CN109684956A - A kind of vehicle damage detection method and system based on deep neural network - Google Patents
A kind of vehicle damage detection method and system based on deep neural network Download PDFInfo
- Publication number
- CN109684956A CN109684956A CN201811531712.0A CN201811531712A CN109684956A CN 109684956 A CN109684956 A CN 109684956A CN 201811531712 A CN201811531712 A CN 201811531712A CN 109684956 A CN109684956 A CN 109684956A
- Authority
- CN
- China
- Prior art keywords
- candidate region
- prospect
- anchor point
- background
- characteristic pattern
- 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
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
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- 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
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle damage detection method and system based on deep neural network, comprising: selection original image extracts characteristic pattern to original image；The anchor point frame of judging characteristic figure is prospect or background, and corrects anchor point frame, obtains candidate region；Candidate region characteristic pattern is obtained based on characteristic pattern and candidate region；Candidate region characteristic pattern is sent into RCNN layers, judges that candidate region for prospect or background, and is corrected in candidate region；Wherein: the anchor point frame of judging characteristic figure is the judgment method of prospect or background are as follows: calculates the intersection of anchor point frame and all mark true value regions and the ratio of anchor point frame, regards as prospect if ratio is higher than threshold value, otherwise regard as background.The present invention has done method improvement when prospect background sample is chosen, and the positive and negative screening sample method of target suitable for a variety of composite vision characteristic morphologies is proposed, so that training process is more easier and stablizes；Higher target recall rate and accuracy can be obtained in vehicle appearance damage check.
Description
Technical field
The present invention relates to vehicle damage detection technique fields, and in particular to a kind of vehicle damage based on deep neural network
Detection method and system.
Background technique
In vehicle appearance picture non-destructive tests system, need to vehicle close shot figure (vehicle appearance picture non-destructive tests mistake
User's shooting at close range vehicle damage picture in journey) damage first carry out target detection, then do type of impairment and degree identification, mesh
The preceding object detection method using classics Two Stage as Faster Rcnn.
Faster Rcnn is the target detection frame proposed in 2016, is still the target detection frame of mainstream up to now
One of frame.In structure, Faster Rcnn is by feature extraction (feature extraction), it is proposed that extracted region
(region proposal), frame returns (bounding box regression) and classification (classification) is all whole
It closes in a network, so that comprehensive performance improves a lot.But Faster Rcnn original, without any changes
It is to be relatively specific for the object target detection of some single vision shapes, such as pedestrian, automobile etc., but be poorly suitable for compound view
Feel the object target detection of form, such as defects detection etc..
The target position that Faster Rcnn detection framework detects all is rectangle, is defined to the positive negative sample of detection target
It is according to IOU method (a usually used concept, the ratio of intersection and union in target detection) Lai Dingyi, by comparing
Threshold value determines whether the candidate region is positive sample, this is for on-fixed visual signature, the object target of complex morphological feature
It is not applicable.And vehicle damage is to be compounded to form process, a kind of damage configuration can be the superposition of a variety of damage configurations, cause to wait
There is mistake in the differentiation of the positive negative sample of favored area.
For example, Vehicular door as shown in Figure 1 damages, the object area in small frame and the IOU index of big frame true value are usual
It is less than threshold value and is judged as negative sample, but same visual signature object area then may in other automobile damage sample
It can be determined as positive sample, therefore IOU method is less suitable.
Summary of the invention
Aiming at the shortcomings existing in the above problems, the present invention provides a kind of vehicle damage based on deep neural network
Detection method and system can preferably improve damage check precision.
The present invention provides a kind of vehicle damage detection method based on deep neural network, comprising:
Original image is selected, characteristic pattern is extracted to original image；
Judge that the anchor point frame of the characteristic pattern for prospect or background, and corrects the anchor point frame, obtains candidate region；
Candidate region characteristic pattern is obtained based on the characteristic pattern and candidate region；
The candidate region characteristic pattern is sent into RCNN layers, judges that candidate regions for prospect or background, and are corrected in candidate region
Domain；
Wherein:
The anchor point frame for judging the characteristic pattern is prospect or the judgment method of background are as follows:
The intersection of the anchor point frame and all mark true value regions and the ratio of the anchor point frame are calculated, if the ratio is high
Prospect is then regarded as in threshold value, otherwise regards as background.
