Search

CN-121999267-A - Method for processing failure of vehicle

CN121999267ACN 121999267 ACN121999267 ACN 121999267ACN-121999267-A

Abstract

The application provides a fault handling method for a vehicle. The method comprises the steps of acquiring a vehicle image set, acquiring target fault detection models corresponding to stations, splicing the acquired vehicle images in the vehicle image set to obtain spliced result data, screening the target fault detection models corresponding to the stations according to the stations by the vehicle image set to perform fault detection on the vehicle image of the stations to obtain station fault detection results, correspondingly obtaining corresponding station fault detection results by each station, processing the vehicle image of the corresponding station by the fault detection models of different stations to obtain station fault detection results which are more accurate, predicting the performance of the vehicle by means of outlier evolution prediction processing according to the station fault detection results of each station in a historical time period, outputting accurate vehicle performance prediction results, and enabling users to timely check, recheck and repair the station fault detection results and the performance prediction results, thereby improving the fault detection accuracy.

Inventors

  • CHEN MINGCAN

Assignees

  • 上海仁童电子科技有限公司

Dates

Publication Date
20260508
Application Date
20251208

Claims (12)

  1. 1. A fault handling method for a vehicle, comprising: Receiving a vehicle image set, splicing at least part of vehicle images in the vehicle image set by using an image splicing algorithm to obtain spliced images, and integrating the spliced images to obtain spliced result data; According to the vehicle images of each station in the vehicle image set, station fault detection is carried out on the vehicle images of the station by utilizing a target fault detection model corresponding to the station, and station fault detection results of the station are obtained and stored, wherein the vehicle images set comprise vehicle images of at least one station; Marking the station fault detection result of at least one station in the spliced result data to obtain a fault marking result; Station fault detection results of all stations in a historical time period are obtained, station fault detection results of all stations in the historical time period are subjected to outlier evolution prediction processing, and performance prediction results of vehicles are determined.
  2. 2. The method of claim 1, wherein the receiving the vehicle image set, stitching at least a portion of the vehicle images in the vehicle image set using an image stitching algorithm to obtain stitched images, and integrating the stitched images to obtain stitched result data comprises: traversing each station based on the vehicle image set, and determining a vehicle image corresponding to each station; Determining at least two vehicle images corresponding to each station, determining an image stitching sequence, and stitching the at least two vehicle images according to the image stitching sequence to obtain a stitched image of the station; In response to determining that the vehicle images of all stations are spliced, determining whether individual vehicle images which do not participate in splicing exist in the vehicle image set; In response to the absence of individual vehicle images in the set of vehicle images that do not participate in stitching, taking the final stitched image as the result data of the stitching, or And in response to the fact that the independent vehicle images which do not participate in stitching exist in the vehicle image set, integrating the independent vehicle images which do not participate in stitching with the stitched image obtained by stitching, and obtaining stitched result data.
  3. 3. The method according to claim 2, wherein the stitching at least two vehicle images according to the image stitching order to obtain a stitched image of the station includes: Turning at least two vehicle images corresponding to the station to obtain at least two turned images; Preprocessing at least two images after the overturning treatment to obtain at least two preprocessed images; Calling the equipment type of the vehicle image acquisition equipment; responding to the equipment type as the original equipment type, performing transverse splicing processing and/or longitudinal splicing processing on at least two preprocessed images through characteristic point matching or suture algorithm to obtain the spliced image of the station, or And responding to the equipment type as a hardware upgrading equipment type, and performing transverse splicing processing and/or longitudinal splicing processing on at least two preprocessed images to obtain the spliced image of the station.
  4. 4. A method according to claim 3, wherein said performing a transverse stitching process on at least two of said preprocessed images by feature point matching comprises: Extracting corresponding characteristic points from at least two preprocessed images; Determining descriptors of the feature points according to the local feature information of the feature points adjacent to each other; according to the descriptors of the feature points, carrying out association matching on the feature points of two transversely adjacent preprocessed images to obtain associated feature point pairs; Optimizing the correlated characteristic point pairs through an error optimization algorithm to obtain optimized characteristic point pairs; Determining a geometric transformation relation between two transversely adjacent preprocessed images according to the optimized characteristic point pairs, and performing alignment processing on the two transversely adjacent preprocessed images according to the geometric transformation relation to obtain aligned two transversely adjacent images; Performing transverse stitching treatment on the two aligned transverse adjacent images to obtain an initial stitched image; and carrying out fusion processing on the initial spliced images to obtain a result of transverse splicing processing.
