CN107316007A - A kind of monitoring image multiclass object detection and recognition methods based on deep learning - Google Patents

A kind of monitoring image multiclass object detection and recognition methods based on deep learning Download PDF

Info

Publication number
CN107316007A
CN107316007A CN201710424764.7A CN201710424764A CN107316007A CN 107316007 A CN107316007 A CN 107316007A CN 201710424764 A CN201710424764 A CN 201710424764A CN 107316007 A CN107316007 A CN 107316007A
Authority
CN
China
Prior art keywords
classification
frame
image
training
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.)
Granted
Application number
CN201710424764.7A
Other languages
Chinese (zh)
Other versions
CN107316007B (en
Inventor
尚凌辉
王弘玥
张兆生
刘小扬
郑永宏
丁连涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Jieshang Safety Equipment Co ltd
Original Assignee
ZHEJIANG ICARE VISION TECHNOLOGY Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by ZHEJIANG ICARE VISION TECHNOLOGY Co Ltd filed Critical ZHEJIANG ICARE VISION TECHNOLOGY Co Ltd
Priority to CN201710424764.7A priority Critical patent/CN107316007B/en
Publication of CN107316007A publication Critical patent/CN107316007A/en
Application granted granted Critical
Publication of CN107316007B publication Critical patent/CN107316007B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/20—Scenes; Scene-specific elements in augmented reality scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Multimedia (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of monitoring image multiclass object detection based on deep learning and recognition methods.The present invention redesigns network structure and corresponding various parameters on this framework, can rapidly detect object concerned in monitor video image by using published SSD deep learnings detection framework.For relatively conventional image processing method, the present invention can learn more effective more rich feature automatically using deep learning, so that with higher robustness.For other deep learning methods, the present invention redesigns object function using the mixed-media network modules mixed-media using some light weights, and introduces residual error module and probability thermodynamic chart, so as to maintain speed and the advantage in performance.In balance, method of the invention can efficiently be quickly detected the attention object in image, and can be generalized to more general object detecting areas.

