CN109948416A - A kind of illegal occupancy bus zone automatic auditing method based on deep learning - Google Patents
A kind of illegal occupancy bus zone automatic auditing method based on deep learning Download PDFInfo
- Publication number
- CN109948416A CN109948416A CN201811654484.6A CN201811654484A CN109948416A CN 109948416 A CN109948416 A CN 109948416A CN 201811654484 A CN201811654484 A CN 201811654484A CN 109948416 A CN109948416 A CN 109948416A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- bus zone
- target vehicle
- frame
- network
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of illegal automatic auditing system of occupancy bus zone based on deep learning, the following steps are included: carry out shooting evidence obtaining for the illegal target vehicle for occupying bus zone using bayonet camera, forensic information includes the license plate number information of n frame (n >=1) evidence figure and target vehicle.Using all vehicles in yolo-v2 vehicle detection model inspection n frame image, license plate number is identified to all vehicle application caffe-ssd model inspection license plate areas detected, and using lstm+ctc model；Position the position of target vehicle in n frame image respectively by comparing the vehicle weight recognizer of the editing distance and application of these license plate numbers and given license plate based on GoogLenet Inception-V2 network later；It recycles deeplab-v2 partitioning algorithm to be partitioned into image bus zone region, accounts for the ratio of target vehicle detection frame by calculating target vehicle detection frame and the intersection of bus zone region, judge whether target vehicle occupies bus zone；Hereafter vehicle classification network of the application based on ResNet18 identifies target vehicle classification, finally judges the whether illegal occupancy bus zone of the target vehicle according to target vehicle classification information and occupancy bus zone situation.The present invention saves police strength, improves illegal review efficiency and accuracy, ensure that the fair and just of audit.
Description
Technical field
The present invention relates to the intelligent images such as target detection, scene cut to identify field, in particular to a kind of to be based on depth
Practise the method and system that automation audit is realized to illegal occupancy bus zone.
Background technique
In recent years, the growth of domestic automobile ownership continuously and healthily brings stern challenge to traffic safety, hand over
Therefore the illegal examination of logical department faces immense pressure.As the important component of break in traffic rules and regulations audit, illegal occupancy
The audit violating the regulations of bus zone occupies the biggish workload of audit violating the regulations concerning public transport efficiency and safety.It is existing illegal
The checking method violating the regulations for occupying bus zone mainly still uses manual examination and verification mode, includes that cost of labor is higher there are problem, imitates
Rate is too low, audit is tired and personnel's subjectivity influences the fair and just property audited.Therefore seek intelligent automaticization audit mode
Substitute or assist existing manual examination and verification to become more more and more urgent.
Summary of the invention
The purpose of the present invention is: propose a kind of intelligent checks that illegal occupancy bus zone is carried out based on deep learning method
Method carries out intelligent checks for illegal occupancy bus zone vehicle pictures, to promote review efficiency and accuracy rate.
The technical solution adopted by the present invention to solve the technical problems is: the illegal occupancy bus zone based on deep learning
Intelligent checks system, it includes the following steps:
S1, obtain one group (n frame, n >=1) pending illegal occupancys bus zone picture and illegal vehicle license plate number believe
Breath；
S2, all vehicles on n frame image are detected using the vehicle detecting algorithm of yolo-v2；
S3, using caffe-ssd detection n frame image all vehicles license plate position, and known using lstm+ctc network
Other license plate number information；
S4, the editing distance for calculating S3 license plate number information and S1 license plate number information, take in n frame image less than threshold value respectively
Vehicle location corresponding to the smallest edit distance of condition, i.e. positioning target vehicle；
S5, to the picture frame of S4 no-fix to target vehicle, extracted using the network of GoogLenet Inception-V2
256 dimensional features of all vehicles of the frame calculate COS distance with 256 dimensional features of the target vehicle navigated in S4, thus real
The target vehicle tracking and positioning of existing n frame image；
S6, usage scenario partitioning algorithm deeplab-v2 are partitioned into bus zone regional location；
S7, the intersection for calculating separately target vehicle detection frame and bus zone region in n frame image, to judge that vehicle is
No occupancy bus zone；If the area ratio in n frame at least in the presence of a frame intersection area and target vehicle detection frame is greater than
Given threshold then determines target vehicle in bus zone, that is, is judged to occupy bus zone, otherwise be judged to vacant；
S8, the vehicle classification network based on ResNet18, identification target vehicle classification, such as private car, taxi, public affairs are used
Hand over vehicle, police car, ambulance etc.；
S9, the judgement that bus zone whether is occupied according to S7 target vehicle, combining target class of vehicle information provide final
Auditing result；If the vacant bus zone of target vehicle or target vehicle occupy bus zone and this vehicle is that bus etc. is not made
Penalty vehicle, then be judged to audit it is not illegal, be otherwise judged to audit illegal；
Further, target vehicle track algorithm of the S5 based on GoogLenet Inception-V2 network includes step
It is rapid as follows:
S10, training GoogLenet Inception-V2 feature extraction network, last full articulamentum dimension, which is arranged, is
256 dimensions, the full articulamentum connects a classification layer when training, and classification layer classifies to different money vehicles, and each classification has collected
With photo of the vehicle under different moments different angle.Using general data amplification mode training network, when trained loss
When value loss is reduced to minimum, classification layer is cropped, retains the 256 full articulamentums of dimension, 256 dimensional features obtained at this time can be fine
Characterization the vehicle feature.
