CN109325418A - Based on pedestrian recognition method under the road traffic environment for improving YOLOv3 - Google Patents

Based on pedestrian recognition method under the road traffic environment for improving YOLOv3 Download PDF

Info

Publication number
CN109325418A
CN109325418A CN201810966748.5A CN201810966748A CN109325418A CN 109325418 A CN109325418 A CN 109325418A CN 201810966748 A CN201810966748 A CN 201810966748A CN 109325418 A CN109325418 A CN 109325418A
Authority
CN
China
Prior art keywords
road traffic
pedestrian
yolov3
traffic environment
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
Application number
CN201810966748.5A
Other languages
Chinese (zh)
Inventor
李巍华
方卓琳
刘晓楠
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.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201810966748.5A priority Critical patent/CN109325418A/en
Publication of CN109325418A publication Critical patent/CN109325418A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • 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
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Human Computer Interaction (AREA)
  • Traffic Control Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses based on pedestrian recognition method under the road traffic environment for improving YOLOv3, comprising steps of S1, Image Acquisition and pretreatment, and pedestrian sample collection is made；S2, pedestrian candidate frame length-width ratio is calculated using clustering algorithm and training set；S3, training set input YOLOv3 network is subjected to multitask training and saves trained weight file；S4, picture to be identified input YOLOv3 network is obtained into multiple dimensioned characteristic pattern；S5, x, y of neural network forecast, confidence level, class probability are activated using logistic function, through threshold decision, obtains coordinate, confidence level and the class probability of all prediction blocks；S6, the above results are handled into the final target detection frame and recognition result of generation by non-maxima suppression.The present invention solves the problems, such as that existing method Detection accuracy is low, realizes multitask training, and without additional storage space, detection accuracy is high, speed is fast.