  5. 5. A method according to claim 3, wherein said transversely stitching at least two of said preprocessed images by a stitching algorithm, comprising: Extracting corresponding characteristic points from at least two preprocessed images; Determining descriptors of the feature points according to the local feature information of the feature points adjacent to each other; according to the descriptors of the feature points, carrying out association matching on the feature points of two transversely adjacent preprocessed images to obtain associated feature point pairs; According to the image positions of the correlated feature point pairs, overlapping analysis is carried out on two adjacent pre-processed images in the transverse direction, and an overlapping area is determined; Determining a difference value of each pixel point in the overlapping region; traversing each row of pixel points of the overlapping area, determining the pixel point corresponding to the smallest difference value as an optimal stitching point, and combining the optimal stitching points of each row to obtain an optimal stitching line; And stitching the two adjacent pre-processed images transversely according to the optimal stitching line to obtain a transverse stitching result.
  6. 6. The method according to claim 1, wherein according to the vehicle image of each station in the vehicle image set, performing station fault detection on the vehicle image of the station by using a target fault detection model corresponding to the station, to obtain and store a station fault detection result of the station, including: performing quality analysis processing on the vehicle images in the vehicle image set, performing station identification on the vehicle images with quality reaching the standard after the quality analysis processing, and determining a target station corresponding to the vehicle images with quality reaching the standard; A target fault detection model corresponding to the target station is called, the quality-up vehicle image is input into the target fault detection model for station fault analysis processing, at least one fault area in the quality-up vehicle image and a fault type corresponding to each fault area are determined, and the fault detection model corresponding to each station is stored in advance; carrying out confidence degree identification on each fault region, and selecting a fault region with the confidence degree larger than or equal to a confidence degree threshold value as a final fault region; And marking the final fault area and the corresponding fault type in the vehicle image with the quality reaching the standard to obtain and store a station fault detection result of the station.
  7. 7. The method according to claim 6, wherein the determining the fault detection model corresponding to each station includes: Acquiring a first sample image corresponding to the station, marking the first sample image to obtain an initial training sample, and performing image enhancement processing on the initial training sample to obtain a first training sample, wherein the marks comprise known fault type marks and/or normal marks; Constructing an initial target detection model, and performing supervised training on the initial target detection model by using the first training sample to obtain a one-stage target detection model; acquiring a second sample image of the unknown fault type corresponding to the station, and taking the second sample image as a second training sample; And adjusting the loss function of the one-stage target detection model, combining the loss function with a proximity algorithm to obtain a two-stage target detection model, and performing unsupervised training on the two-stage target detection model by using the second training sample to obtain a fault detection model corresponding to the station.
  8. 8. The method of claim 7, wherein the adjusting the loss function of the one-stage object detection model and combining with the proximity algorithm to obtain a two-stage object detection model, performing unsupervised training on the two-stage object detection model by using the second training sample to obtain a fault detection model corresponding to the station, includes: Determining semantic information of each fault type, converting the semantic information into attribute vectors to form attribute tables, evaluating the attribute tables by using a proximity algorithm to obtain reasonable attribute tables, and taking the reasonable attribute tables as target attribute tables; Adjusting the focus loss function of the one-stage target detection model to be a polar loss function; extracting a part of target detection models which generate vision related features from the one-stage target detection models, and combining the part of target detection models with a proximity algorithm to obtain two-stage target detection models; inputting the second training sample into the two-stage target detection model, performing: Processing the second training sample by using the partial target detection model to obtain visual characteristics; aligning the visual features with attribute vectors in a target attribute table obtained by the proximity algorithm, determining category similarity of the visual features and the attribute vectors, and outputting a predicted value of the second training sample belonging to a known fault type according to the category similarity and the visual features; Processing the output predicted value of the known fault type by using the polarity loss function, determining a polarity loss value, and adjusting the partial target detection model in the two-stage target detection model according to the polarity loss value; And determining that the two-stage target detection model meets the training ending condition, and taking the finally-trained two-stage target detection model as a fault detection model corresponding to the station.