S11, the target vehicle for navigating to S4 are input to GoogLenet Inception-V2 network, in the defeated of the network
Enter layer, padding is carried out to the vehicle of input, becomes the consistent image of length and width, extra part is with 0 pixel filling；Then right
Pretreated image carries out up-sampling or down-sampling operation, and unified resize is finally obtained at the image of 200*200 resolution ratio
To 256 dimensional feature vectors；
S12, all vehicles to be matched in m (0 < m < n) a picture frame of S4 no-fix to target vehicle are inputted
GoogLenet Inception-V2 network, same S11 obtain several 256 dimensional feature vectors of m group；
S13, each group of several 256 dimensional feature vectors of m group in S12 are successively tieed up with 256 of target vehicle in S11 respectively
Feature vector calculates cosine similarity, obtains 256 dimensional features corresponding to the similarity maximum in each group of m group respectively.Due to
256 dimensional feature vectors illustrate that the vehicle more more dissimilar vector angle of the smaller and opposite vehicle of similarity vector angle is bigger, institute
It can reflect the similarity degree of vehicle with cosine similarity.
S14, due to having detected several vehicles respectively in m frame image, find similarity in m frame respectively with above-mentioned algorithm
The vehicle of highest scoring, vehicle call number corresponding to highest scoring are the vehicle traced into.
Further, the deeplab-v2 segmentation bus zone algorithm of the S6 comprises the following steps that
S15, the bus zone picture for collecting application scenarios, and manually mark out bus zone region, i.e., artificial mark packet
Enclose the closed polygon of bus zone；
S16, artificial mark is converted into label matrix, i.e., it is all pixels point in the closed polygon manually marked is corresponding
Label is set as 1, and the corresponding label of other pixels is set as 0；
S17, bus zone picture and corresponding label matrix input deeplab-v2 partitioning algorithm are trained, deeplab-
For v2 partitioning algorithm using ResNet-34 as backbone network, psp_module and unet module uses skip as decoder
Layer introduces low-dimensional minutia as prototype network structure.Use a*bce_loss+b*lovasz_loss as finally
Loss (0≤a, b≤1 are manually set), and introduce auxiliary loss aux_loss and be trained；
The good deeplab-v2 partitioning algorithm of S18, application training predicts input image pixels point classification, will belong to bus
The pixel coordinate set of road classification exports, to realize the segmentation in bus zone region；
Further, the vehicle classification network algorithm based on ResNet18 of the S8 comprises the following steps that
S19, collect application scenarios vehicle, classification include private car, taxi, passenger train, freight, bus, police car,
Ambulance etc. constitutes model training collection；
S20, the vehicle classification network based on ResNet18, the classification number of setting softmax classification layer vehicle, progress are used
Class of vehicle classification based training；
The vehicle pictures of the good ResNet18 vehicle classification neural network forecast input of S21, application training, export the vehicle of prediction
Classification；
The beneficial effects of the present invention are: present invention is mainly applied to occupancy bus zones illegal in traffic offence auditing system
In automatic audit, the present invention can carry out fully-automatic intelligent audit according to the auditing rule formulated in advance, when due to network training
Used the training data of various complex situations scenes, thus network segmentation and recognition result for night and fuzzy figure
Piece has good robustness, and compared with manual examination and verification, algorithm shows better effect under special circumstances, and when auditing
Between greatly shorten, whole flow process can be completed within 1 second；
Detailed description of the invention
Fig. 1 is illegal occupancy bus zone intelligent checks system flow chart of the invention
Fig. 2 is the present invention by calculating editing distance positioning target vehicle schematic diagram
Fig. 3 is that the vehicle weight recognizer positioning target vehicle of the invention based on GoogLenet Inception-V2 shows
It is intended to
Fig. 4 is the illegal occupancy bus zone schematic diagram of the present invention
Specific embodiment
Below in conjunction with attached drawing, the invention will be further described.
As shown in figure 4, bus zone schematic diagram is occupied for vehicle illegal, yellow " bus zone " printed words in the schematic diagram
Denoting lane where it is bus zone, non-bus etc. do not make penalty vehicle travel or stop in the lane be considered as it is illegal
Occupy bus zone.
The illegal automatic auditing system of occupancy bus zone based on deep learning of the invention mainly include vehicle identify again,
Scene cut, vehicle classification and logic judgment algoritic module.
As shown in Fig. 2, the process using editing distance to 1 frame framing target vehicle is schematically illustrated, it is practical to n frame evidence
Figure requires to undergo this process respectively.Picture frame A is the wherein frame image in n frame evidence figure, is detected first by yolo-v2
Then all vehicle images of frame image out using caffe-ssd detection license plate and make the every width vehicle image that detected
With lstm+ctc Network Recognition license plate number, it is at most only able to detect a license plate on a vehicle, so can find only by license plate
One vehicle determined.
Editing distance is the common quantizer method for measurement for two character string difference degrees, which can only pass through
Increase, delete, replacing three kinds of operations, seeing and at least how many times is needed to operate a character string could be become another character string in this way,
Operand is fewer under identical characters string length, illustrates that two character strings are more similar.Therefore this method can effectively quantify two vehicles
The similarity degree of trade mark information.
Assuming that common recognition is clipped to x license plate images in picture frame A, by calculating the step of editing distance positions target vehicle
It is as follows:
1, successively by picture frame A x license plate number and the license plate number that obtains of bayonet calculate editing distance, obtain x away from
From numerical value；
2, minimum value is taken in x distance values and compared with given threshold；
If 3, minimum value is smaller than threshold value, the corresponding license plate of the minimum value is got, unique vehicle inspection is determined with this license plate
Frame is surveyed, that is, has navigated to the target vehicle in picture frame A；
If 4, minimum value is bigger than threshold value, illustrate to can not find license plate number similar with bayonet acquisition license plate number in picture frame A, because
The positioning failure of this picture frame A target vehicle
If picture frame A no-fix target vehicle, needs subsequent applications based on the vehicle of GoogLenet Inception-V2
Weight recognizer identifies its all vehicle detected again, relocates target vehicle.
1 frame image is known again using the vehicle based on GoogLenet Inception-V2 as shown in figure 3, illustrating
The process of other algorithm positioning target vehicle.Vehicle weight recognizer by extract certain frame image in all vehicle characteristics vectors and according to
It is secondary to calculate with the cosine similarity of target vehicle feature the target vehicle for determining the frame.Assuming that picture frame A and picture frame B are to belong to
Wherein two field pictures in n frame evidence image, the target vehicle in picture frame A are above by the editing distance for calculating license plate
It navigates to, and picture frame B fails to navigate to target carriage in the above manner, it is therefore desirable to vehicle identifies again relocates target
Vehicle.Vehicle is identified again and is comprised the following steps:
1, the target vehicle of picture frame A is by extracting one 256 dimension based on GoogLenet Inception-V2 network
Feature vector
2, the x width vehicle image that all yolo-v2 of picture frame B are detected is by being based on GoogLenet Inception-V2
Network extracts 256 dimensional feature vector of x group
3, successively that 256 dimensional feature vector of x group in step 2 is similar to 256 dimensional feature vectors of target vehicle calculating cosine
Degree, obtains x group similarity numerical value
4, the maximum value in x group similarity numerical value is taken, and by it compared with given threshold
5, when maximum value is greater than given threshold in step 4, the corresponding vehicle detection frame of the maximum value is got, that is, navigates to figure
As the target vehicle in frame B
6, when maximum value is less than given threshold in step 4, illustrate to be not present and picture frame A target vehicle phase in picture frame B
As vehicle, therefore in picture frame B target vehicle positioning failure.
It is artificial since scene cut is the classification to pixel for deeplab-v2 Algorithm of Scene module
It needs to be close to cutting object profile as far as possible when labeled data to be labeled, mark difficulty is larger；Simultaneously as segmentation was trained
Journey is to make classification based training to each pixel, therefore it is very big to divide training expense.The present invention marks workload to reduce data,
And training for promotion efficiency, segmentation training is realized using transfer learning strategy, is comprised the following steps that
1, bus zone region image data collection is expanded using generic way, amplification method includes rotation, translates, is random
Cut, change luminance contrast tone, increase random noise, Gaussian Blur etc., to enrich the diversity of sample；
2, to the data set after amplification, proportionally 9:1 is divided, and is divided into detection data subset and segmentation data subset two
Point, it will test data subset and be used as detection mark, i.e., surround bus zone region using rectangle frame；Segmentation data subset is used as
Segmentation mark is bonded bus zone profile mark using polygon as far as possible；
3, the network structure of design segmentation and detection, makes two task models have public base network portion；
4, using the detection network in detection data trained step 3, training is completed, public base network parameter is obtained
5, public base network parameter is locked, segmentation network unique portion is trained, to training to restraining, decontrols locking
Public base network parameter continue training until convergence.Such training method reaches identical compared with conventional directly training method
In the case where training effect, segmentation data mark amount is original 1/10 can be reduced, the training time saves be original half with
On.
6, in training data, there are the unbalanced situations of data volume of all categories, can obtain by adjusting loss weighted value flat
Weighing apparatus.The distribution that background in image probably meets 8:1 than bus zone area pixel quantitative proportion is marked, such as in order to make to divide network
The bus zone smaller to accounting is partitioned into preferable effect, and the loss weight for needing to adjust both this classification is 0.1:0.8.
For the wherein frame image in n frame evidence figure, the target carriage in picture frame is navigated to by above method
, logic judgment is made into the bus zone region that target vehicle and Algorithm of Scene are divided, so as to judge this target
Whether vehicle combines the target vehicle type information of vehicle classification algorithm identification to can determine whether the target carriage later in bus zone
Whether illegal occupancy bus zone.It is required according to audit, meets illegal condition simply by the presence of a frame image in n frame evidence figure
It can be judged to illegal.
Implementation detailed process of the invention is as shown in Figure 1, a kind of illegal occupancy bus zone based on deep learning is automatic
Auditing system, comprising the following steps:
S1, obtain one group (n frame, n >=1) pending illegal occupancys bus zone picture and illegal vehicle license plate number believe
Breath；
S2, all vehicles on n frame image are detected using the vehicle detecting algorithm of yolo-v2；
S3, using caffe-ssd detection n frame image all vehicles license plate position, and known using lstm+ctc network
Other license plate number information；
S4, the editing distance for calculating S3 license plate number information and S1 license plate number information, take in n frame image less than threshold value respectively
Vehicle location corresponding to the smallest edit distance of condition, i.e. positioning target vehicle；
S5, to the picture frame of S4 no-fix to target vehicle, extracted using the network of GoogLenet Inception-V2
256 dimensional features of all vehicles of the frame calculate COS distance with 256 dimensional features of the target vehicle navigated in S4, thus real
The target vehicle tracking and positioning of existing n frame image；
S6, usage scenario partitioning algorithm deeplab-v2 are partitioned into bus zone regional location；
S7, the intersection for calculating separately target vehicle detection frame and bus zone region in n frame image, to judge that vehicle is
No occupancy bus zone；If the area ratio in n frame at least in the presence of a frame intersection area and target vehicle detection frame is greater than
Given threshold then determines target vehicle in bus zone, that is, is judged to occupy bus zone, otherwise be judged to vacant；
S8, the vehicle classification network based on ResNet18, identification target vehicle classification, such as private car, taxi, public affairs are used
Hand over vehicle, police car, ambulance etc.；
S9, the judgement that bus zone whether is occupied according to S7 target vehicle, combining target class of vehicle information provide final
Auditing result；If the vacant bus zone of target vehicle or target vehicle occupy bus zone and this vehicle is that bus etc. is not made
Penalty vehicle, then be judged to audit it is not illegal, be otherwise judged to audit illegal；
The advantages of basic principles and main features and this programme of this programme have been shown and described above.The technology of the industry
Personnel are it should be appreciated that this programme is not restricted to the described embodiments, and the above embodiments and description only describe this
The principle of scheme, under the premise of not departing from this programme spirit and scope, this programme be will also have various changes and improvements, these changes
Change and improvement is both fallen within the scope of claimed this programme.This programme be claimed range by appended claims and its
Equivalent thereof.
