CN110458225A - A kind of vehicle detection and posture are classified joint recognition methods - Google Patents
A kind of vehicle detection and posture are classified joint recognition methods Download PDFInfo
- Publication number
- CN110458225A CN110458225A CN201910729142.4A CN201910729142A CN110458225A CN 110458225 A CN110458225 A CN 110458225A CN 201910729142 A CN201910729142 A CN 201910729142A CN 110458225 A CN110458225 A CN 110458225A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- posture
- detection
- module
- vehicle detection
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V10/40—Extraction of image or video features
Abstract
The present invention relates to vehicle detection identification technology fields, and disclose a kind of vehicle detection and posture classification joint recognition methods, the vehicle detection and posture classification joint identify that required module has vehicle detection module, detection post-processing module, characteristic size unified modules and posture categorization module, flow processing vehicle image is successively pressed by each module, and each module is all made of alternately training method.The vehicle detection and posture are classified joint recognition methods, by focusing under the premise of not influencing Detection task by after alternately joint training while exporting position and the posture information of vehicle;It positions vehicle location in feature output layer using homing method and candidate frame will test by sensitizing range pond after nms is post-processed and be unified for identical size, the full connection of finally access carries out posture classification;Vehicle attitude, which is classified, has been multiplexed the feature of Detection task, and exports posture classification results together with detection, more efficient.
Description
Technical field
The present invention relates to vehicle detection identification technology field, specially a kind of vehicle detection and posture classification joint identification side
Method.
Background technique
In vehicle monitoring, the vehicle in frame each in video is positioned first, then passes through vehicle position in successive frame
Set association determine its running track, but due in video vehicle frequently staggeredly cause association misjudgment cause identical strip path curve to be closed
The vehicle for joining different directions, by vehicle attitude can auxiliary judgment traffic direction determination its running track, and in static map
Because that can not judge direction of traffic according to related information without video inter motion information as in, vehicle can be judged by posture information
Direction.
Patent of invention (application number: 201811309235.3) discloses a kind of vehicle attitude classification side based on deep learning
Method, this method comprises: training data building, validation data set building, building mobile-net network, network training, prediction;It should
Application has fully considered each vehicle, all angles, various scenes, using mobile-net network, can be quickly obtained knowledge
Not as a result, and possessing higher accuracy rate.
Patent of invention (application number: 201811309235.3) discloses a kind of vehicle classified based on SSD and vehicle attitude
Detection method and system, which comprises vehicle attitude is divided according to the angle of headstock and trunnion axis, original
Vehicle attitude is added on SSD network model to be divided, in original SSD Detectability loss and vehicle attitude classification task, by vehicle
Detectability loss and vehicle attitude classification combine to form multitask loss, and the softmax loss of original SSD model is replaced with
Focal loss loss, training obtain detection model, carry out vehicle detection to picture to be detected using detection model, realize more rulers
The vehicle detection of degree, angle;Preferably resolving multiple dimensioned, multi-angle makes vehicle in video frequently interlock, and association is caused to sentence
Dislocation misleads the vehicle for causing identical strip path curve association different directions.
However, a kind of vehicle attitude classification method based on deep learning disclosed in patent application 1, is an independent task
Posture classification is carried out to each vehicle individual, the feature of Detection task training is not multiplexed, is taken a long time after task is cumulative, logic
Complexity is unfavorable for task integration;Patent application 2 discloses vehicle attitude classification task and the multitask arranged side by side of vehicle detection task
It exports simultaneously, for Detection task, itself is divided into two tasks of classification and positioning, since it trains complexity high, right
Data distribution is more sensitive, thus generally require balance adjustment mechanism between more large batch of data and finer class be trained it is quasi-
It closes.Posture classification task and Detection task are integrated into multitask arranged side by side by this method, and single vehicle detection becomes more from one kind
Posture classification, causes the data of single class to fall sharply and occur equilibrium problem between unstable class, and it is steady to be unfavorable for Detection task training
It is fixed.
Summary of the invention
The present invention provides a kind of vehicle detections and posture classification joint recognition methods, and it is stable to have vehicle detection training
The more preferable advantage of efficiency simultaneously, solves the technical issues of mentioning in background technique.
The joint recognition methods the invention provides the following technical scheme: a kind of vehicle detection and posture are classified, the vehicle inspection
Survey and posture classification joint identify needed for module have vehicle detection module, detection post-processing module, characteristic size unified modules with
And posture categorization module, flow processing vehicle image is successively pressed by each module, and each module is all made of alternately training side
Formula.