  9. 9. The method according to claim 6, wherein the retrieving a target fault detection model corresponding to the target station, inputting the quality-qualified vehicle image into the target fault detection model for station fault analysis processing, and determining at least one fault region in the quality-qualified vehicle image and a fault type corresponding to each fault region, includes: obtaining a similarity matrix of the unknown fault type and the known fault type; Performing station fault detection on the vehicle image of the station by using a target fault detection model corresponding to the station, and outputting a vehicle image detection result comprising at least one boundary box, wherein the label of each boundary box corresponds to a predicted value of a known fault type; Selecting a maximum predicted value in at least one bounding box; Determining a boundary box corresponding to the maximum predicted value as the fault area and a known fault type corresponding to the maximum predicted value as the predicted fault type of the fault area in response to the maximum predicted value being greater than or equal to a first preset threshold value, or And in response to the maximum predicted value being smaller than a first preset threshold value, multiplying the maximum predicted value by the similarity matrix to obtain a predicted matrix of an unknown fault type, selecting a maximum value from the predicted matrix of the unknown fault type, taking a boundary box corresponding to the maximum value as the fault area, and taking the unknown fault type corresponding to the maximum value as the predicted fault type of the fault area, wherein the predicted matrix of the unknown fault type comprises predicted values corresponding to all the unknown fault types.
  10. 10. The method of claim 9, wherein the determining of the similarity matrix comprises: extracting visual characteristics obtained after the target fault detection model processes the first training sample of the known fault type; Determining semantic information of various fault types, converting the semantic information into attribute vectors to form an attribute table, evaluating the attribute table by using a proximity algorithm, and determining whether the attribute table is reasonable; In response to the unreasonable attribute table, optimally adjusting the attribute table until the attribute table after optimization adjustment is evaluated by the proximity algorithm and then determined to be reasonable, taking the final attribute table after optimization adjustment as a target attribute table, or Responding to the attribute table rationally, and taking the attribute table as a target attribute table; And determining the Euclidean distance between the attribute vector of each unknown fault category in the target attribute table and the attribute vector of each known fault category, and determining the similarity matrix according to the Euclidean distance.
  11. 11. The method according to claim 1, wherein the obtaining the station fault detection results of each station in the historical time period, performing outlier evolution prediction processing on the station fault detection results of each station in the historical time period, and determining the performance prediction result of the vehicle includes: Station fault detection results of all stations in a historical time period are obtained, and key feature extraction is carried out on the station fault detection results of all stations in the historical time period by utilizing a fault feature extraction algorithm to obtain key fault detection results; And inputting the key fault detection result into a performance prediction model obtained through pre-training, performing outlier evolution prediction processing, determining the overall performance of the vehicle, and obtaining a performance prediction result of the vehicle.
  12. 12. The method of claim 11, wherein the performance prediction model comprises an input layer, a forward processing layer, a reverse processing layer, and an output layer; Inputting the key fault detection result into a performance prediction model obtained by training in advance, performing outlier evolution prediction processing, determining the overall performance of the vehicle, and obtaining a performance prediction result of the vehicle, wherein the method comprises the following steps: Inputting the key fault detection result from an input layer of the performance prediction model, and performing forward processing on the key fault detection result by using the forward processing layer according to time points to obtain a forward output result corresponding to each time point; Performing reverse processing on the key fault detection result by using the reverse processing layer according to time points to obtain a reverse output result corresponding to each time point; The forward output result and the reverse output result at the same time point are fused to obtain a fusion result; determining a query vector, a matching vector and a target vector based on the fusion result; performing dot product processing according to the query vector and the matching vector, and performing normalization processing to obtain an attention score matrix; And carrying out weighting processing on a target matrix formed by the target vectors through the attention score matrix to obtain an attention processing result, and outputting the attention processing result as a vehicle performance prediction result through an output layer.