It is as a further improvement of the present invention, described to judge candidate region for prospect or the judgment method of background are as follows:
The intersection of the candidate region and all mark true value regions and the ratio of the candidate region are calculated, if the ratio
Value is higher than threshold value and then regards as prospect, otherwise regards as background.
It is as a further improvement of the present invention, described that characteristic pattern is extracted to original image, comprising:
Vehicle appearance damage detection system extracts frame by VGG or ResNet foundation characteristic to extract vehicle damage image
Characteristic pattern.
As a further improvement of the present invention, the anchor point frame for judging the characteristic pattern is prospect or background, and corrects
The anchor point frame, comprising:
The RPN layer of Faster Rcnn judges that the anchor point frame of the characteristic pattern for prospect or background, and leads to by softmax
It crosses frame and returns the device amendment anchor point frame.
As a further improvement of the present invention, the amendment candidate region, comprising:
Device, which is returned, by frame corrects candidate region.
The vehicle damage detection system based on deep neural network that the present invention also provides a kind of, comprising:
Characteristic extracting module extracts characteristic pattern to original image for selecting original image；
Candidate region extraction module, for judging that the anchor point frame of the characteristic pattern for prospect or background, and corrects the anchor
Point frame, obtains candidate region；The anchor point frame for judging the characteristic pattern is prospect or the judgment method of background are as follows: described in calculating
The intersection in anchor point frame and all mark true value regions and the ratio of the anchor point frame, before being regarded as if the ratio is higher than threshold value
Otherwise scape regards as background；
Pond module, for obtaining candidate region characteristic pattern based on the characteristic pattern and candidate region；
Judgment module, for by the candidate region characteristic pattern be sent into RCNN layers, judge candidate region for prospect or background,
And correct candidate region.
It is as a further improvement of the present invention, described to judge candidate region for prospect or the judgment method of background are as follows:
The intersection of the candidate region and all mark true value regions and the ratio of the candidate region are calculated, if the ratio
Value is higher than threshold value and then regards as prospect, otherwise regards as background.
It is as a further improvement of the present invention, described that characteristic pattern is extracted to original image, comprising:
Vehicle appearance damage detection system extracts frame by VGG or ResNet foundation characteristic to extract vehicle damage image
Characteristic pattern.
As a further improvement of the present invention, the anchor point frame for judging the characteristic pattern is prospect or background, and corrects
The anchor point frame, comprising:
The RPN layer of Faster Rcnn judges that the anchor point frame of the characteristic pattern for prospect or background, and leads to by softmax
It crosses frame and returns the device amendment anchor point frame.
As a further improvement of the present invention, the amendment candidate region, comprising:
Device, which is returned, by frame corrects candidate region.
Compared with prior art, the invention has the benefit that
The present invention has done method improvement when prospect background sample is chosen, and proposes suitable for a variety of composite vision feature shapes
The positive and negative screening sample method of the target of state, so that training process is more easier and stablizes；It can be obtained in vehicle appearance damage check
Higher target recall rate and accuracy are obtained, while being also applied for the unfixed object target detection of some visual signatures.
Detailed description of the invention
Fig. 1 is the damage figure of existing Vehicular door；
Fig. 2 is the process of the vehicle damage detection method based on deep neural network disclosed in an embodiment of the present invention
Figure；
Fig. 3 is the frame of the vehicle damage detection system based on deep neural network disclosed in an embodiment of the present invention
Figure；
Fig. 4 is the schematic diagram that query frame disclosed in an embodiment of the present invention is greater than true value frame；
Fig. 5 is the schematic diagram that query frame disclosed in an embodiment of the present invention is less than true value frame.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
The present invention is described in further detail with reference to the accompanying drawing:
The present invention provides a kind of vehicle damage detection method and system based on deep neural network, proposes a kind of new
Definition damages positive and negative Sample Method SIOQ, SIOQ is defined as: the intersection accumulation in query region and true value region and and query region
Ratio；It can preferably improve damage check precision.
