CN111652200A - Processing method, device and equipment for distinguishing multiple vehicles from pictures in vehicle insurance case - Google Patents
Processing method, device and equipment for distinguishing multiple vehicles from pictures in vehicle insurance case Download PDFInfo
- Publication number
- CN111652200A CN111652200A CN202010775095.XA CN202010775095A CN111652200A CN 111652200 A CN111652200 A CN 111652200A CN 202010775095 A CN202010775095 A CN 202010775095A CN 111652200 A CN111652200 A CN 111652200A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- license plate
- vehicle image
- information
- plate number
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a processing method, a device and equipment for distinguishing multiple vehicles from pictures in a vehicle insurance case, and relates to the technical field of vehicle picture distinguishing. The processing method for distinguishing the multiple vehicles from the pictures in the vehicle insurance case comprises the following steps: acquiring first vehicle image information of each picture to be processed; clustering similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the vehicle image areas containing the license plates; and acquiring the image areas of the vehicles without license plates in each cluster group, and correspondingly matching each image area of the vehicles without license plates with the unique license plate number associated with the cluster group. The method is based on a deep learning algorithm, extracts the characteristics of the vehicle images, and measures the similarity between vehicles by using the characteristics. The accurate binding of the vehicle image and the license plate number is realized through a clustering method, and the distinguishing and identifying efficiency of the vehicle is further improved.
Description
Technical Field
The invention relates to the technical field of vehicle picture distinguishing, in particular to a processing method, a device and equipment for distinguishing multiple vehicles from pictures in a vehicle insurance case.
Background
In the insurance claim settlement industry, in the links of vehicle insurance claim settlement loss, loss checking and claim checking, a large number of vehicle photos need to be classified and information identified. The manual identification mode is not only slow in identification speed, but also can cause the problem of identification errors.
In view of this problem, a solution is proposed in the prior art to distinguish different vehicle information by identifying the vehicle color, however, the inventors found that the following problems exist in the prior art: firstly, vehicles with the same or similar colors cannot be accurately distinguished; secondly, the technical scheme is that color features are extracted from the whole picture, and when multiple vehicles or a large number of backgrounds exist in one picture, vehicle information cannot be accurately distinguished; thirdly, the colors of part of appearance parts of the vehicle such as a middle net, a lower grid, tires, glass and the like are not obvious, and if the colors are not excluded, the identification accuracy is influenced; and fourthly, the license plate number corresponding to each vehicle cannot be distinguished, the insurance claim settlement service is closely related to the license plate number, and the subsequent service flow cannot be automated.
Disclosure of Invention
In order to solve at least one problem in the background art, the invention provides a processing method, a device and equipment for distinguishing multiple vehicles from pictures in a vehicle insurance case, which can realize accurate identification and distinguishing of the multiple vehicles in the pictures of the vehicle insurance case.
According to one aspect of the invention, a processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case comprises the following steps: acquiring vehicle insurance claim settlement case information, wherein the vehicle insurance claim settlement case information comprises a picture set to be processed; acquiring at least one piece of first vehicle image information of each picture to be processed, wherein each piece of first vehicle image information corresponds to one vehicle image area on the corresponding picture to be processed, each piece of first vehicle image information comprises vehicle image area position information, a license plate number and a similarity characterization vector, and the similarity characterization vector is used for describing the similarity degree between vehicles; clustering the similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, wherein the second vehicle image information comprises vehicle image area position information, license plate numbers and cluster group identifications to which the vehicle image areas belong; acquiring all license plate-containing vehicle image areas in each cluster group, and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas; and acquiring the image areas of the vehicles without license plates in each clustering group, and matching each image area without license plates with the unique license plate number associated with the clustering group.
Further optionally, the acquiring at least one piece of first vehicle image information of each to-be-processed picture includes: the method comprises the steps of obtaining at least one piece of first vehicle image information of each picture to be processed through a preset neural network model, wherein the preset neural network model is formed by extracting vehicle image features through a convolution neural network and utilizing a triple loss cost measurement function to train based on vehicle triple data, and the vehicle triple data comprise a target vehicle image, a similar vehicle image and a dissimilar vehicle image.
Further optionally, before the obtaining of the information of the car insurance claim case, the method further includes: acquiring historical claim settlement case information; acquiring historical photos of different types of vehicles in different shooting distances, scenes and angles from the historical claim settlement case information, wherein the historical photos belong to different types according to preset attributes; sequentially obtaining the historical photos of different categories to form the vehicle triple data; and training the preset neural network model by using the vehicle triple data.
Further optionally, the sequentially obtaining the historical photos of different categories to form the vehicle triple data includes: and classifying the historical photos by using the body color information and the vehicle model information as basic attributes of the historical photos.
Further optionally, before the acquiring at least one piece of first vehicle image information of each picture to be processed, the method further includes: determining each vehicle image area on each picture to be processed; and carrying out license plate detection in each vehicle image area, and carrying out license plate number identification.
Further optionally, the information of the car insurance claim settlement case further includes a license plate number of an accident vehicle, and clustering the similarity characterizing vectors corresponding to each vehicle image area includes: acquiring the license plate number of the accident vehicle from the vehicle insurance claim settlement case information; and determining the clustering number by combining the license plate number of the accident vehicle and the recognized license plate number.
Further optionally, the performing license plate detection in each vehicle image region and performing license plate number recognition includes: determining a license plate region of each vehicle image region based on a deep learning algorithm; and performing inclination and perspective correction on the license plate area.
Further optionally, determining the unique license plate number associated with each cluster group according to all license plate numbers associated with the image area containing the license plate vehicles includes: judging whether each cluster group contains different license plate numbers; and if the current cluster group contains different license plate numbers, associating the license plate number with the most corresponding vehicle area pictures with the current cluster group.
