CN112241719A - Monitoring video target real-time query method based on edge cloud convolution neural network cascade - Google Patents
Monitoring video target real-time query method based on edge cloud convolution neural network cascade Download PDFInfo
- Publication number
- CN112241719A CN112241719A CN202011152354.XA CN202011152354A CN112241719A CN 112241719 A CN112241719 A CN 112241719A CN 202011152354 A CN202011152354 A CN 202011152354A CN 112241719 A CN112241719 A CN 112241719A
- Authority
- CN
- China
- Prior art keywords
- query
- target
- edge
- server
- cloud
- 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
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000012544 monitoring process Methods 0.000 title claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 8
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 59
- 239000000284 extract Substances 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000001960 triggered effect Effects 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 abstract description 5
- 230000008569 process Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
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
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a monitoring video target real-time query method based on edge cloud convolution neural network cascade, and belongs to the technical field of computer vision pedestrian re-identification. The edge server extracts moving targets from the monitoring video, preferentially identifies the targets through the lightweight CNN, uploads the targets which cannot be judged to the cloud server, and performs secondary identification through the high-precision CNN. Meanwhile, a side cloud task scheduling algorithm and a self-adaptive adjustment mechanism of threshold parameters are set, so that the performance of the query task of the monitoring video target can be effectively improved. Compared with a monitoring video target query method only using a cloud server, the method has the advantages that the bandwidth consumption from the edge server to the cloud server is obviously reduced, the cost is reduced, the average query time delay and the fluctuation of the query time delay are obviously reduced, the requirement on real-time performance is met, and meanwhile, the better query accuracy is kept; compared with a monitoring video target query method only using an edge server, the accuracy of query is obviously improved.
Description
Technical Field
The invention belongs to the technical field of edge computing/deep learning, and particularly relates to a monitoring video target real-time query method based on edge cloud convolution neural network cascade.
Background
With the wide deployment of monitoring cameras, the monitoring video data is increased explosively, and the monitoring video analysis method relying on manual screening cannot meet the actual requirements. On one hand, the speed is low, and the manual screening method is difficult to process a large amount of video data in real time; on the other hand, the cost is high, and huge labor cost expenses are brought by massive video analysis. Therefore, with the development of deep learning, especially the development of Convolutional Neural Network (CNN), it is more and more popular to automatically process and analyze the monitored video data by means of computer resources and video analysis algorithms.
The target query task in the monitoring video is widely concerned and developed due to the important significance of the target query task in intelligent security, intelligent traffic and the like. The target query task in the surveillance video refers to inputting a query target (such as a motorcycle) and surveillance video data and outputting a frame containing the query target in the surveillance video. The existing surveillance video target query method mainly uploads a surveillance video to a cloud server, and processes and analyzes the video by means of abundant computing resources of the cloud server. However, the huge network bandwidth requirement and data transmission delay are brought by uploading a large amount of monitoring videos to the cloud server, the new requirement of the target query task of the monitoring videos on the real-time performance is difficult to meet, and a huge burden is brought to the cloud server. Meanwhile, most edge devices have limited resources, and only a lightweight CNN model (such as MobileNet) can be deployed, but reliable query accuracy cannot be guaranteed. Therefore, there are two basic problems to be solved in the edge cloud system: 1) the query time delay is obviously reduced while the higher query accuracy is ensured, and the requirement of real-time property is met; 2) the method and the device can obviously reduce the bandwidth consumption while ensuring higher query accuracy, thereby reducing the cost.
Disclosure of Invention
The invention aims to provide a monitoring video target real-time query method based on edge cloud convolution neural network cascade connection so as to solve the problems.
