CN110119718A - A kind of overboard detection and Survivable Control System based on deep learning - Google Patents
A kind of overboard detection and Survivable Control System based on deep learning Download PDFInfo
- Publication number
- CN110119718A CN110119718A CN201910404506.1A CN201910404506A CN110119718A CN 110119718 A CN110119718 A CN 110119718A CN 201910404506 A CN201910404506 A CN 201910404506A CN 110119718 A CN110119718 A CN 110119718A
- Authority
- CN
- China
- Prior art keywords
- network
- personnel
- water
- data set
- detection
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- 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
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
- G08B21/02—Alarms for ensuring the safety of persons
- G08B21/08—Alarms for ensuring the safety of persons responsive to the presence of persons in a body of water, e.g. a swimming pool; responsive to an abnormal condition of a body of water
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Human Computer Interaction (AREA)
- Emergency Management (AREA)
- Image Analysis (AREA)
Abstract
The overboard detection and Survivable Control System that the invention discloses a kind of based on deep learning, are related to computer vision field, wherein overboard detection method includes: to establish personnel's image data set in water；Using LabelImg tool in original image acceptance of the bid water filling personnel's classification in the position of personnel and water, and markup information that every picture generates is stored so as to network training with xml document format；Using YOLOv2 deep learning target detection frame training dataset, needs to carry out established data set clustering before network training and data set is divided into training set, verifying collection and test set according to 6:2:2 ratio.Survivable Control System includes monitor video input unit, drowning man's detection unit and alarm rescue unit.The response efficiency of water surface rescue is effectively promoted in the present invention, improves the survival probability of drowning man.
Description
Technical field
This patent is related to computer vision field, particularly relate to it is a kind of in the case where monitoring environment based on the overboard of deep learning
Detection and Survivable Control System.
Background technique
The park man-made lake of enormous amount, reservoir and the patrol in lake dynamics are little, safety measure is not in place and lacks
Profession emergency rescue device waterborne, especially summer, causes drowning accident frequently to occur.In order to avoid the generation of tragedy, have
The round-the-clock monitoring of waters state is had been achieved in a little scenes, however, there remains someone to guard monitored picture, in this way
It does not only time-consuming and laborious, but also not can guarantee and drowning man is succoured in real time, therefore to the real-time automatic prison of drowning man
Surveying and improve emergency reaction ability becomes to be even more important.
The detection Primary Reference for being directed to drowning man at present is ground human body detecting method, using traditional based on figure
As the overboard human testing of processing is faced with, waters background is complicated, distinguishes vulnerable to illumination effect, inverted image and true overboard situation difficulty
And overboard situation the problems such as having differences.The appearance of deep learning solves in target detection due to the variability of object, mesh
Object is marked under different scenes with the difficult point of different colours, size and form.Depth network has the parameter of magnanimity, makes it
With powerful learning ability, complex background, illumination, inverted image etc. can be excluded, bring interference, energy are detected to drowning man
With reaching memory type form of the storing and resolving drowning man under different scenes, the standard of drowning man's detection based on deep learning
True rate will be significantly larger than commonsense method.
Summary of the invention
In view of the above technical problems, the purpose of the present invention is to provide in the case where monitoring environment based on the overboard of deep learning
Detection method.Particularly, according to this method propose a kind of Survivable Control Systems.The present invention is first using people in the water established
Member's image data set training YOLOv2 target detection network, then by extracting the realtime graphic in monitoring waters, and then to these
Video frame images are detected, once discovery has drowning man, it is concurrent drowning man present position to be confined on monitored picture
Signal is controlled out, and then executes corresponding alarm and rescue operation, facilitates the life shortened rescue time, improve drowning man
Deposit probability.
To achieve the above object, the present invention is realized according to following technical scheme:
A kind of overboard detection method based on deep learning, which comprises the steps of:
Step S1: personnel's image data set in water is established；
Step S2: use LabelImg tool, original image acceptance of the bid water filling in personnel position and water in personnel's class
Not, and with xml document format the markup information of every picture generation is stored so as to network training；
Step S3: the use of network frame is YOLOv2 deep learning target detection network, needs to build before network training
The data set stood carries out clustering and data set is divided into training set, verifying collection and test set according to 6:2:2 ratio,
Wherein training set is for training personnel in water to detect network and determining each layer parameter of network, and verifying collection is for adjusting the super of model
Parameter and entry evaluation is carried out for the generalization ability to model, test set is used to assess the generalization ability of mould final mask, no
As the foundation for adjusting ginseng, the relevant selection of selection feature.