Description

A kind of monitoring image multiclass object detection and recognition methods based on deep learning
Technical field
The invention belongs to technical field of video monitoring, it is related to a kind of monitoring image multiclass object detection based on deep learning With recognition methods.
Background technology
As the increase year by year and country of Chinese automobile quantity are to road, the lasting input of the monitoring device of cell, such as The problem of what with the so big quantity monitor video of parsing or image effectively using current urgent need to resolve is turned into.
Analysis to these picture materials either all has with understanding in traffic, security protection, or in terms of video investigation Important application.Image object detection is as first step understood with analysis diagram picture, and its performance directly affects follow-up step Rapid effect.In monitoring image, how people, motor vehicle, this several class of non-motor vehicle are preferably examined as the primary body of concern Survey this type objects be present invention mainly solves the problem of.
Current existing technology mainly has based on traditional image processing method or machine learning method from route, Such as《To the detection -200980137706.X of the vehicle in image》,《A kind of vehicle checking method based on image- 201310259434.9》,《A kind of vehicle detection at night method -201410104987.1 based on image》Deng such method is deposited Shortcoming include to environment and scene heavy dependence, to Time Dependent and can only realize that single class is detected, detect degraded performance Deng this seriously constrains the real practicality of these technologies.
Second of route is to be based on deep learning method, such as《A kind of vehicle based on quick R-CNN deep neural networks Recognition methods -201610563184.1》,《A kind of traffic image polymorphic type vehicle checking method based on deep learning- 201610397819.5》,《A kind of multi-direction vehicle detection identifying system -201610316159.3 based on deep learning》,《Base In the road vehicle real-time detection method -201511183427.5 of deep learning SSD frameworks》.This kind of method no matter from performance or The time-consuming aspect of person is better than foregoing conventional method.
The content of the invention
In order to deal with the video image content of current sharp increase, the effect of video image analysis is improved, the present invention passes through Using published SSD deep learnings detection framework, network structure and corresponding various parameters are redesigned on this framework, are made It can rapidly detect object (people, motor vehicle, non-motor vehicle) concerned in monitor video image, be follow-up image Understanding is laid a solid foundation.
The present invention solves the technological means taken of technical problem：
Step 1, the road for collecting different places, cell monitoring vedio data mark object interested.
Step 2, xml annotation formattings for markup information in step 1 being changed into SSD supports and required for preparing training Labelmap.prototxt and val_name_size.txt, is divided into training set and collects with checking according to a certain percentage.
Step 3, project training network：
3a) combinations VGG, SqueezeNet, ResNet networks redesign network.What it is due to SSD original frames is VGG networks, the network parameter is relatively more, and amount of calculation is huge, causes speed slower.In order to overcome this shortcoming, the present invention is based on The consideration of speed and aspect of performance, has redesigned network structure：
First substituted for former VGG network Cs onv1~Conv5 with SqueezeNet network, and second is to use 1x3,3x1's Network instead of primitive network 3x3 part.Improved by the two, network parameter can be greatly reduced, reduce network calculations Amount.3rd is that the non-convolutional layer for extracting frame introduces ResNet modules in Conv6~Conv8, and this can not increase very On the premise of many amounts of calculation, retain more upper layer informations, so as to lift network performance.Improved by these, the present invention The method used, substantially with maintaining an equal level in disclosed document using VGG, but improves 7-10 times in performance in speed.
3b) counts the information of all marks, sets the aspect_ratio of the different characteristic layers for extracting frame.
Step 4, statistics and practical application request according to step 3b, leveldb/lmdb number is generated by WxH size It is used to train according to library format.
Step 5, design loss functions：
Wherein, c, l, g, x represent to belong to the probability of some classification, prediction block, true frame, and prediction block and true respectively (matching is 1 to the mark of frame matching, otherwise for 0)；What N was represented is the quantity matched with callout box；Lloc(x, l, g) damages for positioning Lose function；Lconf(x, c) presentation class loss function.
In order to eliminate the influence of imbalanced training sets generation, the present invention redefines Classification Loss function Lconf(x, c) is as follows：
Here wiDifferent classes of weight is represented, its calculation formula is：
Expression classification is ciQuantity, M be total frame quantity.
Step 6, the function proposed using Caffe-SSD detection frameworks using previous step are classified as optimization aim training detection Device.