According to another aspect of the present invention, a processing apparatus for distinguishing a plurality of cars from a picture in a car insurance case includes: the system comprises a to-be-processed picture acquisition module, a to-be-processed picture acquisition module and a to-be-processed picture processing module, wherein the to-be-processed picture acquisition module is used for acquiring vehicle insurance claim case information which comprises a to-be-processed picture set; the system comprises a first vehicle image information acquisition module, a second vehicle image information acquisition module and a similarity representation module, wherein the first vehicle image information acquisition module is used for acquiring at least one piece of first vehicle image information of each picture to be processed, each piece of first vehicle image information corresponds to one vehicle image area on the corresponding picture to be processed, each piece of first vehicle image information comprises vehicle image area position information, a license plate number and a similarity representation vector, and the similarity representation vector is used for describing the similarity degree between vehicles; the clustering module is used for clustering the similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, wherein the second vehicle image information comprises vehicle image area position information, a license plate number and a cluster group identifier to which the vehicle image area belongs; the license plate association module is used for acquiring all license plate-containing vehicle image areas in each cluster group and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas; and the matching module is used for acquiring the image areas of the vehicles without the license plates in each clustering group and matching each image area without the license plates with the unique license plate number associated with the clustering group.
Further optionally, the apparatus further comprises: the historical claim settlement case information acquisition module is used for acquiring historical claim settlement case information; a historical photograph obtaining sub-module 4011, configured to obtain historical photographs of the vehicle at different shooting distances, scenes, and angles from the historical claim case information, where the historical photographs belong to different categories according to preset attributes; the vehicle triad composition sub-module is used for sequentially acquiring the historical photos of different categories to form the vehicle triad data; and the preset neural network model construction module is used for training the preset neural network model by utilizing the vehicle triple data.
Further optionally, the vehicle triplet making sub-module includes: and the classification unit is used for classifying the historical photos by using the body color information and the vehicle model information as basic attributes of the historical photos.
Further optionally, the first vehicle image information obtaining module includes: the preset neural network model obtaining sub-module is used for obtaining the similarity characteristic vector through a preset neural network model, the preset neural network model is formed by extracting vehicle image characteristics through a convolution neural network and utilizing a triple Loss cost measurement function to train based on vehicle triple data, and the vehicle triple data comprises a target vehicle image, a similar vehicle image and a dissimilar vehicle image.
Further optionally, the first vehicle image information obtaining module includes: the vehicle detection submodule is used for detecting each vehicle image area on each picture to be processed; and the license plate detection sub-module is used for carrying out license plate detection in each vehicle image area and carrying out license plate number identification.
Further optionally, the license plate detection sub-module includes: the license plate region determining unit is used for determining the license plate region of each vehicle image region based on a deep learning algorithm or a traditional image processing method; and the correction unit is used for carrying out inclination and perspective correction on the license plate area.
Further optionally, the clustering module includes: the license plate number obtaining sub-module is used for obtaining the license plate number of the accident vehicle from the vehicle insurance claim case information; and the clustering number determining submodule is used for determining the clustering number by combining the license plate number of the accident vehicle and the recognized license plate number.
Further optionally, the license plate association module includes: the judging submodule is used for judging whether each clustering group contains different license plate numbers; and if the current cluster group contains different license plate numbers, associating the license plate number with the most corresponding vehicle area pictures with the current cluster group.
According to another aspect of the invention, an apparatus comprises: at least one processor; and a memory coupled to the at least one processor; wherein the memory stores a computer program executable by the at least one processor to implement a process of distinguishing multiple vehicles from a picture in the vehicle insurance case.
The invention has the beneficial effects that:
1. the invention can establish the binding relationship between the vehicle image information and the license plate number, accurately distinguish the picture information of each vehicle and lay a foundation for damage assessment and damage checking of the vehicle insurance claim settlement case based on the appearance.
2. The method extracts the characteristics of the vehicle images through the deep neural network model trained based on the triple data, and measures the similarity between the vehicles by using the characteristics. The accurate binding of the vehicle image and the license plate number is realized through a clustering method, and the distinguishing and identifying efficiency of the vehicle is further improved.
3. The invention realizes the accurate recognition of the license plate number by carrying out multi-dimensional correction on the license plate. Meanwhile, the identified license plate number needs to be compared with case information so as to obtain a more accurate license plate number.
Drawings
FIG. 1 is a flow chart of a processing method for distinguishing multiple cars from pictures in a car insurance case according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for distinguishing multiple vehicles from a picture in another vehicle insurance case according to the present invention;
FIGS. 3 and 4 are flow diagrams illustrating one implementation of steps 204 and 206 in FIG. 2, respectively;
FIG. 5 is a functional structure diagram of a processing device for distinguishing multiple cars from pictures in a car insurance case according to an embodiment of the present invention.
Detailed Description
The content of the invention will now be discussed with reference to a number of exemplary embodiments. It is to be understood that these examples are discussed only to enable those of ordinary skill in the art to better understand and thus implement the teachings of the present invention, and are not meant to imply any limitations on the scope of the invention.
As used herein, the term "include" and its variants are to be read as open-ended terms meaning "including, but not limited to. The term "based on" is to be read as "based, at least in part, on". The terms "one embodiment" and "an embodiment" are to be read as "at least one embodiment". The term "another embodiment" is to be read as "at least one other embodiment".
Example 1:
as shown in fig. 1, an embodiment of the present invention provides a processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case, which is used for a vehicle insurance claim case with two or more vehicles, and includes:
101. acquiring vehicle insurance claim settlement case information, wherein the vehicle insurance claim settlement case information comprises a picture set to be processed;
102. acquiring at least one piece of first vehicle image information of each picture to be processed, wherein each piece of first vehicle image information corresponds to one vehicle image area on the corresponding picture to be processed, each piece of first vehicle image information comprises vehicle image area position information, a license plate number and a similarity characteristic vector, and the similarity characteristic vector is used for describing the similarity degree between vehicles;
103. clustering similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, wherein the second vehicle image information comprises vehicle image area position information, license plate numbers and cluster group identifications to which the vehicle image areas belong;
104. acquiring all license plate-containing vehicle image areas in each cluster group, and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas;
105. and acquiring the image areas of the vehicles without license plates in each cluster group, and matching each image area without license plates with the unique license plate number associated with the cluster group.
By the image processing method in the vehicle insurance claim case, vehicle region detection and license plate number recognition can be performed on the vehicle images in the image set to be processed, vehicle features can be extracted through the preset neural network model, and the vehicle features can be clustered, so that the association between the vehicle image regions and the license plate numbers is realized, the information of each vehicle is accurately distinguished, and a foundation is laid for damage assessment and damage checking of the vehicle insurance claim case based on the appearance.
