface in the corresponding position area of the paint surface initial model is obtained by identification analysis according to the paint surface initial model. The image quality of the corresponding position of the paint surface initial model is numerically valued according to the target identification mode, and the corresponding quality influence value of the corresponding position is obtained. The quality influence value and the target identification mode are associated with the corresponding position area of the paint surface initial model. The current paint surface initial model is marked as the paint surface reference model. 3.The AI and data analytics based vehicle paint identification system of claim 2, wherein, The identification analysis according to the paint surface initial model includes: Obtaining several candidate identification modes for identifying vehicle paint; collecting paint image materials corresponding to each position area in the initial paint model according to camera information; identifying the paint image materials according to the candidate identification modes to obtain corresponding simulation identification results; comparing and analyzing the simulation identification results with standard identification results of the paint image materials to obtain identification accuracy of each position area in the initial paint model by the candidate identification modes; Calculating priority values of the candidate identification modes according to the identification accuracy of each position area, and determining target identification modes of the paint in the corresponding position area in the initial paint model according to the priority values. 4.The AI and data analysis based vehicle paint identification system of claim 3, wherein, The calculation of the priority values includes: Determining a priority analysis area, identifying an area of the priority analysis area, and marking the area of the priority analysis area as an analysis area; identifying identification accuracy of each position in the priority analysis area, merging positions adjacent to each other and having the same identification accuracy in the priority analysis area into one area, and marking the area as a unit area; Marking the unit area as i, i = 1, 2, …, n, n being the number of unit areas in the priority analysis area; identifying an area of the unit area and marking the area as a unit area; Calculating the priority value of the corresponding candidate identification mode according to a priority formula, the priority formula being: ； In the formula, YA is a priority value; DA i represents a unit area of a corresponding unit region, and AF is an analysis area; BL i represents a recognition accuracy of the corresponding unit region. 5.The AI and data analytics based vehicle paint identification system according to claim 3, wherein, The calculation of the priority values includes: Obtaining an abnormal probability of the target vehicle at each paint, determining a priority analysis area, identifying an area of the priority analysis area, and marking the area as an analysis area; Merging positions adjacent to each other and having the same identification accuracy and abnormal probability in the priority analysis area into one area, marking the area as an adjustment area, marking the adjustment area as j, j = 1, 2, …, m, m being the number of adjustment areas in the priority analysis area; identifying an area of the adjustment area and marking the area as an adjustment area; Calculating the priority value of the corresponding candidate identification mode according to a priority formula, the priority formula being: ； In the formula, YB is a priority value; DB j represents an adjustment area of the corresponding adjustment region, and AF is an analysis area; BT j represents a recognition accuracy of the corresponding adjustment region. 6.The AI and data analytics based vehicle paint identification system according to claim 1, wherein, The camera information includes camera equipment information added by a user for monitoring the paint.

Images