Preferably, the vehicle detection model treatment method is as follows:
The rectangular area of vehicle forms positive sample in A1, handmarking's image, and marks posture for each vehicle positive sample,
Including forward, backward, to the left, four direction, collectively forms vehicle detection and posture classification based training image set to the right;
A2, training image collection is normalized to fixed size, and does random shearing, mirror image, corresponding modification when mirror image operation
Posture label;
A3, building CNN (convolutional neural networks) network are that basic network is trained with the full convolutional network of googlenet
Extract the textural characteristics of vehicle in image;
A4, the step A3 feature extracted is activated using sigmod function, and detection label is combined to use cross entropy
Loss function carries out punishment training to "current" model;
A5, vehicle coordinate is returned using 1loss, the sum of the poor absolute value for calculating four coordinates returns coordinate
Return training.
Preferably, the processing method of the detection post-processing module is as follows:
B1, output setting reliability sequence is carried out to the image of vehicle detection module processing, cut with setting super ginseng threshold value
It takes, and gives up candidate lower than the vehicle of threshold value;
B2, removal and the friendship of detection label and smaller frame, retain the higher all vehicle candidate frames of registration, to increase
The quantity of positive sample achievees the purpose that data augmentation;
The forward certain amount candidate frame of B3, selection sequence is output to next stage.
Preferably, the characteristic size unified modules processing method is as follows:
C1, the vehicle candidate frame characteristic obtained 2.3 are identical by sensitizing range pond method pond chemical conversion size
Characteristic pattern;
C2, the scales candidate frame characteristic patterns such as all are spliced into NxCxHxW sequence, using quantity as the first dimension, channel is made
For the second dimension, the height and width of candidate frame characteristic image are as the third dimension and fourth dimension;
C3, using NxCxHxW candidate frame characteristic pattern as input access next step in fully-connected network.
Preferably, the processing method of the posture categorization module is as follows:
D1, the further extraction data image that posture feature is carried out using fully-connected network;
D2, activated using softmax function, in conjunction with posture tag along sort using cross entropy loss function to model into
Row punishment training.
Preferably, posture classification is closed when the trained Detection task, opens posture after training restrains to a certain extent
Identification carries out joint multitask training.
The present invention have it is following the utility model has the advantages that
The vehicle detection and posture classification joint recognition methods, lead to by focusing under the premise of not influencing Detection task
Cross after alternately joint training while exporting position and the posture information of vehicle;Vehicle is positioned in feature output layer using homing method
Position simultaneously will test candidate frame and is unified for identical size after nms is post-processed by sensitizing range pond, and finally access connects entirely
Carry out posture classification;Vehicle attitude, which is classified, has been multiplexed the feature of Detection task, and exports posture classification results, effect together with detection
Rate is higher.
Detailed description of the invention
Fig. 1 is the flow chart of vehicle detection of the present invention and posture classification joint recognition methods.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The joint recognition methods referring to Fig. 1, a kind of vehicle detection and posture are classified, vehicle detection and posture classification joint are known
Not required module has vehicle detection module, detection post-processing module, characteristic size unified modules and posture categorization module, by each
Module successively presses flow processing vehicle image, and modules are all made of alternately training method, close appearance when training Detection task
State classification opens gesture recognition after training restrains to a certain extent and carries out joint multitask training.
Vehicle detection model treatment method is as follows:
The rectangular area of vehicle forms positive sample in A1, handmarking's image, and marks posture for each vehicle positive sample,
Including forward, backward, to the left, four direction, collectively forms vehicle detection and posture classification based training image set, setting classification to the right
Label and target box label;
A2, training image collection is normalized to fixed size, and does random shearing, mirror image, corresponding modification when mirror image operation
Posture label;
A3, building CNN (convolutional neural networks) network are that basic network is trained with the full convolutional network of googlenet
The textural characteristics of vehicle in image are extracted, i.e., vehicle location classification processing are used nms (non-maximum value inhibition) in training
Rejecting of crossing the border is carried out to prediction block, weed out and detects the lesser frame of label degree of overlapping, i.e., only retaining coverage rate is more than default threshold
The target frame (scalping) of value equally carries out rejecting of crossing the border to prediction block using nms (non-maximum value inhibition) in test, rejects
Fall the biggish frame of degree of overlapping, i.e., only retains the target frame that coverage rate is no more than the local maxima score of preset threshold.Carry out under
The modes such as roialign also can be used to guarantee the higher candidate frame of mapping accuracy in the processing of one step, to obtain fixed size candidate
The mode of frame characteristic layer replaces sensitizing range pond;
A4, the step A3 feature extracted is activated using sigmod function, and detection label is combined to use cross entropy
Loss function carries out punishment training to "current" model;
A5, vehicle coordinate is returned using L1loss, calculates the sum of poor absolute value of four coordinates and coordinate is carried out
Regression training.