Example 2:
as shown in fig. 2, as an improvement of the above embodiment, the present invention provides a processing method for distinguishing multiple cars from a picture in a car insurance case, which includes:
201. acquiring historical claim settlement case information;
acquiring historical claim case information, acquiring historical photos of the vehicle at different shooting distances, scenes and angles from the historical claim case information, wherein the historical photos belong to different categories according to preset attributes; sequentially acquiring historical photos of different categories to form vehicle triple data; and training a preset neural network model by using the vehicle triple data.
In some optional embodiments, the step of sequentially obtaining the vehicle triple data composed of the different types of historical photographs comprises: the history photographs are categorized using the body color information and the vehicle model information as basic attributes of the history photographs. The construction of the vehicle three-component group data comprises an offline manual construction mode and the on-line automatic mining of the three-component group data meeting the conditions in the training process. In this embodiment, an online construction method is adopted.
Preferably, a specific embodiment of acquiring the historical claim case information in this embodiment is as follows: a large number of pictures of vehicles of the same category attribute at different shooting distances, scenes and angles are obtained through historical claim settlement case information, and therefore triple training data of 'target vehicle images', 'same vehicle images' and 'different vehicle images' can be constructed.
And (3) training a depth measurement neural network by using a triple loss function, namely presetting a neural network model to obtain a similarity characterization vector of the vehicle in a semantic space. The purpose of the step is to enable the representation vectors of the vehicle pictures with the same category attributes to be as close as possible in the space, the representation vectors of the vehicle pictures with different category attributes to be as separate as possible in the space, and the semantic expression of the information such as the vehicle type structure, the vehicle body color and the like of the vehicle is embedded by the representation vectors learned by the constructed triple training data. When the vehicle type structure and the vehicle body color of the vehicle are similar, the characteristic vectors are as close as possible in the space, and when the vehicle type structure and the vehicle body color of the vehicle are not similar, the characteristic vectors are as separate as possible in the space.
When organizing the training data, only the far, middle and near images in case data should be used. For a distant view image, the image may contain a plurality of vehicles, or excessive background interference, and therefore it is necessary to use a vehicle area image acquired by vehicle detection as training data.
According to the difference of the fine granularity of the measurement, the marking mode of the training data is different. The method adopts the vehicle type and the color as basic attributes to organize and label the data. For example, all "black audi A6L" as a class. The method adopts 1500 common vehicle types and 13 quantized colors as basic attribute organization training samples. Quantifying color includes: black, gray, silver, white, red, orange, brown, yellow, green, cyan, blue, violet, pink. In addition, more than 20 colors consisting of single colors are also included, such as "yellow green" which is common in taxis.
202. Acquiring vehicle insurance claim settlement case information, wherein the vehicle insurance claim settlement case information comprises a picture set to be processed and an accident vehicle license plate number;
203. detecting each vehicle image area on each picture to be processed;
preferably, a specific implementation manner of determining the vehicle image area in this embodiment is as follows:
due to non-standardization of vehicle insurance case image acquisition, a group of case pictures may contain vehicle information, scene information, certificate information, document information and the like, and the embodiment mainly processes the vehicle image information. Therefore, the present embodiment preferably uses deep learning classification algorithms such as Resnet, VGG, etc. to classify the case pictures into car-containing pictures and non-car-containing pictures. For the picture containing the car, the picture is more finely divided into three types of images of a close shot, a medium shot and a long shot.
Vehicle close-up image: the shooting distance is very close, and only a small area, a single appearance piece and a part of the appearance piece of the automobile body can be seen on the picture.
Vehicle medium view image: when the shooting distance is slightly far, a plurality of car body appearance parts can be seen on the picture, and a part of the whole car is exposed. The entire vehicle cannot be seen.
Vehicle perspective image: the picture is taken at a great distance, and substantially the entire vehicle or a plurality of vehicles can be seen on the picture.
Each vehicle image region is preferably detected using a deep learning algorithm such as fasternn/YOLO/SSD for all photographs including the vehicle medium view image and the vehicle distant view image. Since the close-range image probably contains only the same vehicle, the entire image is regarded as a vehicle image region on the close-range image. This determines each vehicle image area on the picture.
204. And carrying out license plate detection in each vehicle image area and carrying out license plate number identification.
Preferably, a specific implementation manner of performing license plate detection in this embodiment is as follows:
and detecting the license plate in the vehicle image region in the picture to be processed, and preferably determining the specific license plate region by adopting a traditional image processing method, a deep learning method and the like.
The traditional image processing method for detecting the license plate area mainly comprises the following processes: detecting a vertical edge of the image by using a canny operator; performing morphological operation on the detected image edge, and calculating a connected domain; forming a plurality of candidate regions with different sizes, aspect ratios, directions and colors; finally determining one or more license plate candidate regions according to the prior knowledge of the license plate; and then, sending the obtained license plate candidate area into classifiers such as a trained SVM, a decision tree and the like for secondary classification, and finally judging whether the license plate candidate area belongs to the license plate area.
As shown in fig. 3, step 204 in some alternative embodiments may be implemented by, but is not limited to, the following processes: 2041. determining a license plate region of each vehicle image region based on a deep learning algorithm;
detecting a license plate region based on a deep learning algorithm, namely, taking a license plate as a target, and detecting by using a target detection algorithm such as FasterRCNN/YOLO/SSD; or detecting the license plate area from the image by using a MaskRCNN method of image segmentation.
2042. And (5) performing inclination and perspective correction on the license plate area.
The detected license plate region needs to be subjected to inclination and perspective correction so as to improve the accuracy of the subsequent license plate recognition. The detection area of the license plate area is divided into two types: a bounding box (bounding box) and a license plate area mask (mask). For the license plate region with a circumscribed rectangle, the image of the license plate region can be firstly segmented by an image segmentation algorithm (such as MSER) and the like, binarization is carried out, and then a connected domain of a character region in the license plate is obtained. And generating an edge contour point set of the connected domain, and analyzing and acquiring four fixed point coordinate points of the connected domain. Thereby establishing an affine matrix of the inclination/perspective transformation and realizing the correction.
After the license plate number is identified, the license plate number needs to be compared and calibrated with the license plate number of the accident vehicle in the vehicle insurance claim case information. In the last step, the corrected license plate area is input into the CRNN to identify the license plate number. And meanwhile, calculating the edit distance (Levenshtein) between the license plate number recognition result and the license plate number of the accident vehicle in the vehicle insurance claim case information in the case, and comparing and calibrating. The robustness of license plate recognition under complex conditions (such as shading, fouling, bending and the like) is improved.
205. And acquiring at least one piece of first vehicle image information of each picture to be processed.
Each piece of first vehicle image information corresponds to one vehicle image area on the corresponding picture to be processed, each piece of first vehicle image information comprises vehicle image area position information, a license plate number and a similarity characteristic vector, and the similarity characteristic vector is used for describing the similarity between vehicles;
in some optional embodiments, at least one piece of first vehicle image information of each to-be-processed picture is acquired through a preset neural network model, the preset neural network model is formed by training vehicle triple data, and the vehicle triple data comprises a target vehicle image, a similar vehicle image and a dissimilar vehicle image.
Preferably, a specific implementation manner of constructing the feature vector of the similarity in this embodiment is as follows:
and for all vehicle image areas, if the license plate is detected, determining the vehicle to which the license plate belongs according to the license plate area and the IOU value of the vehicle image area. After the above steps, an information association table for each vehicle in the image can be obtained: veh _ Info = { Img, Rect, Plate }
Wherein Veh _ Info represents information of each vehicle in the vehicle image area, not information of the entire image; img represents image data; rect represents the area of the vehicle in the image; plate represents the license Plate number of the vehicle. When no license Plate exists, the Plate field is empty.
And sending the detected vehicle area image in each picture to be processed into a similarity measurement deep neural network, and extracting a similarity characterization vector of the vehicle.
A similarity measure deep neural network model, preferably using a network model of the VGG16 or Resnet series as a feature extraction network; specific triple data can be organized according to different similarity measurement requirements, and the triple Loss can be used as a Loss function. And the measurement evaluation of the similarity of the vehicle image areas is realized. The similarity measure deep neural network can be trained through the historical claim settlement case information of the insurance company, that is, the training mode in step 201.
After all images are processed, each vehicle image area generates an associated information table: veh _ Info = { Img, Rect, Plate, Feature }, where Feature represents a similarity characterization vector.
206. Clustering similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, wherein the second vehicle image information comprises vehicle image area position information, license plate numbers and cluster group identifications to which the vehicle image areas belong;
as shown in fig. 4, step 206 in some alternative embodiments may be implemented by, but is not limited to, the following processes:
2061. acquiring the license plate number of an accident vehicle from the vehicle insurance claim settlement case information;
2062. and determining the clustering number by combining the license plate number of the accident vehicle and the recognized license plate number.
Preferably, a specific implementation manner of clustering in this embodiment is as follows:
the similarity calculation of the vehicle image areas can be preferably measured by the size of cosine included angles between vectors or the distance of Euclidean distances. In the embodiment, K-means clustering is adopted as n groups (G1, G2.. Gn), wherein n corresponds to the number of involved vehicles in the damage assessment case information. If n =1, the case is single-vehicle, meaning that clustering is not required, and if n =2, the case is double-vehicle accident.
After clustering, all vehicle image areas corresponding to one vehicle belong to a Group of clustering results, and Veh _ Info = { Img, Rect, Plate, Feature, Group }. Wherein, Group represents the clustering result to which the vehicle belongs.
Sometimes, the case information is not uploaded timely or completely, and the case may only contain images of a single vehicle; or only a single license plate number is recorded in case information. Furthermore, the vehicle image area and the license plate number do not necessarily correspond to the same vehicle.
Therefore, when selecting the number of clusters, the license plate numbers of the accident vehicles recorded in the vehicle insurance claim case information and the license plate numbers actually recognized from the vehicle image area need to be comprehensively referred to. And taking a union set of the license plate numbers identified in the vehicle image area and a set of the license plate numbers recorded in the vehicle insurance claim case information. Obtaining the number of case license plate numbers, namely the number of vehicles involved in the case;
207. acquiring all license plate-containing vehicle image areas in each cluster group, and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas;
in some alternative embodiments, it is determined whether each cluster group contains a different license plate number; and if the current cluster group contains different license plate numbers, associating the license plate number with the most corresponding vehicle area pictures with the current cluster group.
Preferably, a specific implementation manner of associating the license plate number with the cluster group in this embodiment is as follows:
and counting the license plate numbers associated with each vehicle image area in each group of clustering results, and acquiring all the license plate numbers contained in the group of clustering results. If all the license plate numbers are the same, the license plate number is uniquely associated with the cluster group; if there are multiple different license plate numbers, further corrections to the cluster group are needed. And when a plurality of different license plates exist in the clustering result, respectively counting the number of the vehicle images corresponding to each license plate number, and selecting the license plate number corresponding to the vehicle image with the largest number as the unique associated license plate of the clustering result.
208. And acquiring the image areas of the vehicles without license plates in each cluster group, and matching each image area without license plates with the unique license plate number associated with the cluster group.
And traversing all the vehicle information in the cluster group, and if the license plate number information is null, assigning the unique license plate number associated with the cluster group to the vehicle. After processing, all vehicles in all images in the case are bound with an associated license plate number. And comparing the license plate number information of each vehicle in all the images with the license plate number information recorded in the case information, thereby determining the vehicle image corresponding to the license plate number which is really concerned in the case processing process.
The beneficial effect of this embodiment lies in:
1. the embodiment can establish the binding relationship between the vehicle image information and the license plate number, accurately distinguish the picture information of each vehicle, and lay a foundation for damage assessment and damage checking of the vehicle insurance claim settlement case based on the appearance.
2. The method extracts the characteristics of the vehicle images through the deep neural network model trained based on the triple data, and measures the similarity between the vehicles by using the characteristics. The accurate binding of the vehicle image and the license plate number is realized through a clustering method, and the distinguishing and identifying efficiency of the vehicle is further improved.
3. In the embodiment, the license plate is subjected to multi-dimensional correction, so that the license plate number is accurately identified. Meanwhile, the identified license plate number needs to be compared with case information so as to obtain a more accurate license plate number.
Example 3:
for the convenience of the reader to understand, the present embodiment details, with reference to specific examples, the method for processing an image in a car insurance claim case in the foregoing, specifically, the method includes:
the first stage is as follows:
the first step is as follows: because the pictures to be processed in the vehicle insurance claim case information are not standardized in the acquisition process, the pictures to be processed in a group of vehicle insurance claim case information may contain vehicle information, scene information, certificate information, document information and the like, and the invention mainly processes the vehicle information. And classifying the pictures to be processed into the pictures containing the vehicles and the pictures not containing the vehicles by adopting deep learning classification algorithms such as Resnet, VGG and the like. For the picture containing the car, the picture is more finely divided into three types of images of a close shot, a medium shot and a long shot.
Vehicle close-up image: the shooting distance is very close, and only a small area, a single appearance piece and a part of the appearance piece of the automobile body can be seen on the picture.
Vehicle medium view image: when the shooting distance is slightly far, a plurality of car body appearance parts can be seen on the picture, and a part of the whole car is exposed. The entire vehicle cannot be seen.
Vehicle perspective image: the picture is taken at a great distance, and substantially the entire vehicle or a plurality of vehicles can be seen on the picture.
The second step is that: and detecting each vehicle image area by utilizing a deep learning algorithm such as FasterRCNN/YOLO/SSD and the like for all the pictures to be processed including the vehicle intermediate view image and the vehicle distant view image. Since the close-range image of the vehicle probably contains only the same vehicle, the whole image is taken as the vehicle image area on the close-range image, and thus the vehicle image area of each vehicle on the image to be processed is determined.
The third step: and detecting license plate regions in all the pictures to be processed, and determining the specific license plate regions by adopting a traditional image processing method, a deep learning algorithm and the like.
The traditional method for detecting the license plate area mainly comprises the following processes: detecting a vertical edge of the image by using a canny operator; performing morphological operation on the detected image edge, and calculating a connected domain; forming a plurality of candidate regions with different sizes, aspect ratios, directions and colors; finally determining one or more license plate candidate regions according to the prior knowledge of the license plate; and then, sending the obtained license plate candidate area into classifiers such as a trained SVM, a decision tree and the like for secondary classification, and finally judging whether the license plate candidate area belongs to the license plate area.
Based on a deep learning algorithm, a license plate region is taken as a target, and detection is carried out by using a target detection algorithm such as FasterRCNN/YOLO/SSD; or detecting the license plate area from the image by using a MaskRCNN method of image segmentation.
The detected license plate region needs to be subjected to inclination/perspective correction so as to improve the accuracy of the subsequent license plate recognition. The detection area of the license plate area is divided into two types: a bounding box (bounding box) and a license plate area mask (mask). For the license plate region with a circumscribed rectangle, the image of the license plate region can be firstly segmented by an image segmentation algorithm (such as MSER) and the like, binarization is carried out, and then a connected domain of a character region in the license plate is obtained. And generating an edge contour point set of the connected domain, and analyzing and acquiring four fixed point coordinate points of the connected domain. Thereby establishing an affine matrix of the inclination/perspective transformation and realizing the correction.
The fourth step: and comparing and calibrating the license plate number identification with the license plate number of the accident vehicle in case information.
In the last step, the corrected license plate area is input into the CRNN for license plate number recognition. And meanwhile, calculating the edit distance (Levenshtein) between the license plate number recognition result and the license plate number of the accident vehicle in the case, and comparing and calibrating. The robustness of license plate number recognition under complex conditions (such as shading, fouling, bending and the like) is improved.
The fifth step: and for each picture, if the license plate area is detected, determining the vehicle to which the license plate number belongs according to the license plate area and the IOU value of the vehicle image area. After the above steps, the information association table of each vehicle image area in the image can be obtained: veh _ Info = { Img, Rect, Plate };
wherein Veh _ Info represents information of each vehicle in the image, not information of the entire image; img represents image data; rect represents the area of the vehicle in the image; plate represents the license Plate number of the vehicle. When no license Plate exists, the Plate field is empty.
And a sixth step: and (3) sending the vehicle image area detected in each picture to be processed into a similarity measurement deep neural network, and extracting a similarity characterization vector of the vehicle image area.
The similarity measurement deep neural network model uses a network model of VGG16 or Resnet series as a feature extraction network; specific vehicle triple data can be organized according to different similarity measurement requirements, and the triple Loss can be used as a Loss function. And realizing the measurement evaluation of the image similarity.
The similarity metric deep neural network can be trained through historical claim case information of insurance companies. Historical pictures of vehicles with the same category attribute at different shooting distances, scenes and angles are obtained through historical claim settlement case information, triple training data of 'target vehicle images', 'same-category vehicle images' and 'different-category vehicle images' can be constructed, a depth measurement neural network is trained through a triple loss function, similarity characterization vectors of the vehicles in a semantic space are obtained, the goal is to enable the characterization vectors of the vehicle images with the same category attribute to be as close as possible in the space, the characterization vectors of the vehicle images with different category attributes to be as separated as far as possible in the space, and the characterization vectors learned through the constructed triple training data are embedded with semantic expressions of information such as vehicle type structures and vehicle body colors of the vehicles.
When training data is organized, only a vehicle long-range view image, a vehicle medium-range view image and a vehicle short-range view image in case data are adopted. It is possible for the vehicle perspective image to contain a plurality of vehicles, or excessive background interference, and therefore it is necessary to use the vehicle area image acquired by vehicle detection as training data.
According to the difference of the fine granularity of the measurement, the marking mode of the training data is different. The method adopts the vehicle model and the vehicle body color as basic attributes to organize and label the data. For example, all "black audi A6L" as a class. The embodiment adopts 1500 common vehicle models and 13 quantified colors as basic attribute organization training samples. Quantifying color includes: black, gray, silver, white, red, orange, brown, yellow, green, cyan, blue, violet, pink. In addition, more than 20 colors consisting of single colors are also included, such as "yellow green" which is common in taxis.
The seventh step: after all the pictures to be processed are processed, each vehicle generates an associated information table: veh _ Info = { Img, Rect, Plate, Feature }. Wherein Feature represents a similarity characterizing vector.
In the second phase, the above information of the vehicles is used for clustering and the clustering result is corrected.
Second stage
Eighth step: and clustering the extracted similarity characterization vectors of each vehicle.
The similarity can be calculated by measuring the magnitude of cosine included angle between vectors or the distance of Euclidean distance. The method adopts K-means clustering as n groups (G1, G2.. Gn), wherein n corresponds to the number of involved vehicles in the damage assessment case information. If n =1, the case is single-vehicle, meaning that clustering is not required, and if n =2, the case is double-vehicle accident.
After clustering, each vehicle belongs to a Group of clustering results, and Veh _ Info = { Img, Rect, Plate, Feature, Group }. Wherein, Group represents the clustering result to which the vehicle belongs.