Claims (4)
1. a kind of illegal occupancy bus zone automatic auditing method based on deep learning, it includes the following steps:
S1, the pending illegal occupancy bus zone picture for obtaining one group of n frame, wherein n is more than or equal to 1 and illegal vehicle license plate
Number information；
S2, all vehicles on n frame image are detected using the vehicle detecting algorithm of yolo-v2；
S3, using caffe-ssd detection n frame image all vehicles license plate position, and use lstm+ctc Network Recognition vehicle
Trade mark information；
S4, the editing distance for calculating S3 license plate number information and S1 license plate number information, take in n frame image less than threshold condition respectively
Smallest edit distance corresponding to vehicle location, i.e., positioning target vehicle；
S5, to the picture frame of S4 no-fix to target vehicle, which is extracted using the network of GoogLenetInception-V2
There are 256 dimensional features of vehicle, COS distance is calculated with 256 dimensional features of the target vehicle navigated in S4, to realize n frame figure
The target vehicle tracking and positioning of picture；
S6, usage scenario partitioning algorithm deeplab-v2 are partitioned into bus zone regional location；
S7, the intersection for calculating separately target vehicle detection frame and bus zone region in n frame image, to judge whether vehicle accounts for
Use bus zone；If at least there is a frame intersection area in n frame and the area ratio of target vehicle detection frame be greater than setting
Threshold value then determines target vehicle in bus zone, that is, is judged to occupy bus zone, otherwise be judged to vacant；
S8, the vehicle classification network based on ResNet18, identification target vehicle classification, such as private car, taxi, public transport are used
Vehicle, police car, ambulance etc.；
S9, the judgement that bus zone whether is occupied according to S7 target vehicle, combining target class of vehicle information, provide final review
As a result；If the vacant bus zone of target vehicle or target vehicle occupy bus zone and this vehicle is that bus etc. does not make penalty
Vehicle, then be judged to audit it is not illegal, be otherwise judged to audit illegal.
2. a kind of illegal occupancy bus zone automatic auditing method based on deep learning as described in claim 1, feature
It is, the S5 is comprised the following steps that based on the target vehicle track algorithm of GoogLenetInception-V2 network
S10, training GoogLenetInception-V2 feature extraction network, it is 256 dimensions that last full articulamentum dimension, which is arranged,
The full articulamentum connects a classification layer when training, and classification layer classifies to different money vehicles, and each classification has collected with vehicle
Photo under different moments different angle, using general data amplification mode training network, as trained penalty values loss
When being reduced to minimum, classification layer is cropped, retains the 256 full articulamentums of dimension, 256 dimensional features obtained at this time can be good at characterizing
The feature of the vehicle,
S11, the target vehicle for navigating to S4 are input to GoogLenet Inception-V2 network, in the input of the network
Layer carries out padding to the vehicle of input, becomes the consistent image of length and width, extra part is with 0 pixel filling；Then to pre-
Treated, and image carries out up-sampling or down-sampling operation, and unified resize is finally obtained at the image of 200*200 resolution ratio
One 256 dimensional feature vector；
S12, all vehicles to be matched in m picture frame of S4 no-fix to target vehicle are inputted
GoogLenetInception-V2 network, wherein m is greater than 0, n and is greater than m, same to S11, obtains several 256 dimensional feature vectors of m group；
S13, respectively to each group of several 256 dimensional feature vectors of m group in S12 successively 256 dimensional features with target vehicle in S11
Vector calculates cosine similarity, 256 dimensional features corresponding to the similarity maximum in each group of m group is obtained respectively, due to 256 dimensions
Feature vector illustrates that the vehicle more more dissimilar vector angle of the smaller and opposite vehicle of similarity vector angle is bigger, so cosine
Similarity can reflect the similarity degree of vehicle,
S14, due to having detected several vehicles respectively in m frame image, find similarity score in m frame respectively with above-mentioned algorithm
Highest vehicle, vehicle call number corresponding to highest scoring are the vehicle traced into.
3. a kind of illegal occupancy bus zone automatic auditing method based on deep learning as described in claim 1, feature
It is, the deeplab-v2 segmentation bus zone algorithm of the S6 comprises the following steps that
S15, the bus zone picture for collecting application scenarios, and manually mark out bus zone region, i.e., artificial mark surrounds public
Hand over the closed polygon in lane；
S16, artificial mark is converted into label matrix, i.e., by all pixels point corresponding label in the closed polygon manually marked
It is set as 1, the corresponding label of other pixels is set as 0；
S17, bus zone picture and the input deeplab-v2 partitioning algorithm training of corresponding label matrix, deeplab-v2 are divided
Algorithm is cut using ResNet-34 as backbone network, psp_module and unet module uses skip as decoder
Layer introduces low-dimensional minutia as prototype network structure, uses a*bce_loss+b*lovasz_loss as finally
Loss, wherein a and b is both greater than equal to 0；A and b is both less than equal to 1；And it introduces auxiliary loss aux_loss and is trained；
The good deeplab-v2 partitioning algorithm of S18, application training predicts input image pixels point classification, will belong to bus zone class
Other pixel coordinate set output, to realize the segmentation in bus zone region.