As shown in Fig. 2, the present invention provides a kind of vehicle damage detection method based on deep neural network, comprising:
S1, selection original image, extract characteristic pattern to original image；Specifically:
Vehicle damage picture is selected, vehicle appearance damage detection system extracts frame by foundation characteristics such as VGG or ResNet
Frame (convolutional layer+active coating+pond layer) extracts the characteristic pattern of vehicle damage image, and this feature figure is shared for RPN layers subsequent
Full articulamentum.
S2, judging characteristic figure anchor point frame be prospect or background, and correct anchor point frame, obtain candidate region；Specifically:
The anchor point frame that RPN layer based on Faster Rcnn passes through softmax judging characteristic figure is prospect or background, and is led to
It crosses frame and returns device amendment anchor point frame, to obtain accurate candidate region.
Wherein:
The anchor point frame of judging characteristic figure is the judgment method of prospect or background are as follows:
The intersection of anchor point frame and all mark true value regions and the ratio of anchor point frame are calculated, is assert if ratio is higher than threshold value
For prospect, background is otherwise regarded as.
S3, candidate region characteristic pattern is obtained based on characteristic pattern and candidate region；It is specific:
It is extracted after the information of the characteristic pattern of pond layer collection input and candidate region, comprehensive characteristics figure and candidate region candidate
Provincial characteristics figure.
S4, by candidate region characteristic pattern be sent into RCNN layer, judge candidate region for prospect or background, and pass through frame recurrence
Device corrects candidate region, and it is whole to do position accurate adjustment；Wherein:
Judge candidate region for prospect or the judgment method of background are as follows:
The intersection of candidate region and all mark true value regions and the ratio of candidate region are calculated, if ratio is higher than threshold value
Prospect is regarded as, background is otherwise regarded as.
As shown in figure 3, the present invention provides a kind of vehicle damage detection system based on deep neural network, comprising:
Characteristic extracting module extracts characteristic pattern to original image for selecting original image；Specifically:
Vehicle damage picture is selected, vehicle appearance damage detection system extracts frame by foundation characteristics such as VGG or ResNet
Frame (convolutional layer+active coating+pond layer) extracts the characteristic pattern of vehicle damage image, and this feature figure is shared for RPN layers subsequent
Full articulamentum.
Candidate region extraction module, the anchor point frame for judging characteristic figure is prospect or background, and corrects anchor point frame, is obtained
Candidate region；Specifically:
The anchor point frame that RPN layer based on Faster Rcnn passes through softmax judging characteristic figure is prospect or background, and is led to
It crosses frame and returns device amendment anchor point frame, to obtain accurate candidate region.
Wherein:
The anchor point frame of judging characteristic figure is the judgment method of prospect or background are as follows:
The intersection of anchor point frame and all mark true value regions and the ratio of anchor point frame are calculated, is assert if ratio is higher than threshold value
For prospect, background is otherwise regarded as.
Pond module, for obtaining candidate region characteristic pattern based on characteristic pattern and candidate region；It is specific:
It is extracted after the information of the characteristic pattern of pond layer collection input and candidate region, comprehensive characteristics figure and candidate region candidate
Provincial characteristics figure.
Judgment module judges that candidate region for prospect or background, and leads to for candidate region characteristic pattern to be sent into RCNN layers
It crosses frame and returns device amendment candidate region, it is whole to do position accurate adjustment；Wherein:
Judge candidate region for prospect or the judgment method of background are as follows:
The intersection of candidate region and all mark true value regions and the ratio of candidate region are calculated, if ratio is higher than threshold value
Prospect is regarded as, background is otherwise regarded as.
As shown in figure 4, IOU value is equal to A ∩ C/A ∪ C, SIOQ value etc. when judgement " query frame " belongs to prospect or background
In (A ∩ C+B ∩ C)/C, from fig. 4, it can be seen that the SIOQ value of current queries frame is higher than IOU value, therefore when setting threshold value, IOU
The query frame can be regarded background by value, and SIOQ value is relatively high, by SIOQ value be higher than threshold condition should " query frame " rejecting,
It is not involved in training process.
As shown in figure 5, IOU value is equal to A ∩ C/A ∪ C, SIOQ value etc. when judgement " query frame " belongs to prospect or background
In A ∩ C/A, from fig. 5, it can be seen that the SIOQ value of current queries frame is higher than IOU value, therefore when threshold value is arranged, IOU value can should
Query frame regards background, it is clear that for damage, " query frame " in true value frame cannot be guaranteed be background, pass through
SIOQ value be higher than threshold value should " query frame " rejecting, be not involved in training process.