Sometimes, the uploading of the vehicle claim settlement case information is not timely or complete enough, and the case may also contain images of only a single vehicle; or only a single license plate number is recorded in case information. Further, the image data and the license plate number do not necessarily correspond to the same vehicle.
Therefore, when selecting the number of clusters, the license plate numbers of the accident vehicles recorded in the case information and the license plate numbers actually recognized from the to-be-processed images of the vehicle claims case information need to be comprehensively referred to. And taking a union set of the license plate numbers identified in the picture to be processed and a set of the license plate numbers recorded in the vehicle insurance claim case information. Obtaining the number of case license plate numbers, namely the number of vehicles involved in the case;
the ninth step: each group of clustering results needs to be associated with a unique license plate number. And counting the license plate number information associated with each vehicle image area in each group of clustering results, and acquiring all license plate numbers contained in the group of clustering results. If all the license plate numbers are the same, the license plate number is uniquely associated with the cluster group; if there are multiple different license plate numbers, further corrections to the cluster group are needed.
And when a plurality of different license plates exist in the clustering result, respectively counting the number of the vehicle image areas corresponding to each license plate number, and selecting the license plate number corresponding to the vehicle image area with the largest number as the unique associated license plate of the clustering result.
The third stage
The tenth step: based on the associated information table in the second stage, each cluster group has a unique license plate number corresponding to it. And traversing all the second vehicle image information in the cluster group, and if the license plate number is empty, assigning the unique license plate number associated with the cluster group to the vehicle.
After processing, all vehicles in all vehicle image areas in the case are bound with an associated license plate number.
The eleventh step: and comparing the license plate number of each vehicle in all the vehicle image areas with the license plate number information of the accident vehicle recorded in the vehicle insurance claim case information, thereby determining the vehicle image area corresponding to the license plate number which is really concerned in the case processing process.
The beneficial effect of this embodiment lies in:
1. the embodiment can establish the binding relationship between the vehicle image information and the license plate number, accurately distinguish the picture information of each vehicle, and lay a foundation for damage assessment and damage checking of the vehicle insurance claim settlement case based on the appearance.
2. The method extracts the characteristics of the vehicle images through the deep neural network model trained based on the triple data, and measures the similarity between the vehicles by using the characteristics. The accurate binding of the vehicle image and the license plate number is realized through a clustering method, and the distinguishing and identifying efficiency of the vehicle is further improved.
3. In the embodiment, the license plate is subjected to multi-dimensional correction, so that the license plate number is accurately identified. Meanwhile, the identified license plate number needs to be compared with case information so as to obtain a more accurate license plate number.
Example 4:
as shown in fig. 5, in order to implement the above method cooperatively, this embodiment provides a processing apparatus for distinguishing multiple cars from pictures in a car insurance case, and the processing apparatus is used for a car insurance claim case with two cars or multiple cars, and the apparatus includes:
a historical claim settlement case information obtaining module 401, configured to obtain historical claim settlement case information;
the historical photograph obtaining sub-module 4011 is configured to obtain historical photographs of the vehicle at different shooting distances, scenes and angles from the historical claim settlement case information, where the historical photographs belong to different categories according to preset attributes;
the vehicle triple composition sub-module 4012 is used for sequentially obtaining historical photos of different categories to form vehicle triple data; the vehicle triplet composition sub-module 4012 includes:
a classification unit 4012a for classifying the history photograph using the body color information and the vehicle model information as basic attributes of the history photograph.
And a preset neural network model building module 4013, configured to train the preset neural network model using the vehicle triplet data.
A pending picture obtaining module 402, configured to obtain vehicle insurance claim case information, where the vehicle insurance claim case information includes a pending picture set;
a first vehicle image information obtaining module 403, configured to obtain at least one piece of first vehicle image information of each to-be-processed picture, where each piece of first vehicle image information corresponds to one vehicle image area on a corresponding to-be-processed picture, each piece of first vehicle image information includes vehicle image area position information, a license plate number, and a similarity characterization vector, and the similarity characterization vector is used to describe a degree of similarity between vehicles;
the first vehicle image information acquisition module 403 includes:
the preset neural network model obtaining sub-module 4031 is used for obtaining the feature vectors of the similarity through the preset neural network model, the preset neural network model is based on vehicle triple data, vehicle image features are extracted through a convolutional neural network, and the preset neural network model is trained through a triple Loss cost measurement function, and the vehicle triple data comprise a target vehicle image, a similar vehicle image and a dissimilar vehicle image.
A vehicle detection sub-module 4032 for detecting each vehicle image area on each picture to be processed;
and the license plate detection submodule 4033 is used for performing license plate detection in each vehicle image area and performing license plate number identification.
License plate detection submodule 4033 includes:
a license plate region determination unit 4033a for determining the license plate region of each vehicle image region based on a deep learning algorithm or a conventional image processing method;
and a modification unit 4033b for performing inclination and perspective correction on the license plate region.
The clustering module 404 is configured to cluster the similarity characterizing vectors corresponding to each vehicle image region to obtain second vehicle image information of each vehicle image region, where the second vehicle image information includes vehicle image region position information, a license plate number, and a cluster group identifier to which the vehicle image region belongs;
the clustering module 404 includes: the acquiring license plate number sub-module 4041 is used for acquiring the license plate number of the accident vehicle from the insurance claim case information; a clustering number determination submodule 4042, configured to determine a clustering number by combining the license plate number of the accident vehicle and the recognized license plate number;
the license plate association module 405 is used for acquiring all license plate-containing vehicle image areas in each cluster group, and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas;
a judging sub-module 4051 for judging whether each cluster group contains a different license plate number; and if the current cluster group contains different license plate numbers, associating the license plate number with the most corresponding vehicle area pictures with the current cluster group.
And a matching module 406, configured to obtain an image area without a license plate in each cluster group, and match each image area without a license plate with the unique license plate number associated with the cluster group in which the image area without a license plate is located.
The beneficial effect of this embodiment lies in:
1. the embodiment can establish the binding relationship between the vehicle image information and the license plate number, accurately distinguish the picture information of each vehicle, and lay a foundation for damage assessment and damage checking of the vehicle insurance claim settlement case based on the appearance.