4. a kind of illegal automatic auditing method of vehicle crimping based on deep learning as described in claim 1, which is characterized in that
The vehicle classification network algorithm based on ResNet18 of the S8 comprises the following steps that
S19, application scenarios vehicle is collected, classification includes private car, taxi, passenger train, freight, bus, police car, rescue
Vehicle etc. constitutes model training collection；
S20, the vehicle classification network based on ResNet18, the classification number of setting softmax classification layer vehicle, progress vehicle are used
Category classification training；
The vehicle pictures of the good ResNet18 vehicle classification neural network forecast input of S21, application training, export the vehicle class of prediction
Not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811654484.6A CN109948416A (en) | 2018-12-31 | 2018-12-31 | A kind of illegal occupancy bus zone automatic auditing method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811654484.6A CN109948416A (en) | 2018-12-31 | 2018-12-31 | A kind of illegal occupancy bus zone automatic auditing method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109948416A true CN109948416A (en) | 2019-06-28 |
Family
ID=67007171
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811654484.6A Pending CN109948416A (en) | 2018-12-31 | 2018-12-31 | A kind of illegal occupancy bus zone automatic auditing method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948416A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110415529A (en) * | 2019-09-04 | 2019-11-05 | 上海眼控科技股份有限公司 | Automatic processing method, device, computer equipment and the storage medium of vehicle violation |
| CN110706261A (en) * | 2019-10-22 | 2020-01-17 | 上海眼控科技股份有限公司 | Vehicle violation detection method and device, computer equipment and storage medium |
| CN110718071A (en) * | 2019-10-25 | 2020-01-21 | 上海眼控科技股份有限公司 | Verification method and device for image acquisition equipment, computer equipment and storage medium |
| CN110826415A (en) * | 2019-10-11 | 2020-02-21 | 上海眼控科技股份有限公司 | Method and device for re-identifying vehicles in scene image |
| CN110909742A (en) * | 2019-11-19 | 2020-03-24 | 上海眼控科技股份有限公司 | License plate detection method, system, platform and storage medium |
| CN110969860A (en) * | 2019-12-11 | 2020-04-07 | 上海眼控科技股份有限公司 | Background auditing system and method for traffic law violation behaviors |
| CN111126406A (en) * | 2019-12-17 | 2020-05-08 | 北京四维图新科技股份有限公司 | Vehicle driving area identification method and device |
| CN111275982A (en) * | 2020-01-22 | 2020-06-12 | 于志国 | Illegal vehicle big data reporting system based on lane detection |
| CN111368687A (en) * | 2020-02-28 | 2020-07-03 | 成都市微泊科技有限公司 | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation |
| CN111382704A (en) * | 2020-03-10 | 2020-07-07 | 北京以萨技术股份有限公司 | Vehicle line-pressing violation judgment method and device based on deep learning and storage medium |
| CN111768630A (en) * | 2019-12-24 | 2020-10-13 | 上海高德威智能交通系统有限公司 | Violation waste image detection method and device and electronic equipment |
| CN112329724A (en) * | 2020-11-26 | 2021-02-05 | 四川大学 | Real-time detection and snapshot method for lane change of motor vehicle |
| CN112712708A (en) * | 2020-12-28 | 2021-04-27 | 上海眼控科技股份有限公司 | Information detection method, device, equipment and storage medium |
| CN113297910A (en) * | 2021-04-25 | 2021-08-24 | 云南电网有限责任公司信息中心 | Distribution network field operation safety belt identification method |
| CN113642412A (en) * | 2021-07-16 | 2021-11-12 | 盛视科技股份有限公司 | Method, device and equipment for detecting vehicles occupying bus lane |
| CN113851001A (en) * | 2021-09-17 | 2021-12-28 | 同济大学 | Automatic auditing method, system, device and storage medium for multilane merging violation |
| CN113989794A (en) * | 2021-11-12 | 2022-01-28 | 珠海安联锐视科技股份有限公司 | License plate detection and recognition method |
| CN114373139A (en) * | 2022-01-07 | 2022-04-19 | 北京远度互联科技有限公司 | Method, device, electronic equipment and storage medium for identifying road violation |
| CN114495109A (en) * | 2022-01-24 | 2022-05-13 | 山东大学 | Grabbing robot based on matching of target and scene characters and grabbing method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160034778A1 (en) * | 2013-12-17 | 2016-02-04 | Cloud Computing Center Chinese Academy Of Sciences | Method for detecting traffic violation |