In above-mentioned technical proposal, the step S1 includes:
Step S101: should collect no less than 1500 different drowning man's images, detect net as personnel in subsequent water
The basis of data set needed for network training；
Step S102: it is carried out using the method that the saturation degree of color, the data of brightness and contrast enhance and to image
It cuts, the EDS extended data set of translation scaling and rotation, network training is avoided over-fitting occur.
In above-mentioned technical proposal, clustering is carried out to data set using K-means algorithm in the step S3, according to
Cluster result determines that the quantity and size of candidate frame when network training is distributed.
A kind of Survivable Control System based on deep learning, for realizing the overboard detection method of any of the above-described, packet
It includes: monitor video input unit, drowning man's detection unit and alarm rescue unit, wherein monitor video input is single
Member acquires Surface Picture data using infrared camera in real time and by transmission of video images to monitor terminal, realizes two
14 hours real time monitorings to waters；Drowning man's detection unit uses personnel in the water with optimal weights to detect
Network is to the video image real-time detection for being input to monitor terminal, can be in monitored picture once detection discovery has drowning man
In confine drowning man position, and provide control signal；The alarm rescue unit includes salvor, broadcast notice mould
Block, guardroom and the terminal of patrol, salvor moment are located at bank in a dormant state, broadcast notification module according to
Whether there is or not dangerous situation and emergency handling both of which for alarm signal automatic switchover, while drowning man's information can be sent to guardroom
And the terminal of each patrol shortens rescue time to be unfolded to succour in time, the existence for improving drowning man is general
Rate.
Compared with the prior art, the invention has the following advantages:
(1) present invention detects network, Jin Erjin using personnel in YOLOv2 deep learning target detection network struction water
Row drowning man detection, instead of traditional detection method based on image procossing, and the detection framework based on YOLOv2
Real-time and accurate performance reach the requirement that video monitoring system is handled in real time.
(2) on the one hand the present invention increases trained data volume by data enhancement methods, improves the extensive energy of model
Power；On the other hand noise data is increased, facilitates the robustness of lift scheme.Candidate frame weight is carried out using k-means algorithm
New settings is easier to acquire accurate predicted position convenient for network.
(3) the overboard detection method provided by the invention based on deep learning, can exclude due to waters background is complicated,
It is distinguished vulnerable to illumination effect, inverted image and true overboard situation difficulty and the brings such as overboard situation has differences is interfered, improved
The accuracy of detection algorithm.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will to embodiment or
Attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
Some embodiments of the present invention, for those of ordinary skill in the art, without creative efforts, also
Other attached drawings can be obtained according to these attached drawings.
Fig. 1 is overboard detection method workflow schematic diagram provided by the invention；
Fig. 2 is Darknet-19 schematic network structure provided by the invention；
Fig. 3 is YOLOv2 schematic network structure provided by the invention；
Fig. 4 is that pre-selection frame proposed by the present invention resets schematic diagram；
Fig. 5 is the sample calculation figure of YOLOv2 network boundary frame position provided by the invention and size；
Fig. 6 is K-means clustering algorithm schematic diagram provided by the invention；
Fig. 7 is Survivable Control System schematic diagram proposed by the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.
As shown in Figure 1, the overboard detection method proposed by the present invention based on deep learning, includes the following steps:
Step S1: personnel's image data set in water is established；
The training that personnel detect network in water needs to act on a large amount of image data, and the essence of data then determines
Whether the algorithm of application is suitable, and the quality of image data also determines the fine or not degree of algorithm performance.Step S1 includes:
Step S101: should collect no less than 1500 different drowning man's images, detect net as personnel in subsequent water
The basis of data set needed for network training；Image data can be collected by way of intercepting to associated video image data, separately
Outside image document can also be collected by way of being shot on the spot with the help of professional person.