In order to achieve the purpose, the invention adopts the following technical scheme:
a monitoring video target real-time query method based on edge cloud convolution neural network cascade comprises the following steps:
step 1: when a query command is received, fine tuning training is carried out in a cloud server based on a specific training data set and a specific Convolutional Neural Network (CNN) of a specified query target, and then the specific CNN is deployed to an edge server;
step 2: detecting and extracting a moving target in a monitoring video at an edge terminal server;
and step 3: selecting an edge server for executing an identification task for each extracted moving object at the edge server;
and 4, step 4: the method comprises the steps that a light-weight CNN is used on an edge end server to quickly identify a detected moving target, the identification confidence coefficient of a query target is output, and if the identification confidence coefficient is larger than alpha, the moving target is considered as the query target; if the recognition confidence coefficient is smaller than beta, the moving target is not considered as the query target; if the confidence is between alpha and beta, the lightweight CNN is considered to be incapable of judging whether the moving target is a query target, and the moving target is uploaded to a cloud server;
and 5: and for the moving target which cannot be judged by the lightweight CNN on the edge end server, carrying out secondary identification by using the high-precision CNN, judging whether the moving target is a query target, and finishing real-time query of the monitoring video target.
Further, in step 2, a target detection algorithm based on interframe difference is adopted to detect and extract the moving target in the monitoring video.
Further, in step 3, a load-based edge cloud task scheduling algorithm is adopted to select an edge server for executing the recognition task for each extracted moving object.
Further, step 3 specifically comprises:
for a total of N edge servers i, tiFor the inference time of the edge server i, i.e. the time delay of the edge server recognizing an image, QiFor the queue length of the identification task of the edge server i, the query time delay of an image processed by the edge server i is (Q)i+1)ti(ii) a When a moving target is detected, a real-time task scheduling program is triggered immediately, and the moving target is judged to be classified by which edge server d has the least query time delay, wherein the formula is as follows:
and taking the edge server d as an edge server for executing the identification task.
Further, in step 4, the lightweight CNN uses MobileNet _ v 2.
Further, in step 4, according to the load condition of the whole system, α and β are adaptively adjusted, and the adjustment formula is as follows:
αnew＝max{min{αold-γ1(Qdtd-s),1},0.5}
βnew＝γ2(1+αnew)
wherein, γ1、γ2Is a weight parameter, and γ1∈(0,1)，γ2E (0,1), s is query time delay standard, QdQueue length, t, for the identification task of the edge server ddIs the inference time of the edge end server d.
Further, in step 4, ResNet152 is used for high-precision CNN.
Further, in step 4, α and β are threshold parameters of the recognition confidence of the query target, and satisfy 0< β < α < 1.
Compared with the prior art, the invention has the following technical effects:
the invention discloses a monitoring video target real-time query method based on edge cloud convolution neural network cascade, wherein an edge server extracts a moving target from a monitoring video and preferentially identifies the target by a lightweight CNN on the edge server, the target which cannot be judged is uploaded to a cloud server, and high-precision CNN on the cloud server is used for secondary identification. Meanwhile, a side cloud task scheduling algorithm and a self-adaptive adjustment mechanism of threshold parameters are set, so that the performance of the query task of the monitoring video target can be effectively improved. The experimental result shows that compared with the traditional method for querying the monitoring video target only by using the cloud server, the system obviously reduces the bandwidth consumption from the edge server to the cloud server, reduces the cost, obviously reduces the average query time delay and the fluctuation of the query time delay, meets the requirement of real-time performance, and simultaneously keeps better query accuracy; compared with a monitoring video target query method only using an edge server, the system obviously improves the query accuracy.
Furthermore, an online fine tuning method is adopted to perform fine tuning training on the CNN deployed on the edge server, so that the model complexity is reduced, and the capability of recognizing specific types of targets is enhanced.