2. The method extracts the characteristics of the vehicle images through the deep neural network model trained based on the triple data, and measures the similarity between the vehicles by using the characteristics. The accurate binding of the vehicle image and the license plate number is realized through a clustering method, and the distinguishing and identifying efficiency of the vehicle is further improved.
3. In the embodiment, the license plate is subjected to multi-dimensional correction, so that the license plate number is accurately identified. Meanwhile, the identified license plate number needs to be compared with case information so as to obtain a more accurate license plate number.
Example 5
In accordance with another aspect of the present invention, there is provided an apparatus comprising:
at least one processor; and
a memory coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores a computer program that can be executed by the at least one processor to implement the picture processing method in the car insurance claim case of the present invention.
Those of ordinary skill in the art will appreciate that the various illustrative modules and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described apparatuses and devices may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is merely a logical division, and in actual implementation, there may be other divisions, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or modules, and may be in an electrical, mechanical or other form.
The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical modules, may be located in one place, or may be distributed on a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment of the present invention.
In addition, functional modules in the embodiments of the present invention may be integrated into one processing module, or each of the modules may exist alone physically, or two or more modules are integrated into one module.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method for transmitting/receiving the power saving signal according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.
The above description is only a preferred embodiment of the application and is illustrative of the principles of the technology employed. It will be appreciated by a person skilled in the art that the scope of the invention as referred to in the present application is not limited to the embodiments with a specific combination of the above-mentioned features, but also covers other embodiments with any combination of the above-mentioned features or their equivalents without departing from the inventive concept. For example, the above features may be replaced with (but not limited to) features having similar functions disclosed in the present application.
It should be understood that the order of execution of the steps in the summary of the invention and the embodiments of the present invention does not absolutely imply any order of execution, and the order of execution of the steps should be determined by their functions and inherent logic, and should not be construed as limiting the process of the embodiments of the present invention.
Claims (10)
1. A processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case is characterized by comprising the following steps:
acquiring vehicle insurance claim settlement case information, wherein the vehicle insurance claim settlement case information comprises a picture set to be processed;
acquiring at least one piece of first vehicle image information of each picture to be processed, wherein each piece of first vehicle image information corresponds to one vehicle image area on the corresponding picture to be processed, each piece of first vehicle image information comprises vehicle image area position information, a license plate number and a similarity characterization vector, and the similarity characterization vector is used for describing the similarity degree between vehicles;
clustering the similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, wherein the second vehicle image information comprises vehicle image area position information, license plate numbers and cluster group identifications to which the vehicle image areas belong;
acquiring all license plate-containing vehicle image areas in each cluster group, and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas;
and acquiring the image areas of the vehicles without license plates in each clustering group, and matching each image area without license plates with the unique license plate number associated with the clustering group.
2. The processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case according to claim 1, wherein the obtaining at least one piece of first vehicle image information of each picture to be processed comprises:
and obtaining the similarity characteristic vector through a preset neural network model, wherein the preset neural network model is formed by extracting vehicle image characteristics by using a convolution neural network and training by using a triple Loss cost measurement function based on vehicle triple data, and the vehicle triple data comprises a target vehicle image, a similar vehicle image and a dissimilar vehicle image.
3. The processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case according to claim 2, further comprising, before the obtaining of the information of the vehicle insurance claim case:
acquiring historical claim settlement case information;
acquiring historical photos of the vehicle at different shooting distances, scenes and angles from the historical claim settlement case information, wherein the historical photos belong to different categories according to preset attributes;
sequentially obtaining the historical photos of different categories to form the vehicle triple data;
and training the preset neural network model by using the vehicle triple data.
4. The processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case according to claim 3, wherein the sequentially obtaining the historical photos of different categories to form the vehicle triple data comprises:
and classifying the historical photos by using the body color information and the vehicle model information as basic attributes of the historical photos.
5. The method for processing multiple vehicles distinguished from pictures in a vehicle insurance case according to any one of claims 1 to 4, wherein the obtaining at least one piece of first vehicle image information of each picture to be processed further comprises:
detecting each vehicle image area on each picture to be processed;
and carrying out license plate detection in each vehicle image area, and carrying out license plate number identification.
6. The processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case according to claim 5, wherein the vehicle insurance claim case information further includes an accident vehicle license plate number, and the clustering the similarity characterizing vectors corresponding to each vehicle image area includes:
acquiring the license plate number of the accident vehicle from the vehicle insurance claim settlement case information;
and determining the clustering number by combining the license plate number of the accident vehicle and the recognized license plate number.
7. The processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case according to claim 5, wherein the detecting license plates in each vehicle image area and the identifying license plate numbers comprise:
determining a license plate region of each vehicle image region based on a deep learning algorithm;
and performing inclination and perspective correction on the license plate area.
8. The processing method for distinguishing multiple vehicles from pictures in a vehicle insurance case according to claim 6, wherein determining the unique license plate number associated with each cluster group according to all license plate numbers associated with the image area containing the license plate vehicles comprises:
judging whether each cluster group contains different license plate numbers;
and if the current cluster group contains different license plate numbers, associating the license plate number with the most corresponding vehicle area pictures with the current cluster group.
9. A processing device for distinguishing multiple vehicles from pictures in a vehicle insurance case is characterized in that,
the system comprises a to-be-processed picture acquisition module, a to-be-processed picture acquisition module and a to-be-processed picture processing module, wherein the to-be-processed picture acquisition module is used for acquiring vehicle insurance claim case information which comprises a to-be-processed picture set;
the system comprises a first vehicle image information acquisition module, a second vehicle image information acquisition module and a similarity representation module, wherein the first vehicle image information acquisition module is used for acquiring at least one piece of first vehicle image information of each picture to be processed, each piece of first vehicle image information corresponds to one vehicle image area on the corresponding picture to be processed, each piece of first vehicle image information comprises vehicle image area position information, a license plate number and a similarity representation vector, and the similarity representation vector is used for describing the similarity degree between vehicles;
the clustering module is used for clustering the similarity characterization vectors corresponding to each vehicle image area to obtain second vehicle image information of each vehicle image area, wherein the second vehicle image information comprises vehicle image area position information, a license plate number and a cluster group identifier to which the vehicle image area belongs;
the license plate association module is used for acquiring all license plate-containing vehicle image areas in each cluster group and determining a unique license plate number associated with each cluster group according to all license plate numbers associated with the license plate-containing vehicle image areas;
and the matching module is used for acquiring the image areas of the vehicles without the license plates in each clustering group and matching each image area without the license plates with the unique license plate number associated with the clustering group.