Step S102: it is carried out using the method that the saturation degree of color, the data of brightness and contrast enhance and to image
It cuts, the EDS extended data set of translation scaling and rotation, network training is avoided over-fitting occur.In deep learning, only
It just can guarantee the accuracy of testing result in data set comprising a large amount of sample.On the one hand the method for data enhancing is conducive to increase
Add trained data volume, improve the generalization ability of model, on the other hand increase noise data, facilitates the Shandong of lift scheme
Stick can fundamentally promote the performance of network.Image is overturn, random cropping, translation scaling, to image by one
Determine angle do rotate, operation and random superposition noise on the basis of original image such as visible change and piecewise affine.Its
In, overturning includes flip horizontal and flip vertical；Visible change is to image using 4 random perspective transforms；Point
Affine section is to place a regular dot grid on the image, according to these mobile points of the sample size being just distributed very much and surrounding
Image-region.On the one hand the method for data enhancing is conducive to increase the data volume of training, improve the generalization ability of model, separately
On the one hand noise data is increased, facilitates the robustness of lift scheme.
Step S2: use LabelImg tool, original image acceptance of the bid water filling in personnel position and water in personnel's class
Not, and with xml document format the markup information of every picture generation is stored so as to network training；
Step S3: the use of network frame is YOLOv2 deep learning target detection network, needs to build before network training
The data set stood carries out clustering and data set is divided into training set, verifying collection and test set according to 6:2:2 ratio,
Wherein training set is for training personnel in water to detect network and determining each layer parameter of network, and verifying collection is for adjusting the super of model
Parameter and entry evaluation is carried out for the generalization ability to model, test set is used to assess the generalization ability of mould final mask, no
As the foundation for adjusting ginseng, the relevant selection of selection feature.
In step S2, image calibration uses LabelImg tool, in each image in established data set
Personnel are marked with rectangle frame in water, and the coordinate information of personnel in the picture in water can be obtained, personnel in simultaneous selection water
Classification information, the tool can generate corresponding xml document to every picture and store markup information.The coordinate of the rectangle frame
Information includes the upper left corner and the lower right corner coordinate value of rectangle frame, i.e. Xmin, Ymin, Xmax, Ymax.
It is YOLOv2 that personnel, which detect network frame used in network training, in water, and the foundation structure of the detection network is
Darknet-19, as shown in Fig. 2, extracting feature using the convolution kernel of 3*3 size, the convolution kernel of 1*1 size carrys out data fusion
And the maximum value pond layer that step-length is 2 is down-sampled.It is illustrated in figure 3 YOLOv2 detection framework schematic diagram, altogether includes 19
Convolutional layer and 5 maximum value pond layers.Network training is set to be easier to restrain simultaneously energy using BN layers after each convolutional layer
Prevent over-fitting.Parallel link is also used simultaneously, shallow-layer feature is merged with high-layer semantic information, such YOLOv2
Network obtains more fine-grained features there is no because down-sampled operation leads to information loss instead.Output layer uses volume
Lamination substitutes full articulamentum, classification information, confidence level size and the coordinate information of final output object.YOLOv2 is in identification kind
Class, precision, speed and positioning accuracy etc. are attained by application request.
Detection in order to better understand based on YOLOv2 network implementations drowning man, herein to YOLOv2 network
Working principle is illustrated:
As shown in figure 4, the grid that the image segmentation of input is 13*13 by YOLOv2 network, while being generated for each grid
Multiple scales, candidate frame not of uniform size finally predict the recurrence coordinate of these candidate frames on characteristic pattern in a manner of full convolution
Value and confidence score (Confidence Score), while predicting that (C is object to C condition class probability value for each candidate frame
Body classification number).In the network training stage, each grid is responsible for the object that inspection center falls in the grid, specially and object
The maximum one of candidate frame of body actual frames IOU value (hand over and than) is responsible for predicting this object.So-called confidence level includes two
Aspect, first is that a possibility that this bounding box contains target size, is denoted as Pr (object)；Second is that this bounding box is accurate
Degree is characterized with the friendship of candidate frame and actual frames and ratio, is denoted as IOU.Therefore confidence level is defined as Pr (object) * IOU,
Reflection is that have the probability of object to have much and fitting actual frames degree in the pre-selection frame.The size of bounding box and position
It can be characterized with this 4 values of x, y, h, w, wherein x and y is the centre coordinate of bounding box, and h and w are the width and height of bounding box.
As shown in figure 5, the predicted value (x, y) of centre coordinate is the deviant relative to each cell top left co-ordinate point, and single
Position is relative to cell size.And w the and h predicted value of bounding box is width and the high ratio relative to entire picture.