The processing method for detecting post-processing module is as follows:
B1, output setting reliability sequence is carried out to the image of vehicle detection module processing, cut with setting super ginseng threshold value
It takes, and gives up candidate lower than the vehicle of threshold value;
B2, removal and the friendship of detection label and smaller frame, retain the higher all vehicle candidate frames of registration, to increase
The quantity of positive sample achievees the purpose that data augmentation;
The forward certain amount candidate frame of B3, selection sequence is output to next stage.
Characteristic size unified modules processing method is as follows:
C1, the vehicle candidate frame characteristic obtained 2.3 are identical by sensitizing range pond method pond chemical conversion size
Characteristic pattern;
C2, the scales candidate frame characteristic patterns such as all are spliced into NxCxHxW sequence, using quantity as the first dimension, channel is made
For the second dimension, for the height and width of candidate frame characteristic image as the third dimension and fourth dimension, x therein is corresponding four-dimensional coordinate value, is had
Conducive to the accuracy for further determining that candidate frame;
C3, using NxCxHxW candidate frame characteristic pattern as input access next step in fully-connected network.
The processing method of posture categorization module is as follows:
D1, the further extraction data image that posture feature is carried out using fully-connected network;
D2, activated using softmax function, in conjunction with posture tag along sort using cross entropy loss function to model into
Row punishment training.
By focusing under the premise of not influencing Detection task by alternately after joint training while exporting the position of vehicle
It sets and posture information, and vehicle location is positioned in feature output layer using homing method and passes through sensitizing range after nms is post-processed
Domain pond will test candidate frame and be unified for identical size, and the full connection of finally access carries out posture classification, vehicle attitude classification multiplexing
The feature of Detection task, and posture classification results are exported together with detection, it is more efficient.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (6)
- The joint recognition methods 1. a kind of vehicle detection and posture are classified, it is characterised in that: the vehicle detection and posture classification connection It closes and identifies that required module has vehicle detection module, detection post-processing module, characteristic size unified modules and posture categorization module, Flow processing vehicle image is successively pressed by each module, and each module is all made of alternately training method.
- The joint recognition methods 2. a kind of vehicle detection according to claim 1 and posture are classified, it is characterised in that: the vehicle Detection model processing method is as follows:The rectangular area of vehicle forms positive sample in A1, handmarking's image, and marks posture for each vehicle positive sample, including Forward, backward, to the left, four direction to the right, collectively forms vehicle detection and posture classification based training image set;A2, training image collection is normalized to fixed size, and does random shearing, mirror image, corresponding modification posture when mirror image operation Label;A3, building CNN (convolutional neural networks) network are that basic network is trained extraction with the full convolutional network of googlenet The textural characteristics of vehicle in image;A4, the step A3 feature extracted is activated using sigmod function, and combines detection label using intersection entropy loss Function carries out punishment training to "current" model;A5, vehicle coordinate is returned using 1loss, calculates the sum of poor absolute value of four coordinates and recurrence instruction is carried out to coordinate Practice.
- The joint recognition methods 3. a kind of vehicle detection according to claim 1 and posture are classified, it is characterised in that: the inspection The processing method of after logging process module is as follows:B1, output setting reliability sequence is carried out to the image of vehicle detection module processing, intercepted with setting super ginseng threshold value, and Give up candidate lower than the vehicle of threshold value;B2, removal and the friendship of detection label and smaller frame, retain the higher all vehicle candidate frames of registration, to increase positive sample This quantity, achievees the purpose that data augmentation;The forward certain amount candidate frame of B3, selection sequence is output to next stage.
- The joint recognition methods 4. a kind of vehicle detection according to claim 1 and posture are classified, it is characterised in that: the spy It is as follows to levy size unified modules processing method:C1, the vehicle candidate frame characteristic obtained 2.3 are melted into the identical feature of size by sensitizing range pond method pond Figure;C2, the scales candidate frame characteristic patterns such as all are spliced into NxCxHxW sequence, using quantity as the first dimension, channel is as the Two dimension, the height and width of candidate frame characteristic image are as the third dimension and fourth dimension;C3, using NxCxHxW candidate frame characteristic pattern as input access next step in fully-connected network.