Step 7, the frame that obtained detection sorter model, input test picture or video are trained using step 6, then do The non-maxima suppression (Non Maximum Suppression, NMS) of generic, obtains the output of each image/frame, this Output includes the physical quantities detected, and each position, classification and confidence level of object.
The probability that each classification occurs in the picture in step 8, statistics training set, obtains each classification on zoomed image Probability thermodynamic chart.
Step 9, the probability thermodynamic chart obtained using step 8 recalculate the value of the confidence level of each classification, then sharp again Merge different classes of frame with NMS.
Step 10, according to actual conditions to the different confidence threshold values of different classes of design, being less than this class confidence level threshold The result of value is filtered out, so as to obtain final output result.
Beneficial effects of the present invention：For relatively conventional image processing method, the present invention can be certainly using deep learning The dynamic more effective more rich feature of study, so that with higher robustness.For other deep learning methods, the present invention Using the mixed-media network modules mixed-media using some light weights, object function is redesigned, and introduces residual error module and probability thermodynamic chart, so that Maintain speed and the advantage in performance.In balance, method of the invention can efficiently be quickly detected interested in image Object, and more general object detecting areas can be generalized to.
Brief description of the drawings
Fig. 1 is the inventive method flow chart.
Fig. 2 is the network structure after improving.
Embodiment
1 the present invention will be described in detail below in conjunction with the accompanying drawings, and of the invention comprises the following steps that：
Step 1, the road for collecting different places, cell monitoring vedio data, and the people in image, two wheeler, Tricycle, headstock, the tailstock, which is marked out, to be come.
Step 2, xml annotation formattings for markup information in step 1 being changed into SSD supports and required for preparing training Labelmap.prototxt and val_name_size.txt, and by 4:1 ratio point training set collects with checking.
Step 3, project training network:
3a) combinations VGG, SqueezeNet, ResNet networks redesign network, as shown in Figure 2；
3b) counts the information of all marks, the aspect_ratio of the different characteristic layers for extracting frame is set, such as following table institute Show.
Characteristic layer Value
SqueezNet_4 (0.25,0.5,0.75,1,1.25)
SqueezeNet_9 (0.5,0.8,1,1.3,1.6)
ConvBlock_2 (0.4,0.8,1,1.4,1.7)
ConvBlock_3 (0.4,0.8,1,1.4,1.7)
ConvBlock_4 (0.5,0.8,1,1.2,1.8)
GlobalPooling_1 (0.8,1,1.25,1.6,2)
Step 4, statistics and practical application request according to step 3b, are generated by WxH (384x2556) size Leveldb/lmdb database format is used to train.
Step 5, design loss functions：
Wherein Lloc(x, l, g) is the loss of positioning, is specifically defined referring to SSD papers.
LconfThe definition of (x, c) redesigns into such as minor function：
What N was represented here is the quantity matched with callout box.wiDifferent classes of weight is represented, its calculation formula is：
Expression classification is ciQuantity, M be total frame quantity.
In this example, categorical measure is 5 classes, according to formula above wiValue in order for (0.17,0.22,0.34, 0.12,0.15)。
Step 6, utilize Caffe-SSD detection frameworks training detection grader, training solver parameters following table institute used Show.
Parameter Value
base_lr 0.001
max_iter 400000
lr_polcy Step
gamma 0.8
momentum 0.95
weight_decay 0.0005
stepsize 40000
average_loss 10
type SGD
6a) first it is trained 200,000 times with SqueezeNet1.0 pre-training model, obtains a model；
6b) use 6a) the obtained model re -training of training 400,000 times.
Step 7, the frame that obtained detection sorter model, input test picture or video are trained using step 6, then do The non-maxima suppression (Non Maximum Suppression, NMS) of generic, obtains the output of each image/frame, this Output includes the physical quantities detected, and each position of object, classification, confidence level.
The probability that each classification occurs in the picture in step 8, statistics training set, obtains each classification on zoomed image Probability thermodynamic chart.
Step 9, the probability thermodynamic chart obtained using step 8 recalculate the value of the confidence level of each classification, then sharp again Merge different classes of frame with NMS.
Step 10, according to actual conditions to the different confidence threshold values of different classes of design, the confidence level point of five classifications (0.6,0.6,0.4,0.8,0.85) is not set to, the result less than respective class confidence level can be dropped, so as to export this figure Coordinate and confidence level as corresponding to contained object classification and each object.
The present embodiment partial content is with reference to as follows：
SSD:Single Shot MultiBox Detector,ECCV,2016.
SqueezeNet:AlexNet-level accuracy with 50x fewer parameters and<0.5MB model size,Arxiv,2016.
Deep Residual Learning for Image Recognition,CVPR,2016
The present invention can detect in monitoring image all objects interested simultaneously and have and carried in performance and speed It is high.Amount of calculation is greatly decreased simultaneously, people that can simultaneously in detection image, non-motor vehicle and motor vehicle, can be reached in speed 85 frames/second is reached on Nvidia M40, far surpasses last 25 frames of patent/second.