It is detected in the present invention just for personnel in water, C value is 1 in the foregoing.
In step S3, in addition, using k-means algorithm to the bounding of training set before carrying out network training
Boxes does clustering, finds the suitable pre-selection frame size collected for demographic data in this water, preselect in network configuration
The reset of frame is easier to acquire accurate predicted position convenient for network.The cluster process is as shown in fig. 6, cluster centre
K can test of many times setting, K=5, range formula can be used are as follows:
D (box, centroid)=1-IOU (box, centroid)
In addition, established data set is divided into training set according to 6:2:2 ratio before carrying out network training, is tested
Card collection and test set.Training set adjusts each layer parameter of network for training personnel in water to detect network.Verifying collection is for adjusting
The hyper parameter of model and for model generalization ability carry out entry evaluation.Test set is used to assess the extensive of final mask
Ability, not as the foundation for adjusting ginseng, the relevant selection of selection feature scheduling algorithm, the assessment result selection according to test set has
Personnel detect network in the water of optimal weight.
The loss function for carrying out personnel's detection network training in water based on YOLOv2 network is as follows:
Wherein, W and H refers to that the width and height of characteristic pattern (13x13), A refer to that each grid cell (cell) is corresponding
The number (5) of anchor box, various λ indicate the weight of all kinds of losses.What first item calculated is the candidate frame of each network unit
Confidence level when being less than threshold value with the IOU of actual frames is lost, i.e., does not have confidence level loss when target in predicting candidate frame.The
Binomial calculates the coordinate loss that the coordinate value for surveying frame is waited in preceding 12800 photo iterative process with the coordinate value predicted.The
Three calculate and the loss of each section of the matched candidate frame of some actual frames, including error of coordinate, confidence level error and classification
Error.
As shown in fig. 7, the present invention proposes a kind of Survivable Control System according to the method provided, comprising: monitor video is defeated
Enter unit, drowning man's detection unit and alarm rescue unit.
Preferably, as an embodiment, to realize real time monitoring of the twenty four hours to waters, monitor video
Input unit acquires Surface Picture data in real time using infrared camera, and by transmission of video images to monitor terminal.
Wherein exterior aerial integration orientation wireless bridge can be used in monitor video transmitting terminal, and it is external that wireless bridge can be used in receiving end
Antenna omnidirectional receives, mobile convenient for later period point.In view of bandwidth and distance, the networking mode of 4 hairs one receipts can be used, entirely
Series is divided into multiple groups, every group of 4 transmitting terminals, a receiving end；All receiving ends are finally imported into monitoring center together
Interchanger, finally connects monitoring host computer, and upper wall is shown.
Further, drowning man's detection unit is using the trained detection network with optimal weights to being input to
The video image real-time detection of monitor terminal.Once detection discovery has drowning man, drowning man can be confined in monitored picture
Position, and provide control signal.
Further, alarm rescue unit includes salvor, broadcast notification module, guardroom and the end of patrol
End, the salvor moment is located at bank in a dormant state, broadcast notification module according to alarm signal automatic switchover whether there is or not dangerous situation with
And emergency handling both of which, while drowning man's information can be sent to guardroom and the terminal of each patrol, with
Just expansion rescue in time, shortens rescue time, improves the survival probability of drowning man.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (4)
1. a kind of overboard detection method based on deep learning, which comprises the steps of:
Step S1: personnel's image data set in water is established；
Step S2: use LabelImg tool, original image acceptance of the bid water filling in personnel position and water in personnel's classification, and with
Xml document format stores the markup information of every picture generation so as to network training；
Step S3: the use of network frame is YOLOv2 deep learning target detection network, needs to establish before network training
Data set carry out clustering and by data set according to 6:2:2 ratio be divided into training set, verifying collection and test set, wherein instructing
Practicing collection, personnel detect network in water and determining each layer parameter of network, verifying collect hyper parameter and use for adjusting model for training
Entry evaluation is carried out in the generalization ability to model, test set is used to assess the generalization ability of mould final mask, join not as tune,
Select the foundation of the relevant selection of feature.
2. overboard detection method according to claim 1, which is characterized in that the step S1 includes:
Step S101: should collect no less than 1500 different drowning man's images, detect network instruction as personnel in subsequent water
The basis of data set needed for practicing；
Step S102: using the saturation degree of color, the data of brightness and contrast enhance method and image is cut,
The EDS extended data set of translation scaling and rotation, avoids network training from over-fitting occur.