- The joint recognition methods 5. a kind of vehicle detection according to claim 1 and posture are classified, it is characterised in that: the appearance The processing method of state categorization module is as follows:D1, the further extraction data image that posture feature is carried out using fully-connected network;D2, it is activated using softmax function, model is punished using cross entropy loss function in conjunction with posture tag along sort Penalize training.
- The joint recognition methods 6. a kind of vehicle detection according to claim 1 and posture are classified, it is characterised in that: the instruction Posture classification is closed when practicing Detection task, is opened gesture recognition after training restrains to a certain extent and is carried out joint multitask instruction Practice.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910729142.4A CN110458225A (en) | 2019-08-08 | 2019-08-08 | A kind of vehicle detection and posture are classified joint recognition methods |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910729142.4A CN110458225A (en) | 2019-08-08 | 2019-08-08 | A kind of vehicle detection and posture are classified joint recognition methods |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110458225A true CN110458225A (en) | 2019-11-15 |
Family
ID=68485322
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910729142.4A Pending CN110458225A (en) | 2019-08-08 | 2019-08-08 | A kind of vehicle detection and posture are classified joint recognition methods |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110458225A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111008608A (en) * | 2019-12-11 | 2020-04-14 | 湖南大学 | Night vehicle detection method based on deep learning |
CN113408375A (en) * | 2021-06-03 | 2021-09-17 | 南京佑驾科技有限公司 | Road vehicle driving posture classification method, device and system |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324938A (en) * | 2012-03-21 | 2013-09-25 | 日电(中国)有限公司 | Method for training attitude classifier and object classifier and method and device for detecting objects |
CN106250838A (en) * | 2016-07-27 | 2016-12-21 | 乐视控股(北京)有限公司 | vehicle identification method and system |
US20170168586A1 (en) * | 2015-12-15 | 2017-06-15 | Purdue Research Foundation | Method and System for Hand Pose Detection |
CN107871126A (en) * | 2017-11-22 | 2018-04-03 | 西安翔迅科技有限责任公司 | Model recognizing method and system based on deep-neural-network |
CN107895150A (en) * | 2016-11-30 | 2018-04-10 | 奥瞳系统科技有限公司 | Face datection and head pose angle based on the small-scale convolutional neural networks module of embedded system are assessed |
CN108038474A (en) * | 2017-12-28 | 2018-05-15 | 深圳云天励飞技术有限公司 | Method for detecting human face, the training method of convolutional neural networks parameter, device and medium |
CN108062521A (en) * | 2017-12-12 | 2018-05-22 | 深圳大学 | Method for detecting human face, device, terminal and medium based on convolutional neural networks |
CN108596053A (en) * | 2018-04-09 | 2018-09-28 | 华中科技大学 | A kind of vehicle checking method and system based on SSD and vehicle attitude classification |
CN109299644A (en) * | 2018-07-18 | 2019-02-01 | 广东工业大学 | A kind of vehicle target detection method based on the full convolutional network in region |
CN109447171A (en) * | 2018-11-05 | 2019-03-08 | 电子科技大学 | A kind of vehicle attitude classification method based on deep learning |
CN109624850A (en) * | 2017-10-05 | 2019-04-16 | 斯特拉德视觉公司 | Monitor the method for the blind spot of vehicle and the blind spot monitoring device using this method |
CN110096962A (en) * | 2019-04-04 | 2019-08-06 | 苏州千视通视觉科技股份有限公司 | Vehicle Detail based on region convolutional network identifies secondary structure method and device |
-
2019
- 2019-08-08 CN CN201910729142.4A patent/CN110458225A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324938A (en) * | 2012-03-21 | 2013-09-25 | 日电(中国)有限公司 | Method for training attitude classifier and object classifier and method and device for detecting objects |
US20170168586A1 (en) * | 2015-12-15 | 2017-06-15 | Purdue Research Foundation | Method and System for Hand Pose Detection |
CN106250838A (en) * | 2016-07-27 | 2016-12-21 | 乐视控股(北京)有限公司 | vehicle identification method and system |
CN107895150A (en) * | 2016-11-30 | 2018-04-10 | 奥瞳系统科技有限公司 | Face datection and head pose angle based on the small-scale convolutional neural networks module of embedded system are assessed |
CN109624850A (en) * | 2017-10-05 | 2019-04-16 | 斯特拉德视觉公司 | Monitor the method for the blind spot of vehicle and the blind spot monitoring device using this method |