Therefore, from the point of view of foregoing description, the other sample of background classes is chosen using SIOQ method, and background classes can be effectively reduced
Error sample ratio in not, in the training process, the sample of prospect classification still can calculate selection using IOU method,
When IOU is higher than threshold value, it is somebody's turn to do " query frame " and then centainly belongs to foreground target.
Advantages of the present invention are as follows:
The present invention has done method improvement when prospect background sample is chosen, and proposes suitable for a variety of composite vision feature shapes
The positive and negative screening sample method of the target of state, so that training process is more easier and stablizes；It can be obtained in vehicle appearance damage check
Higher target recall rate and accuracy are obtained, while being also applied for the unfixed object target detection of some visual signatures.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of vehicle damage detection method based on deep neural network characterized by comprising
Original image is selected, characteristic pattern is extracted to original image；
Judge that the anchor point frame of the characteristic pattern for prospect or background, and corrects the anchor point frame, obtains candidate region；
Candidate region characteristic pattern is obtained based on the characteristic pattern and candidate region；
The candidate region characteristic pattern is sent into RCNN layers, judges that candidate region for prospect or background, and is corrected in candidate region；
Wherein:
The anchor point frame for judging the characteristic pattern is prospect or the judgment method of background are as follows:
The intersection of the anchor point frame and all mark true value regions and the ratio of the anchor point frame are calculated, if the ratio is higher than threshold
Value then regards as prospect, otherwise regards as background.
2. the vehicle damage detection method based on deep neural network as described in claim 1, which is characterized in that the judgement
Candidate region is the judgment method of prospect or background are as follows:
The intersection of the candidate region and all mark true value regions and the ratio of the candidate region are calculated, if the ratio is high
Prospect is then regarded as in threshold value, otherwise regards as background.
3. the vehicle damage detection method based on deep neural network as described in claim 1, which is characterized in that described to original
Beginning picture extracts characteristic pattern, comprising:
Vehicle appearance damage detection system extracts frame by VGG or ResNet foundation characteristic to extract the spy of vehicle damage image
Sign figure.
4. the vehicle damage detection method based on deep neural network as described in claim 1, which is characterized in that the judgement
The anchor point frame of the characteristic pattern is prospect or background, and corrects the anchor point frame, comprising:
The RPN layer of Faster Rcnn judges that the anchor point frame of the characteristic pattern for prospect or background, and passes through side by softmax
Frame returns device and corrects the anchor point frame.
5. the vehicle damage detection method based on deep neural network as described in claim 1, which is characterized in that the amendment
Candidate region, comprising:
Device, which is returned, by frame corrects candidate region.
6. a kind of vehicle damage detection system based on deep neural network characterized by comprising
Characteristic extracting module extracts characteristic pattern to original image for selecting original image；
Candidate region extraction module, for judging that the anchor point frame of the characteristic pattern for prospect or background, and corrects the anchor point frame,
Obtain candidate region；The anchor point frame for judging the characteristic pattern is prospect or the judgment method of background are as follows: calculates the anchor point
The intersection in frame and all mark true value regions and the ratio of the anchor point frame, regard as prospect if the ratio is higher than threshold value,
Otherwise background is regarded as；
Pond module, for obtaining candidate region characteristic pattern based on the characteristic pattern and candidate region；
Judgment module judges that candidate region for prospect or background, and is repaired for the candidate region characteristic pattern to be sent into RCNN layers
Positive candidate region.
7. the vehicle damage detection system based on deep neural network as described in claim 1, which is characterized in that the judgement
Candidate region is the judgment method of prospect or background are as follows:
The intersection of the candidate region and all mark true value regions and the ratio of the candidate region are calculated, if the ratio is high
Prospect is then regarded as in threshold value, otherwise regards as background.
8. the vehicle damage detection system based on deep neural network as described in claim 1, which is characterized in that described to original
Beginning picture extracts characteristic pattern, comprising:
Vehicle appearance damage detection system extracts frame by VGG or ResNet foundation characteristic to extract the spy of vehicle damage image
Sign figure.
9. the vehicle damage detection system based on deep neural network as described in claim 1, which is characterized in that the judgement
The anchor point frame of the characteristic pattern is prospect or background, and corrects the anchor point frame, comprising:
The RPN layer of Faster Rcnn judges that the anchor point frame of the characteristic pattern for prospect or background, and passes through side by softmax
Frame returns device and corrects the anchor point frame.
10. the vehicle damage detection system based on deep neural network as described in claim 1, which is characterized in that described to repair
Positive candidate region, comprising:
Device, which is returned, by frame corrects candidate region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811531712.0A CN109684956A (en) | 2018-12-14 | 2018-12-14 | A kind of vehicle damage detection method and system based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811531712.0A CN109684956A (en) | 2018-12-14 | 2018-12-14 | A kind of vehicle damage detection method and system based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109684956A true CN109684956A (en) | 2019-04-26 |
Family
ID=66186090
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811531712.0A Pending CN109684956A (en) | 2018-12-14 | 2018-12-14 | A kind of vehicle damage detection method and system based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109684956A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349124A (en) * | 2019-06-13 | 2019-10-18 | 平安科技（深圳）有限公司 | Vehicle appearance damages intelligent detecting method, device and computer readable storage medium |
| CN110705405A (en) * | 2019-09-20 | 2020-01-17 | 阿里巴巴集团控股有限公司 | Target labeling method and device |
| CN110929632A (en) * | 2019-11-19 | 2020-03-27 | 复旦大学 | Complex scene-oriented vehicle target detection method and device |
| WO2021136939A1 (en) * | 2020-01-03 | 2021-07-08 | Tractable Ltd | Method of determining repair operations for a damaged vehicle |
| WO2021151277A1 (en) * | 2020-05-26 | 2021-08-05 | 平安科技（深圳）有限公司 | Method and apparatus for determining severity of damage on target object, electronic device, and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127747A (en) * | 2016-06-17 | 2016-11-16 | 史方 | Car surface damage classifying method and device based on degree of depth study |
| CN106250812A (en) * | 2016-07-15 | 2016-12-21 | 汤平 | A kind of model recognizing method based on quick R CNN deep neural network |
| CN106845432A (en) * | 2017-02-07 | 2017-06-13 | 深圳市深网视界科技有限公司 | The method and apparatus that a kind of face is detected jointly with human body |
| US20170293894A1 (en) * | 2016-04-06 | 2017-10-12 | American International Group, Inc. | Automatic assessment of damage and repair costs in vehicles |
| CN107358596A (en) * | 2017-04-11 | 2017-11-17 | 阿里巴巴集团控股有限公司 | A kind of car damage identification method based on image, device, electronic equipment and system |
| CN108985343A (en) * | 2018-06-22 | 2018-12-11 | 深源恒际科技有限公司 | Automobile damage detecting method and system based on deep neural network |
-
2018
- 2018-12-14 CN CN201811531712.0A patent/CN109684956A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170293894A1 (en) * | 2016-04-06 | 2017-10-12 | American International Group, Inc. | Automatic assessment of damage and repair costs in vehicles |
| CN106127747A (en) * | 2016-06-17 | 2016-11-16 | 史方 | Car surface damage classifying method and device based on degree of depth study |
| CN106250812A (en) * | 2016-07-15 | 2016-12-21 | 汤平 | A kind of model recognizing method based on quick R CNN deep neural network |
| CN106845432A (en) * | 2017-02-07 | 2017-06-13 | 深圳市深网视界科技有限公司 | The method and apparatus that a kind of face is detected jointly with human body |
| CN107358596A (en) * | 2017-04-11 | 2017-11-17 | 阿里巴巴集团控股有限公司 | A kind of car damage identification method based on image, device, electronic equipment and system |
| CN108985343A (en) * | 2018-06-22 | 2018-12-11 | 深源恒际科技有限公司 | Automobile damage detecting method and system based on deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| AI研习社: ""一文带你了解 Faster R-CNN"", 《HTTPS://WWW.SOHU.COM/A/ 258726784_717210》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349124A (en) * | 2019-06-13 | 2019-10-18 | 平安科技（深圳）有限公司 | Vehicle appearance damages intelligent detecting method, device and computer readable storage medium |
| CN110705405A (en) * | 2019-09-20 | 2020-01-17 | 阿里巴巴集团控股有限公司 | Target labeling method and device |
| CN110929632A (en) * | 2019-11-19 | 2020-03-27 | 复旦大学 | Complex scene-oriented vehicle target detection method and device |
| US11244438B2 (en) | 2020-01-03 | 2022-02-08 | Tractable Ltd | Auxiliary parts damage determination |
| WO2021136945A1 (en) * | 2020-01-03 | 2021-07-08 | Tractable Ltd | Vehicle damage state determination method |
| WO2021136940A1 (en) * | 2020-01-03 | 2021-07-08 | Tractable Ltd | Method of determining damage to auxiliary parts of a vehicle |
| WO2021136944A1 (en) * | 2020-01-03 | 2021-07-08 | Tractable Ltd | Method of universal automated verification of vehicle damage |
| WO2021136939A1 (en) * | 2020-01-03 | 2021-07-08 | Tractable Ltd | Method of determining repair operations for a damaged vehicle |
| US11250554B2 (en) | 2020-01-03 | 2022-02-15 | Tractable Ltd | Repair/replace and labour hours determination |
| US11257204B2 (en) | 2020-01-03 | 2022-02-22 | Tractable Ltd | Detailed damage determination with image segmentation |
| US11257203B2 (en) | 2020-01-03 | 2022-02-22 | Tractable Ltd | Inconsistent damage determination |
| US11361426B2 (en) | 2020-01-03 | 2022-06-14 | Tractable Ltd | Paint blending determination |
| US11386543B2 (en) | 2020-01-03 | 2022-07-12 | Tractable Ltd | Universal car damage determination with make/model invariance |
| US11587221B2 (en) | 2020-01-03 | 2023-02-21 | Tractable Limited | Detailed damage determination with image cropping |
| US11636581B2 (en) | 2020-01-03 | 2023-04-25 | Tractable Limited | Undamaged/damaged determination |
| WO2021151277A1 (en) * | 2020-05-26 | 2021-08-05 | 平安科技（深圳）有限公司 | Method and apparatus for determining severity of damage on target object, electronic device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109684956A (en) | A kind of vehicle damage detection method and system based on deep neural network | |
| CN107729818B (en) | Multi-feature fusion vehicle re-identification method based on deep learning | |
| CN110378236B (en) | Vehicle identity recognition model construction and recognition method and system based on deep learning | |
| CN106127747B (en) | Car surface damage classifying method and device based on deep learning | |
| CN105975941B (en) | A kind of multi-direction vehicle detection identifying system based on deep learning | |
| CN105373135B (en) | A kind of method and system of aircraft docking guidance and plane type recognition based on machine vision | |
| CN111898523A (en) | Remote sensing image special vehicle target detection method based on transfer learning | |
| CN105447490B (en) | Vehicle critical point detection method and device based on gradient regression tree | |
| CN105046196A (en) | Front vehicle information structured output method base on concatenated convolutional neural networks | |
| CN110969160A (en) | License plate image correction and recognition method and system based on deep learning | |
| CN101702197A (en) | Method for detecting road traffic signs | |
| CN106778540B (en) | Parking detection is accurately based on the parking event detecting method of background double layer | |
| CN103413046B (en) | Statistical method of traffic flow | |
| CN109508731A (en) | A kind of vehicle based on fusion feature recognition methods, system and device again | |
| CN109543753B (en) | License plate recognition method based on self-adaptive fuzzy repair mechanism | |
| WO2023155483A1 (en) | Vehicle type identification method, device, and system | |
| CN105389561B (en) | A kind of bus zone detection method based on video | |
| CN108491498A (en) | A kind of bayonet image object searching method based on multiple features detection | |
| CN106815583A (en) | A kind of vehicle at night license plate locating method being combined based on MSER and SWT | |
| CN109801220A (en) | Mapping parameters method in a kind of splicing of line solver Vehicular video | |
| CN110084111A (en) | A kind of quick vehicle detection at night method applied to adaptive high beam | |
| CN107944354A (en) | A kind of vehicle checking method based on deep learning | |
| CN104751190A (en) | Vehicle part positioning method for vehicle fine recognition | |
| CN109086803A (en) | A kind of haze visibility detection system and method based on deep learning and the personalized factor | |
| CN109118514A (en) | A kind of method for tracking target |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190426 |