3. overboard detection method according to claim 1, which is characterized in that use K-means algorithm in the step S3
Clustering is carried out to data set, determines that the quantity and size of candidate frame when network training is distributed according to cluster result.
4. a kind of Survivable Control System based on deep learning, for realizing any one of claims 1 to 3 overboard detection side
Method characterized by comprising monitor video input unit, drowning man's detection unit and alarm rescue unit, wherein institute
State monitor video input unit using infrared camera Surface Picture data are acquired in real time and by transmission of video images extremely
Monitor terminal realizes twenty four hours to the real time monitoring in waters；Drowning man's detection unit, which uses, has optimal weights
Water in personnel detect network to the video image real-time detection for being input to monitor terminal, once detection discovery has drowning man,
Drowning man position can be confined in monitored picture, and provide control signal；The alarm rescue unit include salvor,
Broadcast notification module, guardroom and the terminal of patrol, salvor moment are located at bank in a dormant state, broadcast notice
Whether there is or not dangerous situation and emergency handling both of which according to alarm signal automatic switchover for module, while drowning man's information can be sent to
The terminal of guardroom and each patrol shorten rescue time, improve the life of drowning man to be unfolded to succour in time
Deposit probability.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910404506.1A CN110119718A (en) | 2019-05-15 | 2019-05-15 | A kind of overboard detection and Survivable Control System based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910404506.1A CN110119718A (en) | 2019-05-15 | 2019-05-15 | A kind of overboard detection and Survivable Control System based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110119718A true CN110119718A (en) | 2019-08-13 |
Family
ID=67522500
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910404506.1A Pending CN110119718A (en) | 2019-05-15 | 2019-05-15 | A kind of overboard detection and Survivable Control System based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110119718A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110569772A (en) * | 2019-08-30 | 2019-12-13 | 北京科技大学 | Method for detecting state of personnel in swimming pool |
| CN110619365A (en) * | 2019-09-18 | 2019-12-27 | 苏州经贸职业技术学院 | Drowning detection method |
| CN111028480A (en) * | 2019-12-06 | 2020-04-17 | 江西洪都航空工业集团有限责任公司 | Drowning detection and alarm system |
| CN111545955A (en) * | 2020-04-20 | 2020-08-18 | 华南理工大学 | Door plate welding spot identification and welding path planning method |
| CN112232299A (en) * | 2020-11-09 | 2021-01-15 | 江苏科技大学 | Automatic navigation method for rescuing drowning car based on deep learning |
| CN112418181A (en) * | 2020-12-13 | 2021-02-26 | 西北工业大学 | Personnel overboard detection method based on convolutional neural network |
| CN112489371A (en) * | 2020-11-26 | 2021-03-12 | 上海天健体育科技发展有限公司 | Swimming pool drowning prevention early warning system based on computer vision |
| CN113033478A (en) * | 2021-04-19 | 2021-06-25 | 曲阜师范大学 | Pedestrian detection method based on deep learning |
| CN113044184A (en) * | 2021-01-12 | 2021-06-29 | 桂林电子科技大学 | Deep learning-based water rescue robot and drowning detection method |
| CN113188000A (en) * | 2021-05-14 | 2021-07-30 | 太原理工大学 | Lake side downwater personnel identification rescue system and method |
| CN113298029A (en) * | 2021-06-15 | 2021-08-24 | 广东工业大学 | Blind person walking assisting method and system based on deep learning target detection |
| CN113610178A (en) * | 2021-08-17 | 2021-11-05 | 湖南工学院 | Inland ship target detection method and device based on video monitoring image |
| CN115937506A (en) * | 2023-03-09 | 2023-04-07 | 南京邮电大学 | Method, system, device and medium for positioning bridge side falling point hole position information |
| CN116740649A (en) * | 2023-08-07 | 2023-09-12 | 山东科技大学 | Deep learning-based real-time detection method for behavior of crewman falling into water beyond boundary |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103413114A (en) * | 2013-05-17 | 2013-11-27 | 浙江大学 | Near-drowning behavior detection method based on support vector machine |
| CN104992140A (en) * | 2015-05-27 | 2015-10-21 | 上海海事大学 | Sea surface abnormal floating object detecting method based on remote sensing image |
| CN106022230A (en) * | 2016-05-11 | 2016-10-12 | 太原理工大学 | Video-based detection method for drowning event in swimming pool |
| CN108509860A (en) * | 2018-03-09 | 2018-09-07 | 西安电子科技大学 | HOh Xil Tibetan antelope detection method based on convolutional neural networks |
| CN108600701A (en) * | 2018-05-02 | 2018-09-28 | 广州飞宇智能科技有限公司 | A kind of monitoring system and method judging video behavior based on deep learning |
| CN108647575A (en) * | 2018-04-10 | 2018-10-12 | 西北工业大学 | Drowned method for early warning based on optical visual analysis |
| CN108764365A (en) * | 2018-06-06 | 2018-11-06 | 国网福建省电力有限公司厦门供电公司 | A kind of device signboard detection method |
| CN109002841A (en) * | 2018-06-27 | 2018-12-14 | 淮阴工学院 | A kind of building element extracting method based on Faster-RCNN model |
| CN109033934A (en) * | 2018-05-25 | 2018-12-18 | 江南大学 | A kind of floating on water surface object detecting method based on YOLOv2 network |
| US20190012551A1 (en) * | 2017-03-06 | 2019-01-10 | Honda Motor Co., Ltd. | System and method for vehicle control based on object and color detection |
| CN109359207A (en) * | 2018-12-24 | 2019-02-19 | 焦点科技股份有限公司 | A kind of Logo detection method being easy and fast to iteration update |
| CN109460754A (en) * | 2019-01-31 | 2019-03-12 | 深兰人工智能芯片研究院（江苏）有限公司 | A kind of water surface foreign matter detecting method, device, equipment and storage medium |
| CN109637086A (en) * | 2019-01-24 | 2019-04-16 | 北京工业大学 | Alarm method and system |
-
2019
- 2019-05-15 CN CN201910404506.1A patent/CN110119718A/en active Pending
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103413114A (en) * | 2013-05-17 | 2013-11-27 | 浙江大学 | Near-drowning behavior detection method based on support vector machine |
| CN104992140A (en) * | 2015-05-27 | 2015-10-21 | 上海海事大学 | Sea surface abnormal floating object detecting method based on remote sensing image |
| CN106022230A (en) * | 2016-05-11 | 2016-10-12 | 太原理工大学 | Video-based detection method for drowning event in swimming pool |
| US20190012551A1 (en) * | 2017-03-06 | 2019-01-10 | Honda Motor Co., Ltd. | System and method for vehicle control based on object and color detection |
| CN108509860A (en) * | 2018-03-09 | 2018-09-07 | 西安电子科技大学 | HOh Xil Tibetan antelope detection method based on convolutional neural networks |
| CN108647575A (en) * | 2018-04-10 | 2018-10-12 | 西北工业大学 | Drowned method for early warning based on optical visual analysis |
| CN108600701A (en) * | 2018-05-02 | 2018-09-28 | 广州飞宇智能科技有限公司 | A kind of monitoring system and method judging video behavior based on deep learning |
| CN109033934A (en) * | 2018-05-25 | 2018-12-18 | 江南大学 | A kind of floating on water surface object detecting method based on YOLOv2 network |
| CN109460753A (en) * | 2018-05-25 | 2019-03-12 | 江南大学 | A method of detection over-water floats |
| CN108764365A (en) * | 2018-06-06 | 2018-11-06 | 国网福建省电力有限公司厦门供电公司 | A kind of device signboard detection method |
| CN109002841A (en) * | 2018-06-27 | 2018-12-14 | 淮阴工学院 | A kind of building element extracting method based on Faster-RCNN model |
| CN109359207A (en) * | 2018-12-24 | 2019-02-19 | 焦点科技股份有限公司 | A kind of Logo detection method being easy and fast to iteration update |
| CN109637086A (en) * | 2019-01-24 | 2019-04-16 | 北京工业大学 | Alarm method and system |
| CN109460754A (en) * | 2019-01-31 | 2019-03-12 | 深兰人工智能芯片研究院（江苏）有限公司 | A kind of water surface foreign matter detecting method, device, equipment and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| JOSEPH REDMON ET AL.: "YOLO9000: Better, Faster, Stronger", 《HTTPS://ARXIV.ORG/ABS/1612.08242》 * |
| MORTEN B. JENSEN ET AL.: "Swimming Pool Occupancy Analysis using Deep Learning on Low Quality Video", 《IN PROCEEDINGS OF THE 1ST INTERNATIONAL WORKSHOP ON MULTIMEDIA CONTENT ANALYSIS IN SPORTS》 * |
| 高宗 等: "基于YOLO网络的行人检测方法", 《计算机工程》 * |
| 魏湧明 等: "基于YOLO v2的无人机航拍图像定位研究", 《激光与光电子学进展》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110569772B (en) * | 2019-08-30 | 2022-03-08 | 北京科技大学 | Method for detecting state of personnel in swimming pool |
| CN110569772A (en) * | 2019-08-30 | 2019-12-13 | 北京科技大学 | Method for detecting state of personnel in swimming pool |
| CN110619365A (en) * | 2019-09-18 | 2019-12-27 | 苏州经贸职业技术学院 | Drowning detection method |
| CN110619365B (en) * | 2019-09-18 | 2023-09-12 | 苏州经贸职业技术学院 | Method for detecting falling water |
| CN111028480A (en) * | 2019-12-06 | 2020-04-17 | 江西洪都航空工业集团有限责任公司 | Drowning detection and alarm system |
| CN111545955A (en) * | 2020-04-20 | 2020-08-18 | 华南理工大学 | Door plate welding spot identification and welding path planning method |
| CN112232299A (en) * | 2020-11-09 | 2021-01-15 | 江苏科技大学 | Automatic navigation method for rescuing drowning car based on deep learning |
| CN112232299B (en) * | 2020-11-09 | 2023-10-27 | 江苏科技大学 | Automatic navigation method for rescuing water-falling automobile based on deep learning |
| CN112489371A (en) * | 2020-11-26 | 2021-03-12 | 上海天健体育科技发展有限公司 | Swimming pool drowning prevention early warning system based on computer vision |
| CN112418181B (en) * | 2020-12-13 | 2023-05-02 | 西北工业大学 | Personnel falling water detection method based on convolutional neural network |
| CN112418181A (en) * | 2020-12-13 | 2021-02-26 | 西北工业大学 | Personnel overboard detection method based on convolutional neural network |
| CN113044184A (en) * | 2021-01-12 | 2021-06-29 | 桂林电子科技大学 | Deep learning-based water rescue robot and drowning detection method |
| CN113033478A (en) * | 2021-04-19 | 2021-06-25 | 曲阜师范大学 | Pedestrian detection method based on deep learning |
| CN113188000A (en) * | 2021-05-14 | 2021-07-30 | 太原理工大学 | Lake side downwater personnel identification rescue system and method |
| CN113188000B (en) * | 2021-05-14 | 2022-07-01 | 太原理工大学 | System and method for identifying and rescuing people falling into water beside lake |
| CN113298029A (en) * | 2021-06-15 | 2021-08-24 | 广东工业大学 | Blind person walking assisting method and system based on deep learning target detection |
| CN113610178A (en) * | 2021-08-17 | 2021-11-05 | 湖南工学院 | Inland ship target detection method and device based on video monitoring image |
| CN115937506A (en) * | 2023-03-09 | 2023-04-07 | 南京邮电大学 | Method, system, device and medium for positioning bridge side falling point hole position information |
| CN116740649A (en) * | 2023-08-07 | 2023-09-12 | 山东科技大学 | Deep learning-based real-time detection method for behavior of crewman falling into water beyond boundary |
| CN116740649B (en) * | 2023-08-07 | 2023-11-03 | 山东科技大学 | Deep learning-based real-time detection method for behavior of crewman falling into water beyond boundary |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110119718A (en) | A kind of overboard detection and Survivable Control System based on deep learning | |
| US11798132B2 (en) | Image inpainting method and apparatus, computer device, and storage medium | |
| CN109670452A (en) | Method for detecting human face, device, electronic equipment and Face datection model | |
| US9646212B2 (en) | Methods, devices and systems for detecting objects in a video | |
| CN107229929A (en) | A kind of license plate locating method based on R CNN | |
| CN108804394A (en) | A kind of construction method of city noctilucence total amount-urban population regression model | |
| KR20190066873A (en) | Method for rescue using drone and computer readable record-medium on which program for executing method therefor | |
| CN114612769B (en) | Integrated sensing infrared imaging ