CN107301376B - Pedestrian detection method based on deep learning multi-layer stimulation - Google Patents

Pedestrian detection method based on deep learning multi-layer stimulation Download PDF

Info

Publication number
CN107301376B
CN107301376B CN201710385952.3A CN201710385952A CN107301376B CN 107301376 B CN107301376 B CN 107301376B CN 201710385952 A CN201710385952 A CN 201710385952A CN 107301376 B CN107301376 B CN 107301376B
Authority
CN
China
Prior art keywords
pedestrian
frame
candidate
target
neural 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.)
Active
Application number
CN201710385952.3A
Other languages
Chinese (zh)
Other versions
CN107301376A (en
Inventor
李玺
李健
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 University ZJU filed Critical Zhejiang University ZJU
Priority to CN201710385952.3A priority Critical patent/CN107301376B/en
Publication of CN107301376A publication Critical patent/CN107301376A/en
Application granted granted Critical
Publication of CN107301376B publication Critical patent/CN107301376B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • 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
    • 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
    • G06V40/20—Movements or behaviour, e.g. gesture recognition

Landscapes

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

Abstract

The invention discloses a pedestrian detection method based on deep learning multilayer stimulation, which is used for marking the position of a target appearing in a video after video monitoring and the target needing to be detected are given. The method specifically comprises the following steps: acquiring a pedestrian data set used for training a target detection model, and defining an algorithm target; modeling the position deviation and the apparent semantics of the pedestrian target; establishing a pedestrian multilayer stimulation network model according to the modeling result in the step S2; the pedestrian position in the monitoring image is detected using the detection model. The pedestrian detection method is suitable for pedestrian detection in real video monitoring images, and has better effect and robustness in the face of various complex conditions.

Description

Pedestrian detection method based on deep learning multi-layer stimulation
Technical Field
The invention belongs to the field of computer vision, and particularly relates to a pedestrian detection method based on deep learning multi-layer stimulation.
Background
With the development of computer vision since the end of the 20 th century, intelligent video processing technology has gained widespread attention and research. Pedestrian detection is an important and challenging task, with the goal of accurately detecting the location of pedestrians in video surveillance images. The problem has high application value in the fields of video monitoring, intelligent robots and the like, and is the basis of a large number of high-level visual tasks. However, the problem is also more challenging, namely how to express the target region information; secondly, how to uniformly model and optimize the extraction of the candidate region and the target classification, and the challenges put higher requirements on the performance and the robustness of the corresponding algorithm.
The general pedestrian detection algorithm is divided into three parts: 1. candidate regions containing the target in the input image are found. 2. And manually extracting target features based on the candidate regions. 3. And (4) realizing a detection task by using a classification algorithm on the features. The method mainly has the following problems: 1) the pedestrian detection method is based on the traditional visual features, the visual features can only express visual information of a lower layer, but a pedestrian detection task needs a model with high-level abstract semantic understanding capability; 2) the extraction of candidate regions and the classification of features are not optimized by end-to-end learning; 3) the features extracted based on deep learning are not subjected to multi-layer stimulation combination, and the target features are not abstract enough.
Disclosure of Invention
To solve the above problems, it is an object of the present invention to provide a pedestrian detection method based on deep learning multi-layer stimuli for detecting the pedestrian position in a given monitoring image. The method is based on a deep neural network, utilizes the depth visual characteristics of multi-layer stimulation to represent the information of a target area, adopts the Faster R-CNN framework to model pedestrian detection, and can better adapt to the complex situation in a real video monitoring scene.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a pedestrian detection method based on deep learning multi-layer stimulation comprises the following steps:
s1, acquiring a pedestrian data set for training a target detection model, and defining an algorithm target;
s2, modeling the position deviation and the apparent semantic meaning of the pedestrian target;
s3, establishing a pedestrian multilayer stimulation network model according to the modeling result in the step S2;
and S4, detecting the pedestrian position in the monitored image by using the detection model.
Further, in step S1, the pedestrian data set for training the target detection model includes a pedestrian image XtrainA pedestrian position B marked manually;
the algorithm targets are defined as: the pedestrian position P in one monitor image X is detected.
Further, in step S2, the modeling of the position deviation and the apparent semantic meaning of the pedestrian object specifically includes:
s21, according to the pedestrian data set XtrainAnd pedestrian position P modeling position deviation:
Figure BDA0001306371040000021
Figure BDA0001306371040000022
wherein, x, y are the coordinate of the middle point of the pedestrian frame label, w, h are the width and the length of the pedestrian frame label, and xa,yaIs the coordinate of the pedestrian candidate frame, wa,haIs the width and length of the pedestrian candidate frame; t is txAs the ratio of the deviation of the x coordinate of the pedestrian frame relative to the x coordinate of the marking frame to the width of the marking frame, tyAs the proportion of the deviation of the y coordinate of the pedestrian frame relative to the y coordinate of the marking frame corresponding to the length of the marking frame, twAs the ratio of the width of the pedestrian frame to the width of the marking frame, thThe length of the pedestrian frame is in proportion to the length of the marking frame;
s22, according to the pedestrian data set XtrainAnd pedestrian position P modeling appearance semantics:
s＝<w,d>
Figure BDA0001306371040000023
where s represents the projection value of the feature d onto a projection vector w, w is the pedestrian weight projection vector, d is the pedestrian feature descriptor,<.,.>is the inner product operator, p (C ═ k | d) is the softmax function, indicating the probability values belonging to class k; sjIs the projection value of the feature d on the jth projection vector w; c is a discrete random variable with the value number of k; j is the index of the jth w of the total projection vectors w.
Further, in step S3, the step of establishing the pedestrian multi-layer stimulation network model according to the modeling result in step S2 specifically includes:
s31, establishing a multilayer stimulation convolutional neural network, wherein the input of the neural network is a monitoring image X and a pedestrian marking box B, and the output is a probability value p of a corresponding pedestrian candidate box and a pedestrian position deviation O in the X; the structure of the neural network is represented as mapping X → (p, O);
s32, child mapping X → p uses the soft maximum Softmax loss function, expressed as
Figure BDA0001306371040000031
Lcls(X,Y；θ)＝-∑jYjlogp (C | d) formula (3)
Wherein Y is a binary vector, if the k-th class belongs to, the corresponding value is 1, and the rest is 0; l iscls(X, Y; θ) represents the softmax loss function of the entire training data set;
s33, child mapping X → O Using Euclidean loss function, expressed as
Lloc(t,v)＝∑ismooth(ti,vi)
Figure BDA0001306371040000032
Wherein t isiIs a pedestrian position deviation tag, viIs a pedestrian position deviation predicted value; i represents the ith training sample;
s34 loss function of the whole multi-layer stimulation neural network
L＝Lcls+LlocFormula (5)
The entire neural network is trained using a stochastic gradient descent and back propagation algorithm under a loss function L.
Further, in step S4, the detecting the pedestrian position in the monitoring image includes: inputting the monitoring image X to be detected into the trained neural network, judging whether the image X is a pedestrian according to the output candidate box probability value, and finally correcting according to the predicted position deviation O to obtain the pedestrian position P.
Compared with the existing pedestrian detection method, the pedestrian detection method applied to the video monitoring scene has the following beneficial effects:
firstly, the pedestrian detection method of the invention builds a model based on a deep convolutional neural network. The invention unifies the generation of the candidate region and the classification of the characteristics in the same network frame for learning and optimization, thereby improving the final effect of the method.
Secondly, the multi-layer stimulation algorithm provided by the invention can enrich the feature abstract capability, and meanwhile, the features learned by the algorithm enable the classifier to learn more robust classification rules.
The pedestrian detection method applied to the video monitoring scene has good application value in an intelligent video analysis system, and can effectively improve the efficiency and accuracy of pedestrian detection. For example, in traffic video monitoring, the pedestrian detection method can quickly and accurately detect the positions of all pedestrians, provide data for subsequent pedestrian search tasks, and greatly release human resources.
Drawings
FIG. 1 is a schematic flow chart of a pedestrian detection method applied to a video surveillance scene according to the present invention;
FIG. 2 is a schematic diagram of the loss function of the whole multi-layer neural network according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
On the contrary, the invention is intended to cover alternatives, modifications, equivalents and alternatives which may be included within the spirit and scope of the invention as defined by the appended claims. Furthermore, in the following detailed description of the present invention, certain specific details are set forth in order to provide a better understanding of the present invention. It will be apparent to one skilled in the art that the present invention may be practiced without these specific details.
Referring to fig. 1, in a preferred embodiment of the present invention, a pedestrian detection method based on deep learning multi-layer stimulation comprises the following steps:
first, a pedestrian data set including a pedestrian image X for training a target detection model is acquiredtrainA pedestrian position B marked manually;
the algorithm targets are defined as: the pedestrian position P in one monitor image X is detected.
Secondly, modeling the position deviation and the apparent semantics of the pedestrian target specifically comprises:
first, from a pedestrian data set XtrainAnd pedestrian position P modeling position deviation:
Figure BDA0001306371040000041
Figure BDA0001306371040000042
wherein, x, y are the coordinate of the middle point of the pedestrian frame label, w, h are the width and the length of the pedestrian frame label, and xa,yaIs the coordinate of the pedestrian candidate frame, wa,haIs the width and length of the pedestrian candidate frame; t is txAs the ratio of the deviation of the x coordinate of the pedestrian frame relative to the x coordinate of the marking frame to the width of the marking frame, tyAs the proportion of the deviation of the y coordinate of the pedestrian frame relative to the y coordinate of the marking frame corresponding to the length of the marking frame, twAs the ratio of the width of the pedestrian frame to the width of the marking frame, thThe length of the pedestrian frame is in proportion to the length of the marking frame;
second, from the pedestrian data set XtrainAnd pedestrian position P modeling appearance semantics:
s＝<w,d>
Figure BDA0001306371040000051
where s represents the projection value of the feature d onto a projection vector w, w is the pedestrian weight projection vector, d is the pedestrian feature descriptor,<.,.>is the inner product operator, p (C ═ k | d) is the softmax function, indicating the probability values belonging to class k; sjIs the projection value of the feature d on the jth projection vector w; c is a discrete random variable with the value number of k; j is the index of the jth w of the total projection vectors w.
And then, pre-training a detection model of the billboard target according to the complaint modeling result. The method specifically comprises the following steps:
firstly, establishing a multilayer stimulation convolutional neural network, wherein the input of the neural network is a monitoring image X and a pedestrian marking frame B, and the output is a probability value p of a corresponding pedestrian candidate frame and a pedestrian position deviation O in the X; thus, the structure of the neural network can be represented as the mapping X → (p, O);
second, the sub-map X → p uses a soft maximum (Softmax) loss function, denoted as
Figure BDA0001306371040000052
Lcls(X,Y；θ)＝-∑jYjlogp (C | d) formula (3)
Wherein Y is a binary vector, if the k-th class belongs to, the corresponding value is 1, and the rest is 0; l iscls(X, Y; θ) represents the softmax loss function of the entire training data set;
third, the sub-map X → O uses the Euclidean loss function, expressed as
Lloc(t,v)＝∑ismooth(ti,vi)
Figure BDA0001306371040000053
Wherein t isiIs a pedestrian position deviation tag, viIs the predicted value of the pedestrian position deviation, and i represents the ith training sample.
Fourth, referring to FIG. 2, the loss function of the entire multi-layer neural network is
L＝Lcls+LlocFormula (5)
The entire neural network is trained using a stochastic gradient descent and back propagation algorithm under a loss function L.
And finally, detecting the pedestrians in the monitoring image by using the trained detection model. The method specifically comprises the following steps: and (4) placing the preprocessed image on a multi-layer stimulation detection framework for calculation. The multi-layer stimulation detection framework extracts candidate frames by using 3 RPN networks, and the feature information utilized by each RPN network is different, so that the sizes and the scales of the obtained candidate frames are different. And firstly obtaining candidate frames extracted by each RPN network, and filtering according to the respective confidence degrees to obtain 300 candidate regions. Then, the candidate regions in the 3 RPN networks are merged to obtain 900 candidate regions. And then, according to the arrangement of the classification confidence degrees from large to small, filtering to obtain the final 300 target candidate regions. And filtering the candidate frames according to whether the output candidate frame classification probability value is greater than a given threshold value or not, eliminating the crossed and repeated detection frames by adopting a non-maximum value inhibition algorithm, and finally correcting according to the predicted position deviation O to obtain the position P of the pedestrian.
In the above embodiment, the pedestrian detection method of the invention first models the position deviation and the apparent semantic meaning of the pedestrian target. On the basis, the original problem is converted into a multi-task learning problem, and a pedestrian detection model is established based on the deep neural network. And finally, detecting the position of the pedestrian in the monitoring image by using the trained detection model.
Through the technical scheme, the embodiment of the invention develops a pedestrian detection algorithm based on deep learning multi-layer stimulation based on the deep learning technology. The invention can effectively model the position deviation and the apparent semantic information of the target at the same time, thereby detecting the accurate pedestrian position.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (1)

1. A pedestrian detection method based on deep learning multi-layer stimulation is characterized by comprising the following steps:
s1, acquiring a pedestrian data set for training a target detection model, and defining an algorithm target; the pedestrian data set for training the target detection model comprises a pedestrian image XtrainA pedestrian position B marked manually; the algorithm targets are defined as: detecting a pedestrian position P in a monitoring image X;
s2, modeling the position deviation and the apparent semantics of the pedestrian target, specifically comprising:
s21, according to the pedestrian data set XtrainAnd pedestrian position P modeling position deviation:
Figure FDA0002946034640000011
Figure FDA0002946034640000012
wherein, x, y are the coordinate of the middle point of the pedestrian frame label, w, h are the width and the length of the pedestrian frame label, and xa,yaIs the coordinate of the pedestrian candidate frame, wa,haIs the width and length of the pedestrian candidate frame; t is txAs the ratio of the deviation of the x coordinate of the pedestrian frame relative to the x coordinate of the marking frame to the width of the marking frame, tyAs the proportion of the deviation of the y coordinate of the pedestrian frame relative to the y coordinate of the marking frame corresponding to the length of the marking frame, twAs the ratio of the width of the pedestrian frame to the width of the marking frame, thThe length of the pedestrian frame is in proportion to the length of the marking frame;
s22, according to the pedestrian data set XtrainAnd pedestrian position P modeling appearance semantics:
s＝<w，d>
Figure FDA0002946034640000013
where s represents the projection value of the feature d onto a projection vector w, w is the pedestrian weight projection vector, d is the pedestrian feature descriptor,<.，.>is the inner product operator, p (C ═ k | d) is the softmax function, indicating the probability values belonging to class k; sjIs the projection value of the feature d on the jth projection vector w; c is a discrete random variable with the value number of k; j is the index of the jth w of all projection vectors w;
s3, establishing a pedestrian multilayer stimulation network model according to the modeling result in the step S2, which specifically comprises the following steps:
s31, establishing a multilayer stimulation convolutional neural network, wherein the input of the neural network is a monitoring image X and a pedestrian marking box B, and the output is a probability value p of a corresponding pedestrian candidate box and a pedestrian position deviation O in the X; the structure of the neural network is represented as mapping X → (p, O);
s32, child mapping X → p uses the soft maximum Softmax loss function, expressed as
Figure FDA0002946034640000021
Lcls(X，Y；θ)＝-∑jYjLog p (C | d) formula (3)
Wherein Y is a binary vector, if the k-th class belongs to, the corresponding value is 1, and the rest is 0; l iscls(X, Y; θ) represents the softmax loss function of the entire training data set;
s33, child mapping X → O Using Euclidean loss function, expressed as
Lloc(t，v)＝∑ismooth(ti，vi)
Figure FDA0002946034640000022
Wherein t isiIs a pedestrian position deviation tag, viIs a pedestrian position deviation predicted value; i represents the ith training sample;
s34 loss function of the whole multi-layer stimulation neural network
L＝Lcls+LlocFormula (5)
Training the whole neural network under a loss function L by using a random gradient descent and back propagation algorithm;
the multilayer stimulation neural network extracts candidate frames by using 3 RPN networks, the characteristic information utilized by each RPN network is different, so that the sizes and the scales of the obtained candidate frames are different, and each RPN network introduces a loss function L; in the detection process, obtaining candidate frames extracted by each RPN network, and filtering according to the respective confidence degrees to obtain 300 candidate regions; then combining the candidate regions in the 3 RPN networks to obtain 900 candidate regions; then, arranging according to the classification confidence degree from large to small, and filtering to obtain the final 300 target candidate regions; filtering the candidate frames according to whether the output candidate frame classification probability value is larger than a given threshold value or not, eliminating the detection frames with repeated crossing by adopting a non-maximum value inhibition algorithm, and finally correcting according to the predicted position deviation O to obtain the position P of the pedestrian;
s4, detecting the pedestrian position in the monitored image by using the detection model; wherein detecting the pedestrian position in the monitored image comprises: inputting the monitoring image X to be detected into the trained neural network, judging whether the image X is a pedestrian according to the output candidate box probability value, and finally correcting according to the predicted position deviation O to obtain the pedestrian position P.
CN201710385952.3A 2017-05-26 2017-05-26 Pedestrian detection method based on deep learning multi-layer stimulation Active CN107301376B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710385952.3A CN107301376B (en) 2017-05-26 2017-05-26 Pedestrian detection method based on deep learning multi-layer stimulation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710385952.3A CN107301376B (en) 2017-05-26 2017-05-26 Pedestrian detection method based on deep learning multi-layer stimulation

Publications (2)

Publication Number Publication Date
CN107301376A CN107301376A (en) 2017-10-27
CN107301376B true CN107301376B (en) 2021-04-13

Family

ID=60138099

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710385952.3A Active CN107301376B (en) 2017-05-26 2017-05-26 Pedestrian detection method based on deep learning multi-layer stimulation

Country Status (1)

Country Link
CN (1) CN107301376B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110163224B (en) * 2018-01-23 2023-06-20 天津大学 Auxiliary data labeling method capable of online learning
CN108537117B (en) * 2018-03-06 2022-03-11 哈尔滨思派科技有限公司 Passenger detection method and system based on deep learning
CN108446662A (en) * 2018-04-02 2018-08-24 电子科技大学 A kind of pedestrian detection method based on semantic segmentation information
CN110969657B (en) * 2018-09-29 2023-11-03 杭州海康威视数字技术股份有限公司 Gun ball coordinate association method and device, electronic equipment and storage medium
CN111178267A (en) * 2019-12-30 2020-05-19 成都数之联科技有限公司 Video behavior identification method for monitoring illegal fishing
CN111476089B (en) * 2020-03-04 2023-06-23 上海交通大学 Pedestrian detection method, system and terminal for multi-mode information fusion in image
CN111523478B (en) * 2020-04-24 2023-04-28 中山大学 Pedestrian image detection method acting on target detection system

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106022237A (en) * 2016-05-13 2016-10-12 电子科技大学 Pedestrian detection method based on end-to-end convolutional neural network
CN106250812A (en) * 2016-07-15 2016-12-21 汤平 A kind of model recognizing method based on quick R CNN deep neural network

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107430677B (en) * 2015-03-20 2022-04-12 英特尔公司 Target identification based on improving binary convolution neural network characteristics
US10382300B2 (en) * 2015-10-06 2019-08-13 Evolv Technologies, Inc. Platform for gathering real-time analysis

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106022237A (en) * 2016-05-13 2016-10-12 电子科技大学 Pedestrian detection method based on end-to-end convolutional neural network
CN106250812A (en) * 2016-07-15 2016-12-21 汤平 A kind of model recognizing method based on quick R CNN deep neural network

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection;Zhaowei Cai et al.;《European Conference on Computer Vision》;20160917;第354-358页 *
Deep Convolutional Neural Networks for Pedestrian Detection with Skip Pooling;Jie Liu et al.;《 2017 International Joint Conference on Neural Networks》;20170519;第1-9页 *
Fast R-CNN;Ross Girshick;《arXiv:1504.08083v2》;20150927;第2056-2063页 *
R-FCN: Object Detection via Region-based Fully Convolutional Networks;Jifeng Dai et al.;《arXiv:1605.06409v2》;20160621;第4页 *
基于特征共享的高效物体检测;任少卿;《中国博士学位论文全文数据库 信息科技辑》;20160815;第2016年卷(第8期);第四章 *

Also Published As

Publication number Publication date
CN107301376A (en) 2017-10-27

Similar Documents

Publication Publication Date Title
CN107301376B (en) Pedestrian detection method based on deep learning multi-layer stimulation
Zhou et al. Safety helmet detection based on YOLOv5
Wang et al. Actionness estimation using hybrid fully convolutional networks
CN106845487B (en) End-to-end license plate identification method
CN104050471B (en) Natural scene character detection method and system
CN107016357B (en) Video pedestrian detection method based on time domain convolutional neural network
CN111611874B (en) Face mask wearing detection method based on ResNet and Canny
CN111191667B (en) Crowd counting method based on multiscale generation countermeasure network
Deng et al. Amae: Adaptive motion-agnostic encoder for event-based object classification
Li et al. Sign language recognition based on computer vision
CN110298297A (en) Flame identification method and device
CN106815563B (en) Human body apparent structure-based crowd quantity prediction method
CN108648211A (en) A kind of small target detecting method, device, equipment and medium based on deep learning
CN103984955B (en) Multi-camera object identification method based on salience features and migration incremental learning
CN112801019B (en) Method and system for eliminating re-identification deviation of unsupervised vehicle based on synthetic data
CN108898623A (en) Method for tracking target and equipment
CN107330363B (en) Rapid internet billboard detection method
CN103500456B (en) A kind of method for tracing object based on dynamic Bayesian network network and equipment
Liu et al. D-CenterNet: An anchor-free detector with knowledge distillation for industrial defect detection
CN110111365B (en) Training method and device based on deep learning and target tracking method and device
CN108009512A (en) A kind of recognition methods again of the personage based on convolutional neural networks feature learning
Yin Object Detection Based on Deep Learning: A Brief Review
CN114170686A (en) Elbow bending behavior detection method based on human body key points
Pang et al. Dance video motion recognition based on computer vision and image processing
CN105118073A (en) Human body head target identification method based on Xtion camera

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