Description

Based on pedestrian recognition method under the road traffic environment for improving YOLOv3
Technical field
The invention belongs to the driving of automotive safety auxiliary and field of image processings, more particularly, to one kind based on improvement Pedestrian recognition method under the road traffic environment of YOLOv3, can be for pedestrian's accuracy of identification under solving road traffic environment not Height detects the problem of time-consuming.
Background technique
Pedestrian's identification under road traffic environment, because the variability of the complexity of traffic environment, pedestrian's posture is always to count The difficult point of calculation machine vision research.Pedestrian accurately, is rapidly and effectively known as an important component in road environment Preferably people Chu Lai not can be assisted to drive.
Common pedestrian detection method mainly includes establishing deformation site model based on features such as LBP, Haar, HOG The method, the method based on deep learning of (Deformable Part Model, DPM) combining classification device.Identification based on DPM Method needs to establish the partial model of multiple pedestrians, computationally intensive and robustness is poor in complicated road environment.Depth The method of habit can effectively extract the recessive character of data essence using convolutional neural networks, and weight is shared, to road environment Lower different pedestrian's posture has preferable robustness and accuracy of identification.
2018, Redmon J et al. was in document " Redmon J, Farhadi A.YOLOv3:An Incremental YOLOv3 detection framework is proposed in Improvement [J] .2018. ", the frame is at (52 layers+1 layer of convolutional layer of darknet-53 Full articulamentum) on the basis of sorter network, remove full articulamentum, increase convolutional layer and up-sampling layer, network uses Resnet The thought of cross-layer jump connection, obtains the output characteristic pattern of three scales.Herein based on the detection framework of YOLOv3, k- is utilized Means clustering algorithm obtains the priori knowledge of pedestrian's length-width ratio on different scale, then changes loss function in training process and sit The weight for marking error, improves the positioning accuracy and detection accuracy of network.
Summary of the invention
The purpose of the present invention is being directed to the deficiency of existing detection method detection accuracy and speed, provide a kind of based on improvement Pedestrian recognition method under the road traffic environment of YOLOv3, this method can in real time, be accurately identified under road traffic environment Pedestrian facilitates driver under complicated road traffic environment and preferably perceives peripheral vehicle environment, prevents traffic accident Occur.
To achieve the goals above, technical scheme is as follows:
It is a kind of based on improve YOLOv3 road traffic environment under pedestrian recognition method, comprising the following steps:
Pedestrian image acquisition and pretreatment under S1, different road traffic environments, and make the row under road traffic environment People's sample set；
S2, it is based on training set, using improved k-means clustering algorithm, calculates the pedestrian candidate under road traffic environment Frame length-width ratio；
S3, the super ginseng of setting training and network parameter, input network for training set, carry out multitask instruction to YOLOv3 network Practice, and saves trained weight file；
S4, picture to be identified is inputted into YOLOv3 network, by several convolutional layers and up-sampling layer, obtained multiple dimensioned Characteristic pattern；
S5, x, y of neural network forecast, confidence level, class probability are activated using logistic function, is sentenced through threshold value It is disconnected, obtain coordinate, confidence level and the class probability of all prediction blocks；
S6, the above results are handled into the final target detection frame and recognition result of generation by non-maxima suppression.
Further, it is specifically included in the step S1:
S11, vehicle-mounted camera or vehicle-mounted traveling recorder, the running information under captured in real-time road traffic environment are opened；
S12, the running information taken is subjected to sub-frame processing, obtains road traffic environment down train image collection sequence Column；
S13, image collection is screened, chooses different illumination conditions, traffic slot, environmental background and pedestrian's posture Pedestrian image；
S14, the picture of selection is labeled, using annotation tool, outlines target area and tagged；
S15, pedestrian's data set under road traffic environment is randomly divided into training set and test set according to a certain percentage.
Further, it is specifically included in the step S2:
S21, range formula is used as with d (box, centroid)=1-IOU (box, centroid), is based on training set, just Beginning dissolves 15 mutual IOU value candidate frame length-width ratios as small as possible, as cluster centre；
Sample pane is divided into the maximum one kind of IOU value by S22, the IOU value for successively calculating sample pane and cluster centre；
S23, it goes through after all sample panes, recalculates the length and width of every class cluster core candidate frame；
S24, above-mentioned two step is repeated, is until there is no the number of sample pane replacement cluster centre or iteration to reach threshold value Only.
Further, include: in the step S3
S31, the initial learning rate of setting network training, learning strategy, initial input picture size, port number, maximum training Iterative steps, color saturation, rotation angle, exposure rate, tone reversal；
S32, setting network are multiple dimensioned training, and picture size is adjusted at random are as follows: (rand () %10+10) * 32, with suitable Answer the picture of different input sizes；
S33, the weight for increasing coordinate prediction damage error, by the weight of coordinate prediction loss by (2-truth.w* Truth.h) it is set as (5-truth.w*truth.h).
Further, the YOLOv3 network structure includes: 75 convolutional layers, 2 up-sampling layers and 3 different scales Output layer.
Further, the size of 3 different scale output layers is respectively 13x13,26x26,52x52, each ruler 5 candidate frames are distributed on the characteristic pattern of degree, and follow large scale feature frame detection small-size object, and the detection of small size features frame is big Size objects principle.
Further, in the YOLOv3 network, in the detection layers of multitask training, network swashs by logistic function It is living to obtain final coordinate prediction and classification prediction.
Further, in the step S6, all detection blocks that will test carry out non-maxima suppression and reject overlapping time Frame is selected, the detection block coordinate and class prediction of highest scoring are finally obtained.
Compared to pedestrian's technology under existing road traffic environment, the present invention solves pedestrian target detection hardly possible, slow-footed Problem, in order to improve positioning accuracy, propose it is a kind of based on improve YOLOv3 road traffic environment under pedestrian recognition method, should Method effectively verifies network under practical driving environment to the knowledge of pedestrian by pedestrian's data set under production road traffic environment Other effect.The present invention learns the posture feature of pedestrian by deep learning from a large amount of sample, and can extract reflection number According to the stealth characteristics of essence, robustness with higher.The method that the present invention uses step detection, before guaranteeing detection accuracy It puts, improves detection speed, be configured to Intel's i7 Duo 6 generation CPU, memory 8G, NVIDA Geforce GTX 1060 In 16.04 system of Ubuntu, real-time detection is can be achieved in the computer of video card (6G) substantially.
Detailed description of the invention
The present invention provides attached drawings in order to further understand to disclosure, will simply be situated between to attached drawing below It continues.Attached drawing constitutes part of this application, but only as the non-limiting example for embodying concept of the invention, is not intended to take the post as What is limited.
Fig. 1 is the embodiment of the present invention based on pedestrian recognition method whole implementation under the road traffic environment for improving YOLOv3 Process flow diagram flow chart.
Fig. 2 is the YOLOv3 network structure of model embodiment of the present invention.
Fig. 3 is the algorithm flow chart of model embodiment cluster candidate frame length-width ratio of the present invention.
Fig. 4 is prediction of the network to coordinate in model embodiment multitask training of the present invention.
Fig. 5 is that the evaluation improvement YOLOv3 network detection performance index of the invention that is used for illustrates schematic diagram.
Specific embodiment