Further, a target detection algorithm based on interframe difference is adopted to detect and extract the moving target in the monitoring video. Two to three adjacent frames in the video are extracted at regular intervals, the gray value difference between the adjacent frames is calculated, and if the difference value of a certain area is close to zero, namely the area has small change in the time of the two to three frames, the area can be considered to have no moving object which is worthy of being extracted. On the contrary, if the difference value of a certain area is larger and exceeds a certain threshold value, the area is considered to have a moving object which is worth extracting, and then the area is extracted to be used as the detected moving object and is transmitted to the task scheduling module. The target detection algorithm is low in complexity, strong in robustness, low in computing resource consumption and capable of extracting the moving target from the video stream quickly.
Furthermore, a load-based edge cloud task scheduling algorithm is adopted to select an edge server for executing the identification task for each extracted moving target, so that the calculation resources are close to a data source and preferentially localized, the data transmission delay and the cloud data volume are effectively reduced, and the calculation load of each edge server is effectively balanced.
Furthermore, the CNN on the edge end server adopts the MobileNet _ v2 which has simple structure, less consumption of computing resources and relatively insufficient precision, thereby reducing the inference time delay of the detected moving target; the CNN on the cloud server adopts the ResNet152 which is high in precision and complex in structure, so that precision reduction caused by lightweight CNN is made up, and query accuracy of the system is effectively guaranteed.
Furthermore, the self-adaptive adjustment is carried out on alpha and beta according to the load condition of the whole system, when the load of the whole system is overlarge, the value of alpha is properly reduced, the value of beta is increased, the probability of target secondary identification is reduced, and therefore the load of the cloud server is reduced, the load of the whole system is reduced, and low and stable query time delay is maintained. When the load of the whole system is small, the value of alpha is properly increased, the value of beta is reduced, and the probability of secondary identification of the target is increased, so that the identification accuracy of the system is improved. The use of the edge cloud task scheduling algorithm and the threshold parameter adaptive adjustment mechanism obviously reduces the average query time delay and the fluctuation of the query time delay, balances the task load of each computing node, and balances the contradiction between the query delay and the system load.
Drawings
FIG. 1 is a diagram of the overall architecture of a system employing the method of the present invention;
FIG. 2 is a framework diagram of the online query portion of the method of the present invention;
FIG. 3 is a flow chart of a method of the present invention;
FIG. 4 is a histogram comparing the effectiveness of the present invention with other target query systems;
FIG. 5 is a graph of query latency versus line for the present invention and other target query systems.
Detailed Description
The invention will be described in further detail with reference to the following drawings and examples, which are given by way of illustration and not by way of limitation.
As shown in fig. 1, which is a logic block diagram of the process of the present invention, the method for real-time query of a surveillance video target based on edge cloud convolutional neural network cascade connection of the present invention includes the following steps:
step 1: when a user-defined query command is received, fine tuning training is carried out in a cloud server on the basis of a specific training data set and a specific CNN (convolutional neural network) of a specified query target, and then the specific CNN is deployed to an edge server;
step 2: detecting and extracting a moving target in the monitoring video by adopting a target detection algorithm based on interframe difference at an edge terminal server;
and step 3: selecting an edge server for executing an identification task for each extracted moving target by adopting a load-based edge cloud task scheduling algorithm at the edge server; the specific operation is as follows:
for a total of N edge servers i, tiFor the inference time of the edge server i, i.e. the time delay of the edge server recognizing an image, QiFor the queue length of the identification task of the edge server i, the query time delay of an image processed by the edge server i is (Q)i+1)ti(ii) a When a moving target is detected, a real-time task scheduling program is immediately triggered to judge which edge server d classifies the moving target with the least inquiry time delayThe formula is as follows:
and taking the edge server d as an edge server for executing the identification task.
And 4, step 4: the method comprises the steps that a light-weight CNN Mobile Net _ v2 is used on an edge server to quickly identify a detected target, the identification confidence that the target is a query target is output, and if the identification confidence is larger than alpha, the target is considered as the query target; if the recognition confidence is less than beta, the target is not considered as the query target; if the confidence is between alpha and beta, the lightweight CNN is considered to be incapable of judging whether the target is a query target or not, and the target is uploaded to a cloud server; alpha and beta are threshold parameters of the recognition confidence of the query target, and satisfy 0< beta < alpha < 1;
meanwhile, according to the load condition of the whole system, the self-adaptive adjustment is carried out on the alpha and the beta, and the adjustment formula is as follows:
αnew＝max{min{αold-γ1(Qdtd-s),1},0.5}
βnew＝γ2(1+αnew)
wherein, γ1、γ2Is a weight parameter, and γ1∈(0,1)，γ2E (0,1), s is query time delay standard, QdQueue length, t, for the identification task of the edge server ddIs the inference time of the edge end server d.
And 5: and for the target which cannot be judged by the lightweight CNN on the edge terminal server, performing secondary identification by using the high-precision CNN, judging whether the target is an inquiry target, and finishing real-time inquiry of the monitoring video target.
As shown in fig. 1 and 2, considering that huge network bandwidth consumption and data transmission delay are caused when a terminal device uploads a large amount of surveillance videos to a cloud server, and it is difficult to meet the requirements of surveillance video target query tasks on real-time performance, data is stored in an edge server, target identification is performed by using a lightweight CNN first, only an image that the lightweight CNN cannot judge whether the image is a query target is uploaded to the cloud server, and secondary identification is performed by using a high-precision CNN on the cloud server, so that the relationship between query delay and query accuracy is balanced.
As shown in fig. 1, fig. 2 and fig. 3, the edge cloud CNN cascade-based real-time query system for a surveillance video target has the following details:
1) online fine adjustment: when a new user-defined query command is received, fine-tuning training is carried out in the cloud server based on the specific training data set and the specific CNN of the specified query target, and the specific CNN is deployed to the edge server, so that the model complexity is reduced, and the capability of identifying a specific type of object is enhanced;
2) target detection: the module extracts two to three adjacent frames in the video at regular intervals, calculates the gray value difference between the adjacent frames, and if the difference value of a certain area is close to zero, namely the area has small change in the time of the two to three frames, the area can be considered to have no moving object which is worthy of being extracted. On the contrary, if the difference value of a certain area is larger and exceeds a certain threshold value, the area is considered to have a moving object which is worth extracting, and then the area is extracted to be used as the detected moving object and transmitted to the task scheduling module;
3) task scheduling: the module receives the detected moving objects, and selects a server for executing an identification task for each extracted moving object according to the load conditions of the edge server and the cloud server. Comparing the processing time delay of the detected target in the edge server and the cloud server, and selecting a server with smaller processing time delay to identify the target;
4) lightweight CNN target identification: the object is quickly identified on the edge end server by using a lightweight CNN (such as MobileNet _ v2), and the identification confidence that the object is the query object is output. If the confidence coefficient is greater than alpha, the target is considered as a query target; if the confidence is less than beta, the target is not considered as the query target; and if the confidence is between alpha and beta, the lightweight CNN is considered to be incapable of judging whether the target is a query target or not, and uploading the query target to the cloud server. α, β are threshold parameters, satisfying 0< β < α < 1;
5) high-precision CNN target identification process: and identifying a moving target uploaded to the cloud server after task scheduling and a target which cannot be judged by the lightweight CNN on the edge end server by using a high-precision CNN (such as ResNet152), judging whether the target is a query target, and outputting a confidence coefficient that the target is the query target. If the confidence exceeds a certain threshold, the target is considered as the query target.
Fig. 4 and 5 are a histogram and a query latency comparison line graph comparing the effects of the present invention and other target query systems. The SurveilEdge is a monitoring video target real-time query system based on edge cloud CNN cascade connection, the SurveilEdge (fixed) is an edge cloud query system which does not use an edge cloud task scheduling algorithm and a threshold parameter self-adaptive adjustment mechanism, the edge-only is a pure edge target query system with all computations on an edge server, and the close-only is a pure cloud target query system with all computations on a cloud server. As can be seen from fig. 3, compared with the traditional cloud-only query system (cloud-only), the bandwidth overhead of the system (surfiledge) is reduced by 3 times, the query response time is reduced by 14.56 times, and meanwhile, better query accuracy is maintained; compared with a pure edge target query system (edge-only), the query accuracy of the system is improved by 27.5%, and 2.05 times of query acceleration is realized. The edge cloud cooperative structure of the system makes a good balance between the accuracy and the delay of the query, the image is firstly identified and filtered on the edge server, and the data volume uploaded to the cloud server is reduced, so that a large amount of transmission delay is reduced. Meanwhile, images with low recognition confidence coefficient are uploaded to a cloud server for secondary recognition, and accuracy reduction caused by recognition by using a lightweight CNN in an edge server is made up.
It can be observed from fig. 4 that the use of the edge cloud task scheduling algorithm and threshold parameter adaptive scheduling significantly reduces the average delay of the query. When one computing node is overloaded, the edge cloud task scheduling algorithm allocates the next task to another node with relatively less computing pressure, and the continuous increase of waiting delay is avoided. In fact, when the query delay of a frame is too large, the query result of the frame becomes obsolete, and the query result loses meaning. The use of the edge cloud task scheduling algorithm and the threshold parameter adaptive adjustment mechanism enables the high-precision CNN in the cloud server to identify more images in a non-busy time, and reduces the overall load of the system by reducing the proportion of the images identified twice in a busy time, thereby obviously reducing the average query delay and the fluctuation of the query delay and balancing the task load of each computing node.
It should be noted that the above description is only a part of the embodiments of the present invention, and all equivalent changes made according to the present invention are included in the protection scope of the present invention. Those skilled in the art to which the invention relates may substitute similar embodiments for the specific examples described, all falling within the scope of the invention, without thereby departing from the invention or exceeding the scope of the claims defined thereby.
Claims (8)
1. A monitoring video target real-time query method based on edge cloud convolution neural network cascade is characterized by comprising the following steps:
step 1: when a query command is received, fine tuning training is carried out in a cloud server on the basis of a specific training data set and a specific CNN convolutional neural network of a specified query target, and then the specific CNN is deployed to an edge server;
step 2: detecting and extracting a moving target in a monitoring video at an edge terminal server;
and step 3: selecting an edge server for executing an identification task for each extracted moving object at the edge server;
and 4, step 4: the method comprises the steps that a light-weight CNN is used on an edge end server to quickly identify a detected moving target, the identification confidence coefficient of a query target is output, and if the identification confidence coefficient is larger than alpha, the moving target is considered as the query target; if the recognition confidence coefficient is smaller than beta, the moving target is not considered as the query target; if the confidence is between alpha and beta, the lightweight CNN is considered to be incapable of judging whether the moving target is a query target, and the moving target is uploaded to a cloud server;
and 5: and for the moving target which cannot be judged by the lightweight CNN on the edge end server, carrying out secondary identification by using the high-precision CNN, judging whether the moving target is a query target, and finishing real-time query of the monitoring video target.
2. The method for querying the target of the surveillance video in real time based on the edge cloud convolutional neural network cascade connection as claimed in claim 1, wherein in the step 2, a target detection algorithm based on inter-frame difference is adopted to detect and extract the moving target in the surveillance video.
3. The method for querying the surveillance video target in real time based on the edge cloud convolutional neural network cascade connection as claimed in claim 1, wherein in step 3, a load-based edge cloud task scheduling algorithm is adopted to select an edge server for executing an identification task for each extracted moving target.
