CN112989944A - Intelligent video safety supervision method based on federal learning - Google Patents
Intelligent video safety supervision method based on federal learning Download PDFInfo
- Publication number
- CN112989944A CN112989944A CN202110182984.XA CN202110182984A CN112989944A CN 112989944 A CN112989944 A CN 112989944A CN 202110182984 A CN202110182984 A CN 202110182984A CN 112989944 A CN112989944 A CN 112989944A
- Authority
- CN
- China
- Prior art keywords
- model
- training
- data
- federal learning
- enterprise
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a federal learning-based intelligent video safety supervision method, which aims to solve the problems that in the prior art, the intelligent video supervision method based on deep learning increases the risk of revealing user privacy, increases the pressure of a calculation center and consumes a large amount of bandwidth resources. The method adopts a decentralized federal learning mode to train a model, and obtains a final federal learning trained model mainly through the steps of center server global model pre-training and issuing, enterprise node local model training and center server global model iterative updating, wherein the model precision is close to that of a model obtained by a data centralized training mode, and the data privacy of each enterprise node can be effectively protected; and a load balancing push strategy is adopted, and computing resources are reasonably allocated according to the load capacity of the model, so that the pressure of a computing center is reduced, and the utilization rate of the computing resources is improved.
Description
Technical Field
The invention relates to the field of safety production, in particular to a video intelligent safety supervision method based on federal learning.
Background
With the continuous expansion of the internal security supervision range of an enterprise and the continuous refinement of supervision, the exposed potential safety hazards are more and more, so how to take effective measures to carry out security supervision is a problem to be solved urgently.
The traditional safety supervision method is to find hidden dangers and give early warning in a mode of manual inspection or video playback, and due to the fact that supervision points are wide and video data volume is large, the problems that safety abnormity is found out late and rectification is not timely are prone to occurring. With the rise of artificial intelligence, the intelligent video supervision method based on deep learning can effectively solve the problems, but the deep learning is data-driven, and a large amount of illegal materials are needed to be used as training samples, so that a large amount of video data needs to be obtained from each enterprise. This approach has two problems: on one hand, the video data may relate to enterprise privacy and business confidentiality, and uploading the video data directly to the cloud computing center increases the risk of revealing user privacy; on the other hand, when the training sample data is large, the pressure of a calculation center is increased, and a large amount of bandwidth resources are consumed.
Disclosure of Invention
The invention aims to solve the problems that the intelligent video supervision method based on deep learning in the prior art increases the risk of revealing user privacy, increases the pressure of a calculation center and consumes a large amount of bandwidth resources, and provides the intelligent video safety supervision method based on federal learning.
The technical scheme adopted by the invention is as follows:
a video intelligent safety supervision method based on federal learning is characterized by comprising the following steps:
1) central server global model pre-training and issuing
1.1) data preprocessing
Collecting data in similar non-confidential scenes, or crawling relevant image data from a network by adopting a crawler technology; manually screening and marking data, and clearing invalid data to obtain a data set; expanding the data set by adopting a data augmentation mode;
1.2) model Pre-training
Based on a pre-training model of the public data set, training a basic model at a central server by using the data set prepared in the step 1.1);
1.3) model distribution
The central server side issues the model to each enterprise node, and initializes model hyper-parameters C, B, E;
wherein C is a node proportion C which participates in the federal learning and belongs to [0,1 ];
b is the batch _ size of each node during training, and B is int type;
e is an epoch trained by each node, and E is an int type;
2) enterprise node local model training
2.1) data preprocessing
Each enterprise collects data in each scene and completes data preprocessing;
2.2) model training
Selecting int (C x 5) enterprises at random, training and fine-tuning the model by each enterprise node based on the model weight of the central service end, and uploading the optimal model of the enterprise node to the central service end after E epochs are iterated;
3) central server global model iterative update
3.1) model polymerization
The central server side adopts a federal average algorithm to aggregate the uploaded enterprise node models, and takes the proportion of the number of samples of each enterprise node participating in training to the total number of samples as a weight to carry out weighted average on the gradient parameters of the models;
3.2) model update
The gradient parameters after weighted averaging are reversely propagated on the central server model, and the central server model weight is updated; judging whether the maximum updating round number or the expected model precision is reached, if not, issuing the updated model to each enterprise node, and returning to the step 2); if so, finishing updating to obtain a final federal learning training model, and issuing the model to each enterprise node;
4) model deployment
4.1) video stream acquisition and parsing
Acquiring a video stream from a camera in an actual production scene of an enterprise, and analyzing the video stream into an image in real time;
4.2) load-balancing push
When the number of the images to be inferred is larger than the load capacity of the single-path model, the model is loaded for multiple times to provide multiple inference services, and the enterprise nodes adopt a load balancing strategy to push the images to be inferred into the corresponding models;
4.3) reasoning
According to enterprise requirements, the enterprise node model detects and identifies an image to be inferred;
4.4) early warning and display
Reporting the violation data to a platform according to the detection result of the enterprise node model; the platform carries out real-time early warning and display and stores, counts and manages violation data.
Further, in step 3.2), the update center server model weight is as follows:
in the formula, wt+1The central server model weight after t round updating;
wtthe model weight of the central service end before the t round of updating is obtained;
eta is the learning rate;
k is the number of nodes;
m is the number of nodes participating in federal learning;
nkthe number of samples contained for node k;
n is the total number of samples contained in all nodes;
t is the number of model updating rounds;
Further, in step 3.2), after the model of the final federal learning training is issued to each enterprise node, the model is packaged into a web application by using a flash and is made into a docker mirror image.
Further, in step 1.1), the data augmentation mode is Cutout, Mixup, CutMix, color space transformation or image flipping.
Further, in step 1.2), the public data set is a COCO data set.
Further, in step 4.1), the analyzing the video stream into an image in real time specifically includes: and analyzing the video stream of the H264 into an RGB three-channel image sequence in real time by adopting an FFmpeg or camera SDK de-streaming mode.
Further, in step 1.3), the model hyper-parameter C is 0.5, B is 64, and E is 20.
Further, in the steps 1) to 4), the model is a face recognition model, a dressing and wearing detection model, an abnormal behavior detection model or an instrument recognition model.
Compared with the prior art, the invention has the beneficial effects that:
(1) according to the intelligent video safety supervision method based on the federal learning, the decentralized federal learning mode is adopted to train the model, the model is updated and iterated conveniently, the finally obtained model precision is close to that obtained by a data centralized training mode, the data privacy of each enterprise node is effectively protected, and the data barrier in deep learning application is broken; by adopting a load balancing push strategy, computing resources are reasonably allocated according to the load capacity of the model, the pressure of a computing center is reduced, and the utilization rate of the computing resources is improved; and finally, real-time early warning and display are carried out according to the detection result, so that the probability of enterprise safety accidents is greatly reduced.
(2) According to the invention, a docker packing model is adopted, and the containerization deployment mode of the model can achieve the effect of being used after opening the box, so that the calculation and the application are decoupled, the stability of the platform is improved, and the later-stage model is convenient to upgrade and optimize.
Drawings
FIG. 1 is a schematic diagram of Federal learning-based model training in the present invention;
FIG. 2 is a schematic diagram of model deployment in the present invention;
fig. 3 is a diagram of an application system architecture for the method of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
Taking a plurality of similar processing and manufacturing enterprises as an example, a plurality of cameras are installed in production workshops of each enterprise, and video data in the enterprise cannot be leaked out, based on the scene, the federal learning-based video intelligent safety supervision method provided by the invention is adopted to train and deploy a model by taking wearing, wearing and detecting as an object, and as shown in fig. 1 and fig. 2, the specific implementation steps are as follows:
1) central server global model pre-training and issuing
1.1) data preprocessing
Collecting data in similar non-confidential scenes, or crawling relevant image data from a network by adopting a crawler technology; manually screening and labeling the data, and clearing invalid data according to the data labels and the image data to ensure the validity of training samples so as to obtain a data set; expanding a data set by adopting a data augmentation mode, such as Cutout, Mixup, CutMix, color space transformation, image inversion and other strategies, and improving the balance of different classes and the generalization capability of models in training samples;
1.2) model Pre-training
Dressing and wearing detection mainly adopts a target model, a pre-training model of an open data set is used as a basis, such as a COCO data set, and a basic model is trained at a central server by using a prepared data set;
1.3) model distribution
The central server side issues the model structure and the model parameters to each enterprise node, and as the data of each enterprise node is isolated and distributed unevenly, the federal learning method introduces three hyper-parameters C, B, E and initializes the model hyper-parameters (which can be configured according to specific conditions): c is 0.5 (the node proportion participating in federal learning, C belongs to [0,1]), and the parameter can effectively improve the training speed and increase the diversity of training samples; b64 (batch _ size at each node training, B int type); e-20 (epoch trained by each node, E is int type);
2) enterprise node local model training
2.1) data preprocessing
Each enterprise collects data in each scene and completes data preprocessing;
2.2) model training
Selecting int (C x 5) enterprises at random, training and fine-tuning the models by each enterprise node based on the model weight of the central service end, storing one model in each training round, and uploading the optimal model in the models stored in the E-round training to the central service end after E epochs are iterated;
3) central server global model iterative update
3.1) model polymerization
The central server side adopts a federal average algorithm to aggregate the uploaded enterprise node models, and takes the proportion of the number of samples of each enterprise node participating in training to the total number of samples as a weight to carry out weighted average on the gradient parameters of the models;
3.2) model update
The gradient parameters after weighted averaging are reversely propagated on the central server model, and the central server model weight is updated;
assuming that there are K nodes in total, and M nodes participate in federal learning in the t round global model updating process, the weight updating is as follows:
in the formula, wt+1The central server model weight after t round updating;
wtthe model weight of the central service end before the t round of updating is obtained;
eta is the learning rate;
k is the number of nodes;
m is the number of nodes participating in federal learning;
nkthe number of samples contained for node k;
n is the total number of samples contained in all nodes;
t is the number of model updating rounds;
after the central server side model weight is updated, judging whether the maximum updating round number or the expected model precision is reached, if not, issuing the updated model to each enterprise node, and returning to the step 2); if so, finishing updating to obtain a final federal learning training model, and issuing the model to each enterprise node; each enterprise node packages the final federal learning and training model into web application by adopting flash, and in order to simplify model deployment, the model is made into a mirror image by adopting a docker container technology;
4) model deployment
4.1) video stream acquisition and parsing
Acquiring a video stream from a camera in an actual production scene of an enterprise, and analyzing the video stream of H264 into an RGB three-channel image sequence in real time by adopting an FFmpeg or camera SDK (software development kit) de-streaming mode;
4.2) load-balancing push
When the number of images to be inferred in an enterprise is larger than the load capacity of a single-path model, the model is loaded for multiple times to provide multiple inference services, and enterprise nodes adopt a load balancing strategy to push the images to be inferred into corresponding models, so that the computing resources of a server are utilized to the maximum extent;
4.3) reasoning
According to enterprise requirements, the enterprise node model detects and identifies an image to be inferred;
4.4) early warning and display
Reporting the violation data to a platform according to the detection result of the enterprise node model; the platform carries out real-time early warning and display and stores, counts and manages violation data.
As shown in fig. 3, in the data acquisition layer, the application system of the method of the present invention acquires data through a network camera, an industrial camera, a local video, a sensor, etc.; on a data processing layer, after obtaining a deep learning model through weighted federal training, each enterprise compresses the model by adopting a quantization and pruning method, reduces model parameters and improves the reasoning speed of the model by combining a high-performance deep learning reasoning optimizer (TensorRT or TVM), packages the optimized deep learning model into web application, and returns a detection result after a client pushes a frame of image; and the application layer quantizes the model detection result into violation data according to enterprise management regulations, and realizes the functions of violation data statistics, violation early warning, violation data display and the like.
Claims (8)
1. A video intelligent safety supervision method based on federal learning is characterized by comprising the following steps:
1) central server global model pre-training and issuing
1.1) data preprocessing
Collecting data in similar non-confidential scenes, or crawling relevant image data from a network by adopting a crawler technology; manually screening and marking data, and clearing invalid data to obtain a data set; expanding the data set by adopting a data augmentation mode;
1.2) model Pre-training
Based on a pre-training model of the public data set, training a basic model at a central server by using the data set prepared in the step 1.1);
1.3) model distribution
The central server side issues the model to each enterprise node, and initializes model hyper-parameters C, B, E;
wherein C is a node proportion C which participates in the federal learning and belongs to [0,1 ];
b is the batch _ size of each node during training, and B is int type;
e is an epoch trained by each node, and E is an int type;
2) enterprise node local model training
2.1) data preprocessing
Each enterprise collects data in each scene and completes data preprocessing;
2.2) model training
Selecting int (C x 5) enterprises at random, training and fine-tuning the model by each enterprise node based on the model weight of the central service end, and uploading the optimal model of the enterprise node to the central service end after E epochs are iterated;
3) central server global model iterative update
3.1) model polymerization
The central server side adopts a federal average algorithm to aggregate the uploaded enterprise node models, and takes the proportion of the number of samples of each enterprise node participating in training to the total number of samples as a weight to carry out weighted average on the gradient parameters of the models;
3.2) model update
The gradient parameters after weighted averaging are reversely propagated on the central server model, and the central server model weight is updated; judging whether the maximum updating round number or the expected model precision is reached, if not, issuing the updated model to each enterprise node, and returning to the step 2); if so, finishing updating to obtain a final federal learning training model, and issuing the model to each enterprise node;
4) model deployment
4.1) video stream acquisition and parsing
Acquiring a video stream from a camera in an actual production scene of an enterprise, and analyzing the video stream into an image in real time;
4.2) load-balancing push
When the number of the images to be inferred is larger than the load capacity of the single-path model, the model is loaded for multiple times to provide multiple inference services, and the enterprise nodes adopt a load balancing strategy to push the images to be inferred into the corresponding models;
4.3) reasoning
According to enterprise requirements, the enterprise node model detects and identifies an image to be inferred;
4.4) early warning and display
Reporting the violation data to a platform according to the detection result of the enterprise node model; the platform carries out real-time early warning and display and stores, counts and manages violation data.
2. The federal learning-based video intelligent security supervision method according to claim 1, wherein:
in step 3.2), the weight of the update center server model is shown as follows:
in the formula, wt+1The central server model weight after t round updating;
wtthe model weight of the central service end before the t round of updating is obtained;
eta is the learning rate;
k is the number of nodes;
m is the number of nodes participating in federal learning;
nkthe number of samples contained for node k;
n is the total number of samples contained in all nodes;
t is the number of model updating rounds;
3. The federal learning-based video intelligent security supervision method according to claim 2, wherein:
and 3.2), after the final model of the federal learning training is issued to each enterprise node, the model is packaged into web application by adopting flash and is made into a docker mirror image.
4. The intelligent video safety supervision method based on federal learning according to any one of claims 1 to 3, characterized in that:
in step 1.1), the data augmentation mode is Cutout, Mixup, CutMix, color space transformation or image inversion.
5. The federal learning-based video intelligent security supervision method according to claim 4, wherein:
in step 1.2), the public data set is a COCO data set.
6. The federal learning-based video intelligent security supervision method according to claim 5, wherein:
in step 4.1), the analyzing the video stream into the image in real time specifically includes: and analyzing the video stream of the H264 into an RGB three-channel image sequence in real time by adopting an FFmpeg or camera SDK de-streaming mode.
7. The federal learning-based video intelligent security supervision method according to claim 6, wherein:
in step 1.3), the model hyperparameter C is 0.5, B is 64, and E is 20.
8. The federal learning-based video intelligent security supervision method according to claim 7, wherein:
in the steps 1) to 4), the model is a face recognition model, a dressing and wearing detection model, an abnormal behavior detection model or an instrument recognition model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110182984.XA CN112989944A (en) | 2021-02-08 | 2021-02-08 | Intelligent video safety supervision method based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110182984.XA CN112989944A (en) | 2021-02-08 | 2021-02-08 | Intelligent video safety supervision method based on federal learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112989944A true CN112989944A (en) | 2021-06-18 |
Family
ID=76392970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110182984.XA Pending CN112989944A (en) | 2021-02-08 | 2021-02-08 | Intelligent video safety supervision method based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112989944A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113191340A (en) * | 2021-07-01 | 2021-07-30 | 智广海联(天津)大数据技术有限公司 | Community key personnel supervision system and method based on federal learning mode |
| CN113537513A (en) * | 2021-07-15 | 2021-10-22 | 青岛海尔工业智能研究院有限公司 | Model training method, device, system, equipment and medium based on federal learning |
| CN113591974A (en) * | 2021-07-29 | 2021-11-02 | 浙江大学 | Forgetting verification method based on forgetting-prone data subset in federated learning |
| CN114925238A (en) * | 2022-07-20 | 2022-08-19 | 山东大学 | Video clip retrieval method and system based on federal learning |
| WO2023029704A1 (en) * | 2021-08-31 | 2023-03-09 | 华为技术有限公司 | Data processing method, apparatus and system |
| CN117172632A (en) * | 2023-10-30 | 2023-12-05 | 湖南财信数字科技有限公司 | Enterprise abnormal behavior detection method, device, equipment and storage medium |
| WO2024032214A1 (en) * | 2022-08-11 | 2024-02-15 | 华为技术有限公司 | Reasoning method and related device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190227980A1 (en) * | 2018-01-22 | 2019-07-25 | Google Llc | Training User-Level Differentially Private Machine-Learned Models |
| CN111612168A (en) * | 2020-06-30 | 2020-09-01 | 腾讯科技（深圳）有限公司 | Management method and related device for machine learning task |
| CN112101489A (en) * | 2020-11-18 | 2020-12-18 | 天津开发区精诺瀚海数据科技有限公司 | Equipment fault diagnosis method driven by united learning and deep learning fusion |
| WO2021004551A1 (en) * | 2019-09-26 | 2021-01-14 | 深圳前海微众银行股份有限公司 | Method, apparatus, and device for optimization of vertically federated learning system, and a readable storage medium |
-
2021
- 2021-02-08 CN CN202110182984.XA patent/CN112989944A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190227980A1 (en) * | 2018-01-22 | 2019-07-25 | Google Llc | Training User-Level Differentially Private Machine-Learned Models |
| WO2021004551A1 (en) * | 2019-09-26 | 2021-01-14 | 深圳前海微众银行股份有限公司 | Method, apparatus, and device for optimization of vertically federated learning system, and a readable storage medium |
| CN111612168A (en) * | 2020-06-30 | 2020-09-01 | 腾讯科技（深圳）有限公司 | Management method and related device for machine learning task |
| CN112101489A (en) * | 2020-11-18 | 2020-12-18 | 天津开发区精诺瀚海数据科技有限公司 | Equipment fault diagnosis method driven by united learning and deep learning fusion |
Non-Patent Citations (1)
| Title |
|---|
| 赵羽: "面向视频监控基于联邦学习的智能边缘计算技术", 《通信学报》, vol. 41, no. 10, pages 109 - 114 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113191340A (en) * | 2021-07-01 | 2021-07-30 | 智广海联(天津)大数据技术有限公司 | Community key personnel supervision system and method based on federal learning mode |
| CN113537513A (en) * | 2021-07-15 | 2021-10-22 | 青岛海尔工业智能研究院有限公司 | Model training method, device, system, equipment and medium based on federal learning |
| CN113591974A (en) * | 2021-07-29 | 2021-11-02 | 浙江大学 | Forgetting verification method based on forgetting-prone data subset in federated learning |
| WO2023029704A1 (en) * | 2021-08-31 | 2023-03-09 | 华为技术有限公司 | Data processing method, apparatus and system |
| CN114925238A (en) * | 2022-07-20 | 2022-08-19 | 山东大学 | Video clip retrieval method and system based on federal learning |
| WO2024032214A1 (en) * | 2022-08-11 | 2024-02-15 | 华为技术有限公司 | Reasoning method and related device |
| CN117172632A (en) * | 2023-10-30 | 2023-12-05 | 湖南财信数字科技有限公司 | Enterprise abnormal behavior detection method, device, equipment and storage medium |
| CN117172632B (en) * | 2023-10-30 | 2024-01-19 | 湖南财信数字科技有限公司 | Enterprise abnormal behavior detection method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112989944A (en) | Intelligent video safety supervision method based on federal learning | |
| CN107729799A (en) | Crowd's abnormal behaviour vision-based detection and analyzing and alarming system based on depth convolutional neural networks | |
| CN108965826A (en) | Monitoring method, device, processing equipment and storage medium | |
| CN108710875A (en) | A kind of take photo by plane road vehicle method of counting and device based on deep learning | |
| CN113642403B (en) | Crowd abnormal intelligent safety detection system based on edge calculation | |
| CN110677300B (en) | Electric power safety management video intelligent shunting device and method based on mobile edge calculation | |
| Xu et al. | Fine-grained prediction of PM2. 5 concentration based on multisource data and deep learning | |
| Millefiori et al. | A distributed approach to estimating sea port operational regions from lots of AIS data | |
| CN115050177A (en) | Expressway congestion prediction method based on space-time multi-head attention mechanism | |
| CN113099242B (en) | Power transmission line video monitoring data processing method and system | |
| CN116664359A (en) | Intelligent fire early warning decision system and method based on multi-sensor fusion | |
| Song et al. | Deformable YOLOX: detection and rust warning method of transmission line connection fittings based on image processing technology | |
| Matek et al. | GitHub open source project recommendation system | |
| Turukmane et al. | Multispectral image analysis for monitoring by IoT based wireless communication using secure locations protocol and classification by deep learning techniques | |
| CN111178503A (en) | Mobile terminal-oriented decentralized target detection model training method and system | |
| CN109461303A (en) | A kind of traffic congestion state acquisition methods and device | |
| Beiji et al. | Crime hotspot detection and monitoring using video based event modeling and mapping techniques | |
| Tan et al. | Video analytics for indoor crowd estimation | |
| Wang et al. | The prediction of urban road traffic congestion by using a deep stacked long short-term memory network | |
| CN110446012A (en) | A kind of realization method and system of smart city | |
| Yang et al. | Static scene target detection based on VIBE algorithm | |
| Slaughter et al. | Vessel trajectory prediction with recurrent neural networks: An evaluation of datasets, features, and architectures | |
| CN116562824B (en) | Highway engineering full life cycle project management method and system | |
| Xu | Artificial intelligence unlocks ecological environment governance—smart statistical monitoring based on meteorology | |
| Veiga et al. | Container-Based IoT Architectures: Use Case for Visual Person Counting |
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 |