Description

Method for processing failure of vehicle Technical Field The application relates to the technical field of vehicle data processing, in particular to a vehicle fault processing method. Background When a vehicle detects faults, a lot of corresponding collected images exist, and a inspector is generally required to visually identify the corresponding fault conditions of the collected images. The image is identified by means of manpower, the workload is large, and the manpower is likely to be in error due to long-time work. And experience and level of different inspectors are different, so that fault detection results are large in difference and inaccurate. Disclosure of Invention In view of the above, the present application is directed to a vehicle fault handling method for solving or partially solving the above-mentioned problems. Based on the above object, the present application provides a fault handling method for a vehicle, comprising: Receiving a vehicle image set, splicing at least part of vehicle images in the vehicle image set by using an image splicing algorithm to obtain spliced images, and integrating the spliced images to obtain spliced result data; According to the vehicle images of each station in the vehicle image set, station fault detection is carried out on the vehicle images of the stations by utilizing a target fault detection model corresponding to the stations, and station fault detection results of the stations are obtained and stored; Marking the station fault detection result of at least one station in the spliced result data to obtain a fault marking result; Station fault detection results of all stations in a historical time period are obtained, station fault detection results of all stations in the historical time period are subjected to outlier evolution prediction processing, and performance prediction results of vehicles are determined. According to the fault processing method of the vehicle, the acquired vehicle images can be spliced together to obtain spliced images, the spliced images are integrated to obtain spliced result data, the vehicle images of the stations are screened according to each station, the screened vehicle images of the stations are subjected to fault detection by using a target fault detection model corresponding to the station to obtain station fault detection results corresponding to the station, the process is repeated for each station to obtain station fault detection results corresponding to each station, and the station fault detection results are more accurate due to the fact that the fault detection models of different stations process the vehicle images of the corresponding stations, so that the station fault detection results can be marked in spliced result data, the follow-up output is facilitated, the station fault detection results of each station in a historical time period are predicted through the evolution prediction processing of abnormal values, the user can obtain accurate vehicle performance prediction results, the user can acquire the fault detection results and the whole vehicle performance prediction results, and the manual fault detection and the overall performance prediction results can be conveniently, manually checked, comprehensively, and the overall performance of the vehicle can be restored, and the whole vehicle can be restored. Drawings In order to more clearly illustrate the technical solutions of the present application or related art, the drawings that are required to be used in the description of the embodiments or related art will be briefly described below, and it is apparent that the drawings in the following description are only embodiments of the present application, and other drawings may be obtained according to the drawings without inventive effort to those of ordinary skill in the art. Fig. 1 is a schematic diagram of an application scenario according to an embodiment of the present application; FIG. 2 is a flow chart of a method of fault handling for a vehicle according to an embodiment of the present application; FIG. 3 is a training schematic diagram of a fault detection model according to an embodiment of the present application; FIG. 4 is a schematic diagram of training a two-stage object detection model to obtain a fault detection model according to an embodiment of the present application; FIG. 5 is a schematic diagram of verification detection performed by a one-stage object detection model according to an embodiment of the present application; fig. 6 is a block diagram showing a configuration of a failure processing apparatus of a vehicle according to an embodiment of the present application; Fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application. Detailed Description The present application will be further described in detail below with reference to specific embodiments and with reference to the accompanying drawings, in order to make th