In order to clearly show advantages of the present invention, embodiment, be described with reference to the drawings, further to the present invention into Row illustrates.
Fig. 1 is that the present invention is a kind of based on pedestrian recognition method whole implementation process under the road traffic environment for improving YOLOv3 Flow chart, Fig. 2 are YOLOv3 network structure, and figure (3) is candidate frame clustering algorithm flow chart, in conjunction with Fig. 1, Fig. 2 and Fig. 3, originally The step of invention specific embodiment are as follows:
It is a kind of based on improve YOLOv3 road traffic environment under pedestrian recognition method, comprising the following steps:
Pedestrian image acquisition and pretreatment under S1, different road traffic environments, and make the row under road traffic environment People's sample set；
S2, it is based on training set, using improved k-means clustering algorithm, calculates the pedestrian candidate under road traffic environment Frame length-width ratio；
S3, the super ginseng of setting training and network parameter, input network for training set, carry out multitask instruction to YOLOv3 network Practice, and saves trained weight file；
S4, picture to be identified is inputted into YOLOv3 network, by several convolutional layers and up-sampling layer, obtained multiple dimensioned Characteristic pattern；
S5, x, y of neural network forecast, confidence level, class probability are activated using logistic function, is sentenced through threshold value It is disconnected, obtain coordinate, confidence level and the class probability of all prediction blocks；
S6, the above results are handled into the final target detection frame and recognition result of generation by non-maxima suppression.
Specifically, being specifically included in the step S1:
S11, vehicle-mounted camera or vehicle-mounted traveling recorder, the running information under captured in real-time road traffic environment are opened；
S12, the running information taken is subjected to sub-frame processing, obtains road traffic environment down train image collection sequence Column；
S13, image collection is screened, chooses different illumination conditions, traffic slot, environmental background and pedestrian's posture Pedestrian image；
S14, the picture of selection is labeled, using annotation tool, outlines target area and tagged；
S15, pedestrian's data set under road traffic environment is randomly divided into training set and test set according to a certain percentage.
Specifically, as shown in figure 3, being specifically included in the step S2:
S21, range formula is used as with d (box, centroid)=1-IOU (box, centroid), is based on training set, just Beginning dissolves 15 mutual IOU value candidate frame length-width ratios as small as possible, as cluster centre；
Sample pane is divided into the maximum one kind of IOU value by S22, the IOU value for successively calculating sample pane and cluster centre；
S23, it goes through after all sample panes, recalculates the length and width of every class cluster core candidate frame；
S24, above-mentioned two step is repeated, is until there is no the number of sample pane replacement cluster centre or iteration to reach threshold value Only.
Specifically, including: in the step S3
S31, the initial learning rate of setting network training, learning strategy, initial input picture size, port number, maximum training Iterative steps, color saturation, rotation angle, exposure rate, tone reversal；
S32, setting network are multiple dimensioned training, and picture size is adjusted at random are as follows: (rand () %10+10) * 32, with suitable Answer the picture of different input sizes；
S33, the weight for increasing coordinate prediction damage error, by the weight of coordinate prediction loss by (2-truth.w* Truth.h) it is set as (5-truth.w*truth.h).
Specifically, as shown in Fig. 2, the YOLOv3 network structure includes: 75 convolutional layers, 2 up-sampling layers and 3 The output layer of different scale, the size of 3 different scale output layers are respectively 13x13,26x26,52x52, each ruler 5 candidate frames are distributed on the characteristic pattern of degree, and follow large scale feature frame detection small-size object, and the detection of small size features frame is big Size objects principle.
In the YOLOv3 network, in the detection layers of multitask training, net of the network after the activation of logistic function Network output matching obtains final coordinate prediction and classification prediction.
Specifically, all detection blocks that will test carry out non-maxima suppression and reject overlapping time in the step S6 Frame is selected, the detection block coordinate and class prediction of highest scoring are finally obtained.
Target frame and which candidate frame IOU value maximum determine to be responsible for detecting target frame, each sample by which output characteristic pattern This frame at most only matches a prediction block, which grid the center of sample pane falls in, this grid is just responsible for predicting this mesh Mark frame.
Wherein, the calculation formula of IOU are as follows:
The method that the training error of the YOLOv3 network uses smooth L2loss；
Wherein, the confidence level loss of the prediction block not comprising target are as follows:
It is responsible for detecting the detection block confidence level loss of sample pane are as follows:
Classification loss are as follows:
Coordinate prediction loss are as follows:
Wherein,For the predicted value after network output conversion, b is used respectively in attached drawingx、by、bw、bhTable Show,For neural network forecast value, x, y, w, h, P, C are training standard value.Indicate j-th of object and i grid Anchor box matching.
The network reality output of coordinate prediction is tx、ty、tw、th, b is obtained by conversionx、by、bw、bh, formula are as follows:
Combining classification, confidence level and coordinate prediction loss, the whole loss function of the yolov3 network is, wherein coordinate Predicted value is as shown in Figure 4:
Sample pane targetedly trains network output, while classification, confidence level and coordinate prediction are realized all by returning The global optimization of network objectives function, until Loss restrains, a step detection framework can also greatly reduce calculation amount.
The present invention is evaluated with mAP and improves YOLOv3 network objectives detection performance, mAP (mean Average Precision) Be cumulative in recall rate of the detection accuracy of each classification and, be an important indicator of evaluation goal detection network performance, The present invention is just for one object of pedestrian, it is only necessary to calculate the AP of pedestrian.
The calculation formula of AP are as follows:
A P=∫ P d R
Wherein P is detection accuracy (precision), and R is recall rate Recall.As shown in Figure 5.P, R is defined as:
Table 1 is the performance comparison result for improving YOLOv3 network and other detection frameworks；
model Training set Test set fps AP Device name
Faster R-CNN_VGG16 COCO COCO 7.7 54.35 GeForce GTX TITAN X
300x300 SSD_VGG16 COCO COCO 25 51.7 GeForce GTX TITAN X
416x416 YOLOv3 COCO COCO 29.6 71.68 GeForce GTX TITAN X
COCO data set is randomly divided into training set and test set with certain proportion, training Faster R-CNN, SSD, YOLOv3 detection framework, and tested on GeForce GTX TITAN X (12G) video card, Faster R-CNN detection speed is 7.7fps, precision are examined up to 54.35%, 300x300SSD VGG16 detection speed up to 25fps, AP 51.70%, YOLOv3 Degree of testing the speed be 29.6fps, AP 71.68%, as shown in table 1.
According to table 1, it is compared to Faster R-CNN and SSD, YOLOv3 has in terms of detecting speed and precision Obvious advantage.
To sum up, the invention proposes a kind of based on pedestrian recognition method under the road traffic environment for improving YOLOv3.The party Method learning characteristic from great amount of samples by the method for deep learning can extract the recessive character of reflection data essence, tool There are higher learning efficiency and accuracy of identification, improve the robustness of algorithm, effectively increases pedestrian under road traffic environment and know Other accuracy.It can largely solve that pedestrian serious shielding, posture under complicated road traffic environment be changeable, illumination variation The problem of factors bring detection difficult such as big.Part Methods step herein and process may need to be executed by computer, To by hardware, software, firmware and its it is any combination of in a manner of implement.
The above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be to the present invention Embodiment restriction.For those of ordinary skill in the art, it can also make on the basis of the above description Other various forms of variations or variation.There is no necessity and possibility to exhaust all the enbodiments.It is all of the invention Made any modifications, equivalent replacements, and improvements etc., should be included in the protection of the claims in the present invention within spirit and principle Within the scope of.