| CN105740855A (en) * | 2016-03-24 | 2016-07-06 | 博康智能信息技术有限公司 | Front and rear license plate detection and recognition method based on deep learning |
| CN107085696A (en) * | 2016-10-15 | 2017-08-22 | 安徽百诚慧通科技有限公司 | A kind of vehicle location and type identifier method based on bayonet socket image |
| CN107730905A (en) * | 2017-06-13 | 2018-02-23 | 银江股份有限公司 | Multitask fake license plate vehicle vision detection system and method based on depth convolutional neural networks |
| WO2018130016A1 (en) * | 2017-01-10 | 2018-07-19 | 哈尔滨工业大学深圳研究生院 | Parking detection method and device based on monitoring video |
-
2018
- 2018-12-31 CN CN201811654484.6A patent/CN109948416A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160034778A1 (en) * | 2013-12-17 | 2016-02-04 | Cloud Computing Center Chinese Academy Of Sciences | Method for detecting traffic violation |
| CN105740855A (en) * | 2016-03-24 | 2016-07-06 | 博康智能信息技术有限公司 | Front and rear license plate detection and recognition method based on deep learning |
| CN107085696A (en) * | 2016-10-15 | 2017-08-22 | 安徽百诚慧通科技有限公司 | A kind of vehicle location and type identifier method based on bayonet socket image |
| WO2018130016A1 (en) * | 2017-01-10 | 2018-07-19 | 哈尔滨工业大学深圳研究生院 | Parking detection method and device based on monitoring video |
| CN107730905A (en) * | 2017-06-13 | 2018-02-23 | 银江股份有限公司 | Multitask fake license plate vehicle vision detection system and method based on depth convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 全晓艳: "车辆违章跟踪识别方法的研究与优化", 《计算机仿真》 * |
| 韩功等: "使用物体交互模型的车辆违停事件检测", 《电视技术》 * |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110415529A (en) * | 2019-09-04 | 2019-11-05 | 上海眼控科技股份有限公司 | Automatic processing method, device, computer equipment and the storage medium of vehicle violation |
| CN110415529B (en) * | 2019-09-04 | 2021-09-28 | 上海眼控科技股份有限公司 | Automatic processing method and device for vehicle violation, computer equipment and storage medium |
| CN110826415A (en) * | 2019-10-11 | 2020-02-21 | 上海眼控科技股份有限公司 | Method and device for re-identifying vehicles in scene image |
| CN110706261A (en) * | 2019-10-22 | 2020-01-17 | 上海眼控科技股份有限公司 | Vehicle violation detection method and device, computer equipment and storage medium |
| CN110718071B (en) * | 2019-10-25 | 2021-09-03 | 上海眼控科技股份有限公司 | Verification method and device for image acquisition equipment, computer equipment and storage medium |
| CN110718071A (en) * | 2019-10-25 | 2020-01-21 | 上海眼控科技股份有限公司 | Verification method and device for image acquisition equipment, computer equipment and storage medium |
| CN110909742A (en) * | 2019-11-19 | 2020-03-24 | 上海眼控科技股份有限公司 | License plate detection method, system, platform and storage medium |
| CN110969860A (en) * | 2019-12-11 | 2020-04-07 | 上海眼控科技股份有限公司 | Background auditing system and method for traffic law violation behaviors |
| CN111126406A (en) * | 2019-12-17 | 2020-05-08 | 北京四维图新科技股份有限公司 | Vehicle driving area identification method and device |
| CN111126406B (en) * | 2019-12-17 | 2023-04-07 | 北京四维图新科技股份有限公司 | Vehicle driving area identification method and device |
| CN111768630A (en) * | 2019-12-24 | 2020-10-13 | 上海高德威智能交通系统有限公司 | Violation waste image detection method and device and electronic equipment |
| CN111275982A (en) * | 2020-01-22 | 2020-06-12 | 于志国 | Illegal vehicle big data reporting system based on lane detection |
| CN111368687A (en) * | 2020-02-28 | 2020-07-03 | 成都市微泊科技有限公司 | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation |
| CN111382704A (en) * | 2020-03-10 | 2020-07-07 | 北京以萨技术股份有限公司 | Vehicle line-pressing violation judgment method and device based on deep learning and storage medium |
| CN111382704B (en) * | 2020-03-10 | 2023-12-15 | 以萨技术股份有限公司 | Vehicle line pressing violation judging method and device based on deep learning and storage medium |
| CN112329724B (en) * | 2020-11-26 | 2022-08-05 | 四川大学 | Real-time detection and snapshot method for lane change of motor vehicle |
| CN112329724A (en) * | 2020-11-26 | 2021-02-05 | 四川大学 | Real-time detection and snapshot method for lane change of motor vehicle |
| CN112712708A (en) * | 2020-12-28 | 2021-04-27 | 上海眼控科技股份有限公司 | Information detection method, device, equipment and storage medium |