Claims (4)

1. a kind of monitoring image multiclass object detection and recognition methods based on deep learning, it is characterised in that this method include with Lower step：
Step 1, road, the cell monitoring vedio data for collecting different places, mark object interested；
Step 2, xml annotation formattings for markup information in step 1 being changed into SSD supports and required for preparing training Labelmap.prototxt and val_name_size.txt, is divided into training set and collects with checking according to a certain percentage；
Step 3, project training network：
3a) combinations VGG, SqueezeNet, ResNet networks redesign network：
Original VGG network Cs onv1~Conv5 is replaced with SqueezeNet network；
Replace original VGG networks 3x3 part with 1x3,3x1 network；
The non-convolutional layer for extracting frame introduces ResNet modules in original VGG network Cs onv6~Conv8；
3b) counts the information of all marks, sets the aspect_ratio of the different characteristic layers for extracting frame；
Step 4, statistics and practical application request according to step 3b, leveldb/lmdb database is generated by WxH size Form is used to train；
Step 5, design loss functions：
Wherein：Represent to belong to respectively probability, prediction block, true frame and the prediction block of some classification with it is true The mark of frame matching；N represents the quantity matched with callout box；For positioning loss function；Presentation class loses letter Number；
Defining classification loss function is as follows：
Different classes of weight is represented, its calculation formula is
Represent that classification isQuantity, M be total frame quantity；
Step 6, the function proposed using Caffe-SSD detection frameworks using previous step are trained as optimization aim and detect grader；
Step 7, the frame that obtained detection sorter model, input test picture or video are trained using step 6, then do similar The non-maxima suppression of other, obtains the output of each image/frame；
The probability that each classification occurs in the picture in step 8, statistics training set, obtains each classification general on zoomed image Rate thermodynamic chart；
Step 9, the probability thermodynamic chart obtained using step 8 recalculate the value of the confidence level of each classification, then recycle non- Maximum suppresses to merge different classes of frame；
Step 10, according to actual conditions to the different confidence threshold values of different classes of design, less than this class confidence threshold value As a result filter out, so as to obtain final output result.
2. monitoring image multiclass object detection according to claim 1 and recognition methods, it is characterised in that：In step 2 The ratio of training set and checking collection is 4:1.
3. monitoring image multiclass object detection according to claim 1 and recognition methods, it is characterised in that：Step 6 is specific It is：First it is trained with SqueezeNet1.0 pre-training model, frequency of training is 200,000 times, obtains a model；Then use Obtained model re -training 400,000 times.
4. monitoring image multiclass object detection according to claim 1 and recognition methods, it is characterised in that：In step 7 Output includes the physical quantities detected, and each position, classification and confidence level of object.
CN201710424764.7A 2017-06-07 2017-06-07 Monitoring image multi-class object detection and identification method based on deep learning Active CN107316007B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710424764.7A CN107316007B (en) 2017-06-07 2017-06-07 Monitoring image multi-class object detection and identification method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710424764.7A CN107316007B (en) 2017-06-07 2017-06-07 Monitoring image multi-class object detection and identification method based on deep learning

Publications (2)

Publication Number Publication Date
CN107316007A true CN107316007A (en) 2017-11-03
CN107316007B CN107316007B (en) 2020-04-03

Family

ID=60181710

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710424764.7A Active CN107316007B (en) 2017-06-07 2017-06-07 Monitoring image multi-class object detection and identification method based on deep learning

Country Status (1)

Country Link
CN (1) CN107316007B (en)