Claims (8)

1. a kind of based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which comprises the following steps:
Pedestrian image acquisition and pretreatment under S1, different road traffic environments, and make pedestrian's sample under road traffic environment This collection；
S2, it is based on training set, using improved k-means clustering algorithm, the pedestrian candidate frame calculated under road traffic environment is long Wide ratio；
S3, the super ginseng of setting training and network parameter, input network for training set, carry out multitask training to YOLOv3 network, and Save trained weight file；
S4, picture to be identified is inputted into YOLOv3 network, by several convolutional layers and up-sampling layer, obtains multiple dimensioned feature Figure；
S5, x, y of neural network forecast, confidence level, class probability are activated using logistic function, through threshold decision, is obtained To the coordinate of all prediction blocks, confidence level and class probability；
S6, the above results are handled into the final target detection frame and recognition result of generation by non-maxima suppression.
2. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that It is specifically included in the step S1:
S11, vehicle-mounted camera or vehicle-mounted traveling recorder, the running information under captured in real-time road traffic environment are opened；
S12, the running information taken is subjected to sub-frame processing, obtains road traffic environment down train image collection sequence；
S13, image collection is screened, chooses the pedestrian of different illumination conditions, traffic slot, environmental background and pedestrian's posture Image；
S14, the picture of selection is labeled, using annotation tool, outlines target area and tagged；
S15, pedestrian's data set under road traffic environment is randomly divided into training set and test set according to a certain percentage.
3. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that It is specifically included in the step S2:
S21, with QUOTE As range formula, it is based on training set, initially dissolves 15 phases IOU value candidate frame length-width ratio as small as possible between mutually, as cluster centre；
Sample pane is divided into the maximum one kind of IOU value by S22, the IOU value for successively calculating sample pane and cluster centre；
S23, it goes through after all sample panes, recalculates the length and width of every class cluster core candidate frame；
S24, above-mentioned two step is repeated, until there is no the number of sample pane replacement cluster centre or iteration to reach threshold value.
4. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that Include: in the step S3
S31, the initial learning rate of setting network training, learning strategy, initial input picture size, port number, maximum training iteration Step number, color saturation, rotation angle, exposure rate, tone reversal；
S32, setting network are multiple dimensioned training, picture size are adjusted at random are as follows: QUOTE , to adapt to the picture of different input sizes；
S33, the weight for increasing coordinate prediction damage error, by the weight of coordinate prediction loss by QUOTE It is set as QUOTE 。
5. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that The YOLOv3 network structure includes: 75 convolutional layers, the output layer of 2 up-samplings layers and 3 different scales.
6. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that The size of 3 different scale output layers is respectively QUOTE , 5 candidate frames are distributed on the characteristic pattern of each scale, and it is small to follow the detection of large scale feature frame Size objects, small size features frame detect large sized object principle.
7. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that In the YOLOv3 network, in the detection layers of multitask training, network activates to obtain final coordinate by logistic function Prediction and classification prediction.
8. as described in claim 1 based on pedestrian recognition method under the road traffic environment for improving YOLOv3, which is characterized in that In the step S6, all detection blocks that will test carry out non-maxima suppression and reject overlapping candidate frame, finally obtain score Highest detection block coordinate and class prediction.
CN201810966748.5A 2018-08-23 2018-08-23 Based on pedestrian recognition method under the road traffic environment for improving YOLOv3 Pending CN109325418A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810966748.5A CN109325418A (en) 2018-08-23 2018-08-23 Based on pedestrian recognition method under the road traffic environment for improving YOLOv3

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810966748.5A CN109325418A (en) 2018-08-23 2018-08-23 Based on pedestrian recognition method under the road traffic environment for improving YOLOv3

Publications (1)

Publication Number Publication Date
CN109325418A true CN109325418A (en) 2019-02-12

Family

ID=65264280

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810966748.5A Pending CN109325418A (en) 2018-08-23 2018-08-23 Based on pedestrian recognition method under the road traffic environment for improving YOLOv3

Country Status (1)

Country Link
CN (1) CN109325418A (en)