| CN113297910A (en) * | 2021-04-25 | 2021-08-24 | 云南电网有限责任公司信息中心 | Distribution network field operation safety belt identification method |
| CN113297910B (en) * | 2021-04-25 | 2023-04-18 | 云南电网有限责任公司信息中心 | Distribution network field operation safety belt identification method |
| CN113642412A (en) * | 2021-07-16 | 2021-11-12 | 盛视科技股份有限公司 | Method, device and equipment for detecting vehicles occupying bus lane |
| CN113642412B (en) * | 2021-07-16 | 2023-12-26 | 盛视科技股份有限公司 | Method, device and equipment for detecting vehicle occupying bus lane |
| CN113851001B (en) * | 2021-09-17 | 2023-08-29 | 同济大学 | Multi-lane merging illegal automatic auditing method, system, device and storage medium |
| CN113851001A (en) * | 2021-09-17 | 2021-12-28 | 同济大学 | Automatic auditing method, system, device and storage medium for multilane merging violation |
| CN113989794B (en) * | 2021-11-12 | 2022-08-02 | 珠海安联锐视科技股份有限公司 | License plate detection and recognition method |
| CN113989794A (en) * | 2021-11-12 | 2022-01-28 | 珠海安联锐视科技股份有限公司 | License plate detection and recognition method |
| CN114373139A (en) * | 2022-01-07 | 2022-04-19 | 北京远度互联科技有限公司 | Method, device, electronic equipment and storage medium for identifying road violation |
| CN114495109A (en) * | 2022-01-24 | 2022-05-13 | 山东大学 | Grabbing robot based on matching of target and scene characters and grabbing method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948416A (en) | A kind of illegal occupancy bus zone automatic auditing method based on deep learning | |
| CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
| CN104298976B (en) | Detection method of license plate based on convolutional neural networks | |
| CN109949579A (en) | A kind of illegal automatic auditing method that makes a dash across the red light based on deep learning | |
| CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
| CN103971097B (en) | Vehicle license plate recognition method and system based on multiscale stroke models | |
| CN109948418A (en) | A kind of illegal automatic auditing method of violation guiding based on deep learning | |
| CN103324920A (en) | Method for automatically identifying vehicle type based on vehicle frontal image and template matching | |
| CN109766769A (en) | A kind of road target detection recognition method based on monocular vision and deep learning | |
| CN107085696A (en) | A kind of vehicle location and type identifier method based on bayonet socket image | |
| CN106127107A (en) | The model recognizing method that multi-channel video information based on license board information and vehicle's contour merges | |
| CN109637151A (en) | A kind of recognition methods that highway Emergency Vehicle Lane is driven against traffic regulations | |
| CN110689724B (en) | Automatic motor vehicle zebra crossing present pedestrian auditing method based on deep learning | |
| CN106682586A (en) | Method for real-time lane line detection based on vision under complex lighting conditions | |
| CN112712057B (en) | Traffic signal identification method and device, electronic equipment and storage medium | |
| CN115717894B (en) | Vehicle high-precision positioning method based on GPS and common navigation map | |
| CN110226170A (en) | A kind of traffic sign recognition method in rain and snow weather | |
| CN103279756A (en) | Vehicle detecting analysis system and detecting analysis method thereof based on integrated classifier | |
| CN109389046A (en) | Round-the-clock object identification and method for detecting lane lines for automatic Pilot | |
| CN109948417A (en) | A kind of vehicle based on deep learning drives in the wrong direction illegal automatic auditing method | |
| CN105184301B (en) | A kind of method that vehicle heading is differentiated using four-axle aircraft | |
| CN106919939B (en) | A kind of traffic signboard tracks and identifies method and system | |
| CN113553916B (en) | Orbit dangerous area obstacle detection method based on convolutional neural network | |
| CN112381101B (en) | Infrared road scene segmentation method based on category prototype regression | |
| CN108960175A (en) | A kind of licence plate recognition method based on deep learning |
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 | ||
| AD01 | Patent right deemed abandoned | ||
| AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20230509 |