4. The monitoring video target real-time query method based on edge cloud convolutional neural network cascade connection as claimed in claim 3, wherein step 3 specifically comprises:
for a total of N edge servers i, tiFor the inference time of the edge server i, i.e. the time delay of the edge server recognizing an image, QiFor the queue length of the identification task of the edge server i, the query time delay of an image processed by the edge server i is (Q)i+1)ti(ii) a When a moving target is detected, a real-time task scheduling program is triggered immediately, and the moving target is judged to be classified by which edge server d has the least query time delay, wherein the formula is as follows:
and taking the edge server d as an edge server for executing the identification task.
5. The method for real-time query of surveillance video targets based on edge cloud convolutional neural network cascade as claimed in claim 1, wherein in step 4, the lightweight CNN employs MobileNet _ v 2.
6. The method for real-time query of surveillance video targets based on edge cloud convolutional neural network cascade as claimed in claim 1, wherein in step 4, according to the load condition of the whole system, the adaptive adjustment is performed on α and β, and the adjustment formula is as follows:
αnew＝max{min{αold-γ1(Qdtd-s)，1}，0.5}
βnew＝γ2(1+αnew)
wherein, γ1、γ2Is a weight parameter, and γ1∈(0，1)，γ2E (0,1), s is query time delay standard, QdQueue length, t, for the identification task of the edge server ddIs the inference time of the edge end server d.
7. The method for real-time query of monitored video targets based on edge cloud convolutional neural network cascade as claimed in claim 1, wherein in step 4, ResNet152 is adopted as the high-precision CNN.
8. The method for real-time query of the surveillance video target based on the edge cloud convolutional neural network cascade connection as claimed in claim 1, wherein in step 4, α and β are threshold parameters of the recognition confidence of the query target, and satisfy 0< β < α < 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011152354.XA CN112241719A (en) | 2020-10-23 | 2020-10-23 | Monitoring video target real-time query method based on edge cloud convolution neural network cascade |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011152354.XA CN112241719A (en) | 2020-10-23 | 2020-10-23 | Monitoring video target real-time query method based on edge cloud convolution neural network cascade |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112241719A true CN112241719A (en) | 2021-01-19 |
Family
ID=74169514
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011152354.XA Pending CN112241719A (en) | 2020-10-23 | 2020-10-23 | Monitoring video target real-time query method based on edge cloud convolution neural network cascade |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112241719A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113298063A (en) * | 2021-07-28 | 2021-08-24 | 江苏电力信息技术有限公司 | Dynamic object detection method based on cloud-edge |
| CN114154018A (en) * | 2022-02-08 | 2022-03-08 | 中国电子科技集团公司第二十八研究所 | Cloud-edge collaborative video stream processing method and system for unmanned system |
| CN114445411A (en) * | 2022-04-11 | 2022-05-06 | 广东电网有限责任公司佛山供电局 | Unmanned aerial vehicle line patrol defect identification system and control method |
| WO2023093053A1 (en) * | 2021-11-25 | 2023-06-01 | 达闼科技（北京）有限公司 | Inference implementation method, network, electronic device, and storage medium |
| CN116366649A (en) * | 2023-06-01 | 2023-06-30 | 中电云脑(天津)科技有限公司 | Side cloud cooperative electroencephalogram data task scheduling method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506432A (en) * | 2017-08-22 | 2017-12-22 | 安徽大学 | A kind of object detection method and system based on edge calculations towards Vehicular video |
| CN109543829A (en) * | 2018-10-15 | 2019-03-29 | 华东计算技术研究所（中国电子科技集团公司第三十二研究所） | Method and system for hybrid deployment of deep learning neural network on terminal and cloud |
-
2020
- 2020-10-23 CN CN202011152354.XA patent/CN112241719A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506432A (en) * | 2017-08-22 | 2017-12-22 | 安徽大学 | A kind of object detection method and system based on edge calculations towards Vehicular video |
| CN109543829A (en) * | 2018-10-15 | 2019-03-29 | 华东计算技术研究所（中国电子科技集团公司第三十二研究所） | Method and system for hybrid deployment of deep learning neural network on terminal and cloud |
Non-Patent Citations (1)
| Title |
|---|
| SHIBO WANG: "《SurveilEdge: Real-time Video Query based on Collaborative Cloud-Edge Deep Learning》", 《ARXIV:2001.01043V2》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113298063A (en) * | 2021-07-28 | 2021-08-24 | 江苏电力信息技术有限公司 | Dynamic object detection method based on cloud-edge |
| CN113298063B (en) * | 2021-07-28 | 2021-10-15 | 江苏电力信息技术有限公司 | Dynamic object detection method based on cloud-edge |
| WO2023093053A1 (en) * | 2021-11-25 | 2023-06-01 | 达闼科技（北京）有限公司 | Inference implementation method, network, electronic device, and storage medium |
| CN114154018A (en) * | 2022-02-08 | 2022-03-08 | 中国电子科技集团公司第二十八研究所 | Cloud-edge collaborative video stream processing method and system for unmanned system |
| CN114154018B (en) * | 2022-02-08 | 2022-05-10 | 中国电子科技集团公司第二十八研究所 | Cloud-edge collaborative video stream processing method and system for unmanned system |
| CN114445411A (en) * | 2022-04-11 | 2022-05-06 | 广东电网有限责任公司佛山供电局 | Unmanned aerial vehicle line patrol defect identification system and control method |
| CN116366649A (en) * | 2023-06-01 | 2023-06-30 | 中电云脑(天津)科技有限公司 | Side cloud cooperative electroencephalogram data task scheduling method and system |
| CN116366649B (en) * | 2023-06-01 | 2023-09-05 | 中电云脑(天津)科技有限公司 | Side cloud cooperative electroencephalogram data task scheduling method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112241719A (en) | Monitoring video target real-time query method based on edge cloud convolution neural network cascade | |
| Li et al. | Reducto: On-camera filtering for resource-efficient real-time video analytics | |
| CN112949673B (en) | Feature fusion target detection and identification method based on global attention | |
| CN108009473B (en) | Video structuralization processing method, system and storage device based on target behavior attribute | |
| WO2022127180A1 (en) | Target tracking method and apparatus, and electronic device and storage medium | |
| CN107483887B (en) | Early warning detection method for emergency in smart city video monitoring | |
| CN111311630A (en) | Method and system for intelligently counting quantity of goods through videos in warehousing management | |
| CN104658152B (en) | A kind of moving object intrusion alarm method based on video | |
| WO2011097795A1 (en) | Method and system for population flow statistics | |
| WO2023155482A1 (en) | Identification method and system for quick gathering behavior of crowd, and device and medium | |
| KR20140095333A (en) | Method and apparratus of tracing object on image | |
| CN108334831A (en) | A kind of monitoring image processing method, monitoring terminal and system | |
| CN112257569A (en) | Target detection and identification method based on real-time video stream | |
| CN111553213A (en) | Real-time distributed identity-aware pedestrian attribute identification method in mobile edge cloud | |
| CN109919033B (en) | Self-adaptive city people searching method based on edge calculation | |
| TW201308254A (en) | Motion detection method for comples scenes | |
| CN113673311A (en) | Traffic abnormal event detection method, equipment and computer storage medium | |
| CN111507391A (en) | Intelligent identification method for nonferrous metal broken materials based on machine vision | |
| CN104853060A (en) | High-definition video preprocessing method and system | |
| CN115130649A (en) | Deep learning model partitioning method and device for pipeline distributed end cloud collaborative reasoning | |
| CN114741185A (en) | Edge computing system for multi-target video monitoring and working method thereof | |
| CN113822110B (en) | Target detection method and device | |
| CN110163125B (en) | Real-time video identification method based on track prediction and size decision | |
| CN117237867A (en) | Self-adaptive field monitoring video target detection method and system based on feature fusion | |
| CN104301722A (en) | Video streaming fuzzy detection method based on frequency domain |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210119 |