10. An electronic device, comprising:
at least one processor; and
a memory coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores a computer program executable by the at least one processor to implement a method of distinguishing multiple vehicles from a picture in a car insurance case of any of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010775095.XA CN111652200A (en) | 2020-08-05 | 2020-08-05 | Processing method, device and equipment for distinguishing multiple vehicles from pictures in vehicle insurance case |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010775095.XA CN111652200A (en) | 2020-08-05 | 2020-08-05 | Processing method, device and equipment for distinguishing multiple vehicles from pictures in vehicle insurance case |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111652200A true CN111652200A (en) | 2020-09-11 |
Family
ID=72350657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010775095.XA Pending CN111652200A (en) | 2020-08-05 | 2020-08-05 | Processing method, device and equipment for distinguishing multiple vehicles from pictures in vehicle insurance case |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111652200A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112232971A (en) * | 2020-10-14 | 2021-01-15 | 太平金融科技服务（上海）有限公司 | Anti-fraud detection method, anti-fraud detection device, computer equipment and storage medium |
| CN112949751A (en) * | 2021-03-25 | 2021-06-11 | 深圳市商汤科技有限公司 | Vehicle image clustering and track restoring method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103235940A (en) * | 2013-05-06 | 2013-08-07 | 南京新奕天智能视频技术有限公司 | Method for detecting multiple license plates in high-resolution image sequence |
| CN108256541A (en) * | 2016-12-29 | 2018-07-06 | 杭州海康威视数字技术股份有限公司 | Car plate index identification method and device |
| US10255691B2 (en) * | 2016-10-20 | 2019-04-09 | Sun Yat-Sen University | Method and system of detecting and recognizing a vehicle logo based on selective search |
| CN109858484A (en) * | 2019-01-22 | 2019-06-07 | 电子科技大学 | A kind of multi-class transformation VLP correction algorithm based on deflection evaluation |
| CN110263864A (en) * | 2019-06-24 | 2019-09-20 | 北京精英路通科技有限公司 | Matching process, device, computer equipment and the storage medium of vehicle |
| CN110569692A (en) * | 2018-08-16 | 2019-12-13 | 阿里巴巴集团控股有限公司 | multi-vehicle identification method, device and equipment |
| CN111488875A (en) * | 2020-06-24 | 2020-08-04 | 爱保科技有限公司 | Vehicle insurance claim settlement loss checking method and device based on image recognition and electronic equipment |
-
2020
- 2020-08-05 CN CN202010775095.XA patent/CN111652200A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103235940A (en) * | 2013-05-06 | 2013-08-07 | 南京新奕天智能视频技术有限公司 | Method for detecting multiple license plates in high-resolution image sequence |
| US10255691B2 (en) * | 2016-10-20 | 2019-04-09 | Sun Yat-Sen University | Method and system of detecting and recognizing a vehicle logo based on selective search |
| CN108256541A (en) * | 2016-12-29 | 2018-07-06 | 杭州海康威视数字技术股份有限公司 | Car plate index identification method and device |
| CN110569692A (en) * | 2018-08-16 | 2019-12-13 | 阿里巴巴集团控股有限公司 | multi-vehicle identification method, device and equipment |
| CN109858484A (en) * | 2019-01-22 | 2019-06-07 | 电子科技大学 | A kind of multi-class transformation VLP correction algorithm based on deflection evaluation |
| CN110263864A (en) * | 2019-06-24 | 2019-09-20 | 北京精英路通科技有限公司 | Matching process, device, computer equipment and the storage medium of vehicle |
| CN111488875A (en) * | 2020-06-24 | 2020-08-04 | 爱保科技有限公司 | Vehicle insurance claim settlement loss checking method and device based on image recognition and electronic equipment |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112232971A (en) * | 2020-10-14 | 2021-01-15 | 太平金融科技服务（上海）有限公司 | Anti-fraud detection method, anti-fraud detection device, computer equipment and storage medium |
| CN112949751A (en) * | 2021-03-25 | 2021-06-11 | 深圳市商汤科技有限公司 | Vehicle image clustering and track restoring method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596166B (en) | Container number identification method based on convolutional neural network classification | |
| CN107545239B (en) | Fake plate detection method based on license plate recognition and vehicle characteristic matching | |
| CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| CN110991465B (en) | Object identification method, device, computing equipment and storage medium | |
| US8139852B2 (en) | Color classification method, color recognition method, color classification apparatus, color recognition apparatus, color recognition system, computer program, and recording medium | |
| Xie et al. | A robust license plate detection and character recognition algorithm based on a combined feature extraction model and BPNN | |
| CN112699876B (en) | Automatic reading method for various meters of gas collecting station | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN103729631B (en) | Vision-based connector surface feature automatically-identifying method | |
| CN110021028B (en) | Automatic clothing making method based on clothing style drawing | |
| CN110705952A (en) | Contract auditing method and device | |
| CN111639629B (en) | Pig weight measurement method and device based on image processing and storage medium | |
| CN108846831B (en) | Band steel surface defect classification method based on combination of statistical characteristics and image characteristics | |
| Rejeb Sfar et al. | Vantage feature frames for fine-grained categorization | |
| CN108073940B (en) | Method for detecting 3D target example object in unstructured environment | |
| CN111652200A (en) | Processing method, device and equipment for distinguishing multiple vehicles from pictures in vehicle insurance case | |
| CN104751475A (en) | Feature point optimization matching method for static image object recognition | |
| CN110619336A (en) | Goods identification algorithm based on image processing | |
| Forczmański et al. | Stamps detection and classification using simple features ensemble | |
| CN114170418A (en) | Automobile wire harness connector multi-feature fusion image retrieval method by searching images through images | |
| WO2007026951A1 (en) | Image search method and device | |
| CN113963295A (en) | Method, device, equipment and storage medium for recognizing landmark in video clip | |
| CN117636421A (en) | Face deep pseudo detection method based on edge feature acquisition | |
| CN110826551B (en) | Intelligent screening method for VIN code topology printing mode image of motor vehicle | |
| CN103136536A (en) | System and method for detecting target and method for exacting image features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200911 |
|
| RJ01 | Rejection of invention patent application after publication |