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107911429A (en) * 2017-11-04 2018-04-13 南京奇蛙智能科技有限公司 A kind of online traffic flow monitoring method in unmanned plane high in the clouds based on video
CN107944457A (en) * 2017-11-23 2018-04-20 浙江清华长三角研究院 Drawing object identification and extracting method under a kind of complex scene
CN108010030A (en) * 2018-01-24 2018-05-08 福州大学 A kind of Aerial Images insulator real-time detection method based on deep learning
CN108256428A (en) * 2017-12-19 2018-07-06 歌尔股份有限公司 Interpretation method and mobile terminal in kind, server
CN108764148A (en) * 2018-05-30 2018-11-06 东北大学 Multizone real-time action detection method based on monitor video
CN108830170A (en) * 2018-05-24 2018-11-16 杭州电子科技大学 A kind of end-to-end method for tracking target indicated based on layered characteristic
CN108960076A (en) * 2018-06-08 2018-12-07 东南大学 Ear recognition and tracking based on convolutional neural networks
CN109003262A (en) * 2018-06-29 2018-12-14 炬大科技有限公司 Stain clean method and device
CN109034190A (en) * 2018-06-15 2018-12-18 广州深域信息科技有限公司 A kind of object detecting system and method for the active sample excavation of dynamic select strategy
CN109063544A (en) * 2018-06-12 2018-12-21 深圳中兴网信科技有限公司 A kind of method, system, computer equipment and the storage medium of dump truck detection
CN109142374A (en) * 2018-08-15 2019-01-04 广州市心鉴智控科技有限公司 Method and system based on the efficient Checking model of extra small sample training
CN109255815A (en) * 2018-08-30 2019-01-22 天津理工大学 A kind of object detection and recognition methods based on order spherical harmonic
CN109359573A (en) * 2018-09-30 2019-02-19 天津天地伟业投资管理有限公司 A kind of warning method and device based on the separation of accurate people's vehicle
CN109389046A (en) * 2018-09-11 2019-02-26 昆山星际舟智能科技有限公司 Round-the-clock object identification and method for detecting lane lines for automatic Pilot
CN109614914A (en) * 2018-12-05 2019-04-12 北京纵目安驰智能科技有限公司 Parking stall vertex localization method, device and storage medium
CN109670450A (en) * 2018-12-20 2019-04-23 天津天地伟业信息系统集成有限公司 A kind of people's vehicle object detecting method based on video
CN109671236A (en) * 2018-12-28 2019-04-23 广东美电贝尔科技集团股份有限公司 The detection method and its system of circumference target object
CN109737974A (en) * 2018-12-14 2019-05-10 中国科学院深圳先进技术研究院 A kind of 3D navigational semantic map updating method, device and equipment
CN110008910A (en) * 2019-04-10 2019-07-12 江西理工大学 A kind of movie and television play is exposed the false the detection method of camera lens
CN110136198A (en) * 2018-02-09 2019-08-16 腾讯科技（深圳）有限公司 Image processing method and its device, equipment and storage medium
CN110163225A (en) * 2018-02-11 2019-08-23 顺丰科技有限公司 It is a kind of that package detection and recognition methods, apparatus and system are mixed based on cloud platform
CN110349148A (en) * 2019-07-11 2019-10-18 电子科技大学 A kind of image object detection method based on Weakly supervised study
WO2020007287A1 (en) * 2018-07-05 2020-01-09 第四范式（北京）技术有限公司 Machine learning process implementation method and apparatus, device, and storage medium
CN110827261A (en) * 2019-11-05 2020-02-21 泰康保险集团股份有限公司 Image quality detection method and device, storage medium and electronic equipment
CN111080608A (en) * 2019-12-12 2020-04-28 哈尔滨市科佳通用机电股份有限公司 Method for recognizing closing fault image of automatic brake valve plug handle of railway wagon in derailment
CN111369501A (en) * 2020-02-24 2020-07-03 武汉大学 Deep learning method for identifying oral squamous cell carcinoma based on visual features
CN112151167A (en) * 2020-05-14 2020-12-29 余红兵 Intelligent screening method for six-age dental caries of children based on deep learning
CN112183460A (en) * 2020-10-20 2021-01-05 武汉光谷联合集团有限公司 Method and device for intelligently identifying environmental sanitation
CN112418278A (en) * 2020-11-05 2021-02-26 中保车服科技服务股份有限公司 Multi-class object detection method, terminal device and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105975941A (en) * 2016-05-31 2016-09-28 电子科技大学 Multidirectional vehicle model detection recognition system based on deep learning
CN106611162A (en) * 2016-12-20 2017-05-03 西安电子科技大学 Method for real-time detection of road vehicle based on deep learning SSD frame

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105975941A (en) * 2016-05-31 2016-09-28 电子科技大学 Multidirectional vehicle model detection recognition system based on deep learning
CN106611162A (en) * 2016-12-20 2017-05-03 西安电子科技大学 Method for real-time detection of road vehicle based on deep learning SSD frame

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
CHENG-YANG FU等: "DSSD : Deconvolutional Single Shot Detector", 《ARXIV:1701.06659V1》 *
FORREST N. IANDOLA等: "SQUEEZENET: ALEXNET-LEVEL ACCURACY WITH 50X FEWER PARAMETERS AND <0.5MB MODEL SIZE", 《ARXIV:1602.07360V4》 *
KAIMING HE等: "Deep Residual Learning for Image Recognition", 《ARXIV:1512.03385V1》 *
WEI LIU等: "SSD: Single Shot MultiBox Detector", 《ARXIV:1512.02325V5》 *