Cited By (58)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109977797A (en) * 2019-03-06 2019-07-05 上海交通大学 The optimization method of single order object detector based on sequence loss function
CN109993122A (en) * 2019-04-02 2019-07-09 中国石油大学(华东) A kind of pedestrian based on depth convolutional neural networks multiplies staircase anomaly detection method
CN110008853A (en) * 2019-03-15 2019-07-12 华南理工大学 Pedestrian detection network and model training method, detection method, medium, equipment
CN110011727A (en) * 2019-04-09 2019-07-12 浩鲸云计算科技股份有限公司 A kind of detection system towards ODF device port
CN110059556A (en) * 2019-03-14 2019-07-26 天津大学 A kind of transformer substation switch division condition detection method based on deep learning
CN110084166A (en) * 2019-04-19 2019-08-02 山东大学 Substation's smoke and fire intelligent based on deep learning identifies monitoring method
CN110110650A (en) * 2019-05-02 2019-08-09 西安电子科技大学 Face identification method in pedestrian
CN110147771A (en) * 2019-05-23 2019-08-20 南京农业大学 Sow side-lying position real-time detecting system based on sow key position Yu environment joint partition
CN110188807A (en) * 2019-05-21 2019-08-30 重庆大学 Tunnel pedestrian target detection method based on cascade super-resolution network and improvement Faster R-CNN
CN110245689A (en) * 2019-05-23 2019-09-17 杭州有容智控科技有限公司 Shield cutter identification and position finding and detection method based on machine vision
CN110263660A (en) * 2019-05-27 2019-09-20 魏运 A kind of traffic target detection recognition method of adaptive scene changes
CN110287822A (en) * 2019-06-10 2019-09-27 浙江大学城市学院 The snail pest control method of view-based access control model target detection in a kind of officinal dendrobium stem plantation
CN110472597A (en) * 2019-07-31 2019-11-19 中铁二院工程集团有限责任公司 Rock image rate of decay detection method and system based on deep learning
CN110503098A (en) * 2019-08-29 2019-11-26 西安电子科技大学 A kind of object detection method and equipment of quick real-time lightweight
CN110533640A (en) * 2019-08-15 2019-12-03 北京交通大学 Based on the track circuit disease discrimination method for improving YOLOv3 network model
CN110555425A (en) * 2019-09-11 2019-12-10 上海海事大学 Video stream real-time pedestrian detection method
CN110659585A (en) * 2019-08-31 2020-01-07 电子科技大学 Pedestrian detection method based on interactive attribute supervision
CN110766726A (en) * 2019-10-17 2020-02-07 重庆大学 Visual positioning and dynamic tracking method for moving target of large bell jar container under complex background
CN110781964A (en) * 2019-10-28 2020-02-11 兰州交通大学 Human body target detection method and system based on video image
CN110852177A (en) * 2019-10-17 2020-02-28 北京全路通信信号研究设计院集团有限公司 Obstacle detection method and system based on monocular camera
CN110929577A (en) * 2019-10-23 2020-03-27 桂林电子科技大学 Improved target identification method based on YOLOv3 lightweight framework
CN110929593A (en) * 2019-11-06 2020-03-27 哈尔滨工业大学（威海） Real-time significance pedestrian detection method based on detail distinguishing and distinguishing
CN110929646A (en) * 2019-11-22 2020-03-27 国网福建省电力有限公司 Power distribution tower reverse-off information rapid identification method based on unmanned aerial vehicle aerial image
CN110942005A (en) * 2019-11-21 2020-03-31 网易（杭州）网络有限公司 Object recognition method and device
CN110992714A (en) * 2019-12-18 2020-04-10 佛山科学技术学院 Intelligent traffic signal lamp control method and system
CN111046797A (en) * 2019-12-12 2020-04-21 天地伟业技术有限公司 Oil pipeline warning method based on personnel and vehicle behavior analysis
CN111091110A (en) * 2019-12-24 2020-05-01 山东仁功智能科技有限公司 Wearing identification method of reflective vest based on artificial intelligence
CN111104965A (en) * 2019-11-25 2020-05-05 河北科技大学 Vehicle target identification method and device
CN111191621A (en) * 2020-01-03 2020-05-22 北京同方软件有限公司 Rapid and accurate identification method for multi-scale target under large-focus monitoring scene
CN111259736A (en) * 2020-01-08 2020-06-09 上海海事大学 Real-time pedestrian detection method based on deep learning in complex environment
CN111310622A (en) * 2020-02-05 2020-06-19 西北工业大学 Fish swarm target identification method for intelligent operation of underwater robot
CN111353393A (en) * 2020-02-19 2020-06-30 桂林电子科技大学 Dog only detects and early warning system based on neural network
CN111401148A (en) * 2020-02-27 2020-07-10 江苏大学 Road multi-target detection method based on improved multilevel YO L Ov3
CN111444809A (en) * 2020-03-23 2020-07-24 华南理工大学 Power transmission line abnormal target detection method based on improved YO L Ov3
CN111597967A (en) * 2020-05-12 2020-08-28 北京大学 Infrared image multi-target pedestrian identification method
CN111626128A (en) * 2020-04-27 2020-09-04 江苏大学 Improved YOLOv 3-based pedestrian detection method in orchard environment
CN111652134A (en) * 2020-06-02 2020-09-11 电子科技大学中山学院 Vehicle-mounted pedestrian detection system and method based on microprocessor
CN111709489A (en) * 2020-06-24 2020-09-25 广西师范大学 Citrus identification method based on improved YOLOv4
CN111738088A (en) * 2020-05-25 2020-10-02 西安交通大学 Pedestrian distance prediction method based on monocular camera
CN111832489A (en) * 2020-07-15 2020-10-27 中国电子科技集团公司第三十八研究所 Subway crowd density estimation method and system based on target detection
CN111832608A (en) * 2020-05-29 2020-10-27 上海海事大学 Multi-abrasive-particle identification method for ferrographic image based on single-stage detection model yolov3
CN111860072A (en) * 2019-04-30 2020-10-30 广州汽车集团股份有限公司 Parking control method and device, computer equipment and computer readable storage medium
CN111860323A (en) * 2020-07-20 2020-10-30 北京华正明天信息技术股份有限公司 Method for identifying initial fire in monitoring picture based on yolov3 algorithm
CN111860679A (en) * 2020-07-29 2020-10-30 浙江理工大学 Vehicle detection method based on YOLO v3 improved algorithm
CN112001339A (en) * 2020-08-27 2020-11-27 杭州电子科技大学 Pedestrian social distance real-time monitoring method based on YOLO v4
CN112084870A (en) * 2020-08-10 2020-12-15 同济大学 YOLO-based multi-target detection method and device in traffic scene
CN112149476A (en) * 2019-06-28 2020-12-29 北京海益同展信息科技有限公司 Target detection method, device, equipment and storage medium
CN112233175A (en) * 2020-09-24 2021-01-15 西安交通大学 Chip positioning method based on YOLOv3-tiny algorithm and integrated positioning platform
CN112307853A (en) * 2019-08-02 2021-02-02 成都天府新区光启未来技术研究院 Detection method of aerial image, storage medium and electronic device
CN112365324A (en) * 2020-12-02 2021-02-12 杭州微洱网络科技有限公司 Commodity picture detection method suitable for E-commerce platform
CN112381032A (en) * 2020-11-24 2021-02-19 华南理工大学 Indoor unattended rapid detection method resisting human posture interference
CN112434681A (en) * 2021-01-27 2021-03-02 武汉星巡智能科技有限公司 Intelligent camera self-training confidence threshold selection method, device and equipment
CN112633159A (en) * 2020-12-22 2021-04-09 北京迈格威科技有限公司 Human-object interaction relation recognition method, model training method and corresponding device
CN112686128A (en) * 2020-12-28 2021-04-20 南京览众智能科技有限公司 Classroom desk detection method based on machine learning
CN112733679A (en) * 2020-12-31 2021-04-30 南京视察者智能科技有限公司 Case logic reasoning-based early warning system and training method
CN113011405A (en) * 2021-05-25 2021-06-22 南京柠瑛智能科技有限公司 Method for solving multi-frame overlapping error of ground object target identification of unmanned aerial vehicle
CN113205028A (en) * 2021-04-26 2021-08-03 河海大学 Pedestrian detection method and system based on improved YOLOv3 model
CN113420716A (en) * 2021-07-16 2021-09-21 南威软件股份有限公司 Improved Yolov3 algorithm-based violation behavior recognition and early warning method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017206066A1 (en) * 2016-05-31 2017-12-07 Nokia Technologies Oy Method and apparatus for detecting small objects with an enhanced deep neural network
US20180211117A1 (en) * 2016-12-20 2018-07-26 Jayant Ratti On-demand artificial intelligence and roadway stewardship system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017206066A1 (en) * 2016-05-31 2017-12-07 Nokia Technologies Oy Method and apparatus for detecting small objects with an enhanced deep neural network
US20180211117A1 (en) * 2016-12-20 2018-07-26 Jayant Ratti On-demand artificial intelligence and roadway stewardship system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JOSEPH REDMON等: "YOLOv3: An Incremental Improvement", 《COMPUTER VISION AND PATTERN RECOGNITION》 *
JOSEPH REDMON等: "You Only Look Once: Unified, Real-Time Object Detection", 《HTTPS://ARXIV.ORG/PDF/1506.02640.PDF》 *
王殿伟等: "改进的YOLOv3红外视频图像行人检测算法", 《西安邮电大学学报》 *

Cited By (78)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109977797B (en) * 2019-03-06 2023-06-20 上海交通大学 Optimization method of first-order target detector based on sorting loss function
CN109977797A (en) * 2019-03-06 2019-07-05 上海交通大学 The optimization method of single order object detector based on sequence loss function
CN110059556A (en) * 2019-03-14 2019-07-26 天津大学 A kind of transformer substation switch division condition detection method based on deep learning
CN110008853A (en) * 2019-03-15 2019-07-12 华南理工大学 Pedestrian detection network and model training method, detection method, medium, equipment
CN109993122A (en) * 2019-04-02 2019-07-09 中国石油大学(华东) A kind of pedestrian based on depth convolutional neural networks multiplies staircase anomaly detection method
CN110011727A (en) * 2019-04-09 2019-07-12 浩鲸云计算科技股份有限公司 A kind of detection system towards ODF device port
CN110084166A (en) * 2019-04-19 2019-08-02 山东大学 Substation's smoke and fire intelligent based on deep learning identifies monitoring method
CN111860072A (en) * 2019-04-30 2020-10-30 广州汽车集团股份有限公司 Parking control method and device, computer equipment and computer readable storage medium
CN110110650A (en) * 2019-05-02 2019-08-09 西安电子科技大学 Face identification method in pedestrian
CN110188807A (en) * 2019-05-21 2019-08-30 重庆大学 Tunnel pedestrian target detection method based on cascade super-resolution network and improvement Faster R-CNN
CN110188807B (en) * 2019-05-21 2023-04-21 重庆大学 Tunnel pedestrian target detection method based on cascading super-resolution network and improved Faster R-CNN
CN110245689A (en) * 2019-05-23 2019-09-17 杭州有容智控科技有限公司 Shield cutter identification and position finding and detection method based on machine vision
CN110147771A (en) * 2019-05-23 2019-08-20 南京农业大学 Sow side-lying position real-time detecting system based on sow key position Yu environment joint partition
CN110263660A (en) * 2019-05-27 2019-09-20 魏运 A kind of traffic target detection recognition method of adaptive scene changes
CN110287822A (en) * 2019-06-10 2019-09-27 浙江大学城市学院 The snail pest control method of view-based access control model target detection in a kind of officinal dendrobium stem plantation
CN112149476A (en) * 2019-06-28 2020-12-29 北京海益同展信息科技有限公司 Target detection method, device, equipment and storage medium
CN110472597A (en) * 2019-07-31 2019-11-19 中铁二院工程集团有限责任公司 Rock image rate of decay detection method and system based on deep learning
CN112307853A (en) * 2019-08-02 2021-02-02 成都天府新区光启未来技术研究院 Detection method of aerial image, storage medium and electronic device
CN110533640B (en) * 2019-08-15 2022-03-01 北京交通大学 Improved YOLOv3 network model-based track line defect identification method
CN110533640A (en) * 2019-08-15 2019-12-03 北京交通大学 Based on the track circuit disease discrimination method for improving YOLOv3 network model
CN110503098A (en) * 2019-08-29 2019-11-26 西安电子科技大学 A kind of object detection method and equipment of quick real-time lightweight
CN110659585B (en) * 2019-08-31 2022-03-15 电子科技大学 Pedestrian detection method based on interactive attribute supervision
CN110659585A (en) * 2019-08-31 2020-01-07 电子科技大学 Pedestrian detection method based on interactive attribute supervision
CN110555425A (en) * 2019-09-11 2019-12-10 上海海事大学 Video stream real-time pedestrian detection method
CN110766726A (en) * 2019-10-17 2020-02-07 重庆大学 Visual positioning and dynamic tracking method for moving target of large bell jar container under complex background
CN110766726B (en) * 2019-10-17 2022-03-01 重庆大学 Visual positioning and dynamic tracking method for moving target of large bell jar container under complex background
CN110852177A (en) * 2019-10-17 2020-02-28 北京全路通信信号研究设计院集团有限公司 Obstacle detection method and system based on monocular camera
CN110929577A (en) * 2019-10-23 2020-03-27 桂林电子科技大学 Improved target identification method based on YOLOv3 lightweight framework
CN110781964A (en) * 2019-10-28 2020-02-11 兰州交通大学 Human body target detection method and system based on video image
CN110929593B (en) * 2019-11-06 2023-06-20 哈尔滨工业大学（威海） Real-time significance pedestrian detection method based on detail discrimination
CN110929593A (en) * 2019-11-06 2020-03-27 哈尔滨工业大学（威海） Real-time significance pedestrian detection method based on detail distinguishing and distinguishing
CN110942005A (en) * 2019-11-21 2020-03-31 网易（杭州）网络有限公司 Object recognition method and device
CN110929646A (en) * 2019-11-22 2020-03-27 国网福建省电力有限公司 Power distribution tower reverse-off information rapid identification method based on unmanned aerial vehicle aerial image
CN111104965A (en) * 2019-11-25 2020-05-05 河北科技大学 Vehicle target identification method and device
CN111046797A (en) * 2019-12-12 2020-04-21 天地伟业技术有限公司 Oil pipeline warning method based on personnel and vehicle behavior analysis
CN110992714A (en) * 2019-12-18 2020-04-10 佛山科学技术学院 Intelligent traffic signal lamp control method and system
CN111091110A (en) * 2019-12-24 2020-05-01 山东仁功智能科技有限公司 Wearing identification method of reflective vest based on artificial intelligence
CN111091110B (en) * 2019-12-24 2023-11-17 山东仁功智能科技有限公司 Reflection vest wearing recognition method based on artificial intelligence
CN111191621A (en) * 2020-01-03 2020-05-22 北京同方软件有限公司 Rapid and accurate identification method for multi-scale target under large-focus monitoring scene
CN111259736B (en) * 2020-01-08 2023-04-07 上海海事大学 Real-time pedestrian detection method based on deep learning in complex environment
CN111259736A (en) * 2020-01-08 2020-06-09 上海海事大学 Real-time pedestrian detection method based on deep learning in complex environment
CN111310622A (en) * 2020-02-05 2020-06-19 西北工业大学 Fish swarm target identification method for intelligent operation of underwater robot
CN111353393A (en) * 2020-02-19 2020-06-30 桂林电子科技大学 Dog only detects and early warning system based on neural network
CN111401148A (en) * 2020-02-27 2020-07-10 江苏大学 Road multi-target detection method based on improved multilevel YO L Ov3
CN111444809A (en) * 2020-03-23 2020-07-24 华南理工大学 Power transmission line abnormal target detection method based on improved YO L Ov3
CN111444809B (en) * 2020-03-23 2023-02-14 华南理工大学 Power transmission line abnormal target detection method based on improved YOLOv3
CN111626128B (en) * 2020-04-27 2023-07-21 江苏大学 Pedestrian detection method based on improved YOLOv3 in orchard environment
CN111626128A (en) * 2020-04-27 2020-09-04 江苏大学 Improved YOLOv 3-based pedestrian detection method in orchard environment
CN111597967B (en) * 2020-05-12 2023-04-07 北京大学 Infrared image multi-target pedestrian identification method
CN111597967A (en) * 2020-05-12 2020-08-28 北京大学 Infrared image multi-target pedestrian identification method
CN111738088A (en) * 2020-05-25 2020-10-02 西安交通大学 Pedestrian distance prediction method based on monocular camera
CN111738088B (en) * 2020-05-25 2022-10-25 西安交通大学 Pedestrian distance prediction method based on monocular camera
CN111832608B (en) * 2020-05-29 2023-09-12 上海海事大学 Iron spectrum image multi-abrasive particle identification method based on single-stage detection model yolov3
CN111832608A (en) * 2020-05-29 2020-10-27 上海海事大学 Multi-abrasive-particle identification method for ferrographic image based on single-stage detection model yolov3
CN111652134A (en) * 2020-06-02 2020-09-11 电子科技大学中山学院 Vehicle-mounted pedestrian detection system and method based on microprocessor
CN111709489A (en) * 2020-06-24 2020-09-25 广西师范大学 Citrus identification method based on improved YOLOv4
CN111709489B (en) * 2020-06-24 2022-04-08 广西师范大学 Citrus identification method based on improved YOLOv4
CN111832489A (en) * 2020-07-15 2020-10-27 中国电子科技集团公司第三十八研究所 Subway crowd density estimation method and system based on target detection
CN111860323A (en) * 2020-07-20 2020-10-30 北京华正明天信息技术股份有限公司 Method for identifying initial fire in monitoring picture based on yolov3 algorithm
CN111860679A (en) * 2020-07-29 2020-10-30 浙江理工大学 Vehicle detection method based on YOLO v3 improved algorithm
CN112084870A (en) * 2020-08-10 2020-12-15 同济大学 YOLO-based multi-target detection method and device in traffic scene
CN112001339B (en) * 2020-08-27 2024-02-23 杭州电子科技大学 Pedestrian social distance real-time monitoring method based on YOLO v4
CN112001339A (en) * 2020-08-27 2020-11-27 杭州电子科技大学 Pedestrian social distance real-time monitoring method based on YOLO v4
CN112233175A (en) * 2020-09-24 2021-01-15 西安交通大学 Chip positioning method based on YOLOv3-tiny algorithm and integrated positioning platform
CN112233175B (en) * 2020-09-24 2023-10-24 西安交通大学 Chip positioning method and integrated positioning platform based on YOLOv3-tiny algorithm
CN112381032A (en) * 2020-11-24 2021-02-19 华南理工大学 Indoor unattended rapid detection method resisting human posture interference
CN112381032B (en) * 2020-11-24 2024-03-22 华南理工大学 Indoor unattended rapid detection method for resisting human body posture interference
CN112365324A (en) * 2020-12-02 2021-02-12 杭州微洱网络科技有限公司 Commodity picture detection method suitable for E-commerce platform
CN112633159B (en) * 2020-12-22 2024-04-12 北京迈格威科技有限公司 Human-object interaction relation identification method, model training method and corresponding device
CN112633159A (en) * 2020-12-22 2021-04-09 北京迈格威科技有限公司 Human-object interaction relation recognition method, model training method and corresponding device
CN112686128A (en) * 2020-12-28 2021-04-20 南京览众智能科技有限公司 Classroom desk detection method based on machine learning
CN112733679B (en) * 2020-12-31 2023-09-01 南京视察者智能科技有限公司 Early warning system and training method based on case logic reasoning
CN112733679A (en) * 2020-12-31 2021-04-30 南京视察者智能科技有限公司 Case logic reasoning-based early warning system and training method
CN112434681A (en) * 2021-01-27 2021-03-02 武汉星巡智能科技有限公司 Intelligent camera self-training confidence threshold selection method, device and equipment
CN113205028A (en) * 2021-04-26 2021-08-03 河海大学 Pedestrian detection method and system based on improved YOLOv3 model
CN113011405A (en) * 2021-05-25 2021-06-22 南京柠瑛智能科技有限公司 Method for solving multi-frame overlapping error of ground object target identification of unmanned aerial vehicle
CN113420716A (en) * 2021-07-16 2021-09-21 南威软件股份有限公司 Improved Yolov3 algorithm-based violation behavior recognition and early warning method
CN113420716B (en) * 2021-07-16 2023-07-28 南威软件股份有限公司 Illegal behavior identification and early warning method based on improved Yolov3 algorithm

Similar Documents

Publication Publication Date Title
CN109325418A (en) Based on pedestrian recognition method under the road traffic environment for improving YOLOv3
Pereira et al. A deep learning-based approach for road pothole detection in timor leste
CN104700099B (en) The method and apparatus for recognizing traffic sign
CN107563372B (en) License plate positioning method based on deep learning SSD frame
CN108898047B (en) Pedestrian detection method and system based on blocking and shielding perception
CN110796168A (en) Improved YOLOv 3-based vehicle detection method
CN107633226B (en) Human body motion tracking feature processing method
CN108960198A (en) A kind of road traffic sign detection and recognition methods based on residual error SSD model
CN107679531A (en) Licence plate recognition method, device, equipment and storage medium based on deep learning
CN103699905B (en) Method and device for positioning license plate
CN105931269A (en) Tracking method for target in video and tracking device thereof
CN105787482A (en) Specific target outline image segmentation method based on depth convolution neural network
Gong et al. Object detection based on improved YOLOv3-tiny
CN110569738A (en) natural scene text detection method, equipment and medium based on dense connection network
CN109087337B (en) Long-time target tracking method and system based on hierarchical convolution characteristics
CN109886147A (en) A kind of more attribute detection methods of vehicle based on the study of single network multiple-task
CN110378239A (en) A kind of real-time traffic marker detection method based on deep learning
CN110245587B (en) Optical remote sensing image target detection method based on Bayesian transfer learning
CN112417931B (en) Method for detecting and classifying water surface objects based on visual saliency
CN109784155B (en) Visual target tracking method based on verification and error correction mechanism and intelligent robot
CN106845458B (en) Rapid traffic sign detection method based on nuclear overrun learning machine
Sun et al. Adaptive saliency biased loss for object detection in aerial images
CN110008900A (en) A kind of visible remote sensing image candidate target extracting method by region to target
CN115457277A (en) Intelligent pavement disease identification and detection method and system
Adiwinata et al. Fish species recognition with faster r-cnn inception-v2 using qut fish dataset

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20190212

RJ01 Rejection of invention patent application after publication