CN107871126A (en) * | 2017-11-22 | 2018-04-03 | 西安翔迅科技有限责任公司 | Model recognizing method and system based on deep-neural-network |
CN108062521A (en) * | 2017-12-12 | 2018-05-22 | 深圳大学 | Method for detecting human face, device, terminal and medium based on convolutional neural networks |
CN108038474A (en) * | 2017-12-28 | 2018-05-15 | 深圳云天励飞技术有限公司 | Method for detecting human face, the training method of convolutional neural networks parameter, device and medium |
CN108596053A (en) * | 2018-04-09 | 2018-09-28 | 华中科技大学 | A kind of vehicle checking method and system based on SSD and vehicle attitude classification |
CN109299644A (en) * | 2018-07-18 | 2019-02-01 | 广东工业大学 | A kind of vehicle target detection method based on the full convolutional network in region |
CN109447171A (en) * | 2018-11-05 | 2019-03-08 | 电子科技大学 | A kind of vehicle attitude classification method based on deep learning |
CN110096962A (en) * | 2019-04-04 | 2019-08-06 | 苏州千视通视觉科技股份有限公司 | Vehicle Detail based on region convolutional network identifies secondary structure method and device |
Non-Patent Citations (5)
Title |
---|
DI WU 等: "6D-VNet: End-to-end 6DoF Vehicle Pose Estimation from Monocular RGB Images", 《CVPRW 2019》 * |
MARKUS BRAUN 等: "Pose-RCNN: Joint Object Detection and Pose Estimation Using 3D Object Proposals", 《2016 IEEE 19TH INTERNATIONAL CONFERENCE ON INTELLIGENT TRANSPORTATION SYSTEMS (ITSC)》 * |
MICHAEL HO¨DLMOSER 等: "Classification and Pose Estimation of Vehicles in Videos by 3D Modeling within Discrete-Continuous Optimization", 《2012 SECOND INTERNATIONAL CONFERENCE ON 3D IMAGING,MODELING,PROCESSING,VISUALIZATION & TRANSMISSION》 * |
PATRICK POIRSON 等: "Fast Single Shot Detection and Pose Estimation", 《ARXIV》 * |
阮航 等: "基于Faster R-CNN的车辆多属性识别", 《计算机技术与发展》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111008608A (en) * | 2019-12-11 | 2020-04-14 | 湖南大学 | Night vehicle detection method based on deep learning |
CN111008608B (en) * | 2019-12-11 | 2023-08-01 | 湖南大学 | Night vehicle detection method based on deep learning |
CN113408375A (en) * | 2021-06-03 | 2021-09-17 | 南京佑驾科技有限公司 | Road vehicle driving posture classification method, device and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107134144B (en) | A kind of vehicle checking method for traffic monitoring | |
CN108830188B (en) | Vehicle detection method based on deep learning | |
CN106127204B (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
CN109711320B (en) | Method and system for detecting violation behaviors of staff on duty | |
CN103324937A (en) | Method and device for labeling targets | |
CN106097346B (en) | A kind of video fire hazard detection method of self study | |
CN106897738B (en) | A kind of pedestrian detection method based on semi-supervised learning | |
CN109102025A (en) | Pedestrian based on deep learning combined optimization recognition methods again | |
CN101447082B (en) | Detection method of moving target on a real-time basis | |
Mallikarjuna et al. | Traffic data collection under mixed traffic conditions using video image processing | |
CN110335270A (en) | Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features | |
CN109859163A (en) | A kind of LCD defect inspection method based on feature pyramid convolutional neural networks | |
CN106373143A (en) | Adaptive method and system | |
CN101221623B (en) | Object type on-line training and recognizing method and system thereof | |
CN110348437A (en) | It is a kind of based on Weakly supervised study with block the object detection method of perception | |
CN109508664A (en) | A kind of vegetable identification pricing method based on deep learning | |
CN111368690A (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
CN101872414B (en) | People flow rate statistical method and system capable of removing false targets | |
CN105069477A (en) | Method for AdaBoost cascade classifier to detect image object | |
CN107145862A (en) | A kind of multiple features matching multi-object tracking method based on Hough forest | |
CN105608431A (en) | Vehicle number and traffic flow speed based highway congestion detection method | |
CN108764373A (en) | A kind of sensing data filtering and fusion method in automatic Pilot | |
CN103246896A (en) | Robust real-time vehicle detection and tracking method | |
CN102915433A (en) | Character combination-based license plate positioning and identifying method | |
CN110458225A (en) | A kind of vehicle detection and posture are classified joint recognition methods |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191115 |
|
WD01 | Invention patent application deemed withdrawn after publication |