Cited By (36)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107911429A (en) * 2017-11-04 2018-04-13 南京奇蛙智能科技有限公司 A kind of online traffic flow monitoring method in unmanned plane high in the clouds based on video
CN107944457A (en) * 2017-11-23 2018-04-20 浙江清华长三角研究院 Drawing object identification and extracting method under a kind of complex scene
CN108256428A (en) * 2017-12-19 2018-07-06 歌尔股份有限公司 Interpretation method and mobile terminal in kind, server
CN108010030A (en) * 2018-01-24 2018-05-08 福州大学 A kind of Aerial Images insulator real-time detection method based on deep learning
CN110136198A (en) * 2018-02-09 2019-08-16 腾讯科技（深圳）有限公司 Image processing method and its device, equipment and storage medium
CN110163225A (en) * 2018-02-11 2019-08-23 顺丰科技有限公司 It is a kind of that package detection and recognition methods, apparatus and system are mixed based on cloud platform
CN108830170A (en) * 2018-05-24 2018-11-16 杭州电子科技大学 A kind of end-to-end method for tracking target indicated based on layered characteristic
CN108830170B (en) * 2018-05-24 2022-03-18 杭州电子科技大学 End-to-end target tracking method based on layered feature representation
CN108764148A (en) * 2018-05-30 2018-11-06 东北大学 Multizone real-time action detection method based on monitor video
CN108764148B (en) * 2018-05-30 2020-03-10 东北大学 Multi-region real-time action detection method based on monitoring video
CN108960076A (en) * 2018-06-08 2018-12-07 东南大学 Ear recognition and tracking based on convolutional neural networks
CN108960076B (en) * 2018-06-08 2022-07-12 东南大学 Ear recognition and tracking method based on convolutional neural network
CN109063544A (en) * 2018-06-12 2018-12-21 深圳中兴网信科技有限公司 A kind of method, system, computer equipment and the storage medium of dump truck detection
CN109034190B (en) * 2018-06-15 2022-04-12 拓元(广州)智慧科技有限公司 Object detection system and method for active sample mining by dynamically selecting strategy
CN109034190A (en) * 2018-06-15 2018-12-18 广州深域信息科技有限公司 A kind of object detecting system and method for the active sample excavation of dynamic select strategy
CN109003262B (en) * 2018-06-29 2022-06-21 炬大科技有限公司 Stubborn stain cleaning method and device
CN109003262A (en) * 2018-06-29 2018-12-14 炬大科技有限公司 Stain clean method and device
WO2020007287A1 (en) * 2018-07-05 2020-01-09 第四范式（北京）技术有限公司 Machine learning process implementation method and apparatus, device, and storage medium
CN109142374A (en) * 2018-08-15 2019-01-04 广州市心鉴智控科技有限公司 Method and system based on the efficient Checking model of extra small sample training
CN109255815A (en) * 2018-08-30 2019-01-22 天津理工大学 A kind of object detection and recognition methods based on order spherical harmonic
CN109389046A (en) * 2018-09-11 2019-02-26 昆山星际舟智能科技有限公司 Round-the-clock object identification and method for detecting lane lines for automatic Pilot
CN109389046B (en) * 2018-09-11 2022-03-29 昆山星际舟智能科技有限公司 All-weather object identification and lane line detection method for automatic driving
CN109359573A (en) * 2018-09-30 2019-02-19 天津天地伟业投资管理有限公司 A kind of warning method and device based on the separation of accurate people's vehicle
CN109614914A (en) * 2018-12-05 2019-04-12 北京纵目安驰智能科技有限公司 Parking stall vertex localization method, device and storage medium
CN109737974A (en) * 2018-12-14 2019-05-10 中国科学院深圳先进技术研究院 A kind of 3D navigational semantic map updating method, device and equipment
CN109670450A (en) * 2018-12-20 2019-04-23 天津天地伟业信息系统集成有限公司 A kind of people's vehicle object detecting method based on video
CN109671236A (en) * 2018-12-28 2019-04-23 广东美电贝尔科技集团股份有限公司 The detection method and its system of circumference target object
CN110008910A (en) * 2019-04-10 2019-07-12 江西理工大学 A kind of movie and television play is exposed the false the detection method of camera lens
CN110008910B (en) * 2019-04-10 2022-01-28 江西理工大学 Detection method for movie and television series cut-to-help lens
CN110349148A (en) * 2019-07-11 2019-10-18 电子科技大学 A kind of image object detection method based on Weakly supervised study
CN110827261A (en) * 2019-11-05 2020-02-21 泰康保险集团股份有限公司 Image quality detection method and device, storage medium and electronic equipment
CN111080608A (en) * 2019-12-12 2020-04-28 哈尔滨市科佳通用机电股份有限公司 Method for recognizing closing fault image of automatic brake valve plug handle of railway wagon in derailment
CN111369501A (en) * 2020-02-24 2020-07-03 武汉大学 Deep learning method for identifying oral squamous cell carcinoma based on visual features
CN112151167A (en) * 2020-05-14 2020-12-29 余红兵 Intelligent screening method for six-age dental caries of children based on deep learning
CN112183460A (en) * 2020-10-20 2021-01-05 武汉光谷联合集团有限公司 Method and device for intelligently identifying environmental sanitation
CN112418278A (en) * 2020-11-05 2021-02-26 中保车服科技服务股份有限公司 Multi-class object detection method, terminal device and storage medium

Also Published As

Publication number Publication date
CN107316007B (en) 2020-04-03

Similar Documents

Publication Publication Date Title
CN107316007A (en) A kind of monitoring image multiclass object detection and recognition methods based on deep learning
CN109359559B (en) Pedestrian re-identification method based on dynamic shielding sample
CN107134144B (en) A kind of vehicle checking method for traffic monitoring
WO2019140767A1 (en) Recognition system for security check and control method thereof
CN111563557B (en) Method for detecting target in power cable tunnel
CN108509954A (en) A kind of more car plate dynamic identifying methods of real-time traffic scene
CN109766769A (en) A kind of road target detection recognition method based on monocular vision and deep learning
CN108491797A (en) A kind of vehicle image precise search method based on big data
CN110084165A (en) The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations
CN105808610A (en) Internet picture filtering method and device
CN105608454A (en) Text structure part detection neural network based text detection method and system
CN111444939A (en) Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field
CN109740676A (en) Object detection moving method based on similar purpose
CN111428558A (en) Vehicle detection method based on improved YO L Ov3 method
CN103279738B (en) Automatic identification method and system for vehicle logo
CN112287896A (en) Unmanned aerial vehicle aerial image target detection method and system based on deep learning
CN110287806A (en) A kind of traffic sign recognition method based on improvement SSD network
CN107038416A (en) A kind of pedestrian detection method based on bianry image modified HOG features
CN107392251B (en) Method for improving target detection network performance by using classified pictures
CN107330027A (en) A kind of Weakly supervised depth station caption detection method
CN102867183A (en) Method and device for detecting littered objects of vehicle and intelligent traffic monitoring system
CN108198202A (en) A kind of video content detection method based on light stream and neural network
CN108256462A (en) A kind of demographic method in market monitor video
CN107818321A (en) A kind of watermark date recognition method for vehicle annual test
CN109800756A (en) A kind of text detection recognition methods for the intensive text of Chinese historical document

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
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20230519

Address after: Room 319-2, 3rd Floor, Building 2, No. 262 Wantang Road, Xihu District, Hangzhou City, Zhejiang Province, 310012

Patentee after: Hangzhou Jieshang Safety Equipment Co.,Ltd.

Address before: 311121 East Building, building 7, No. 998, Wenyi West Road, Wuchang Street, Yuhang District, Hangzhou City, Zhejiang Province

Patentee before: ZHEJIANG ICARE VISION TECHNOLOGY Co.,Ltd.

TR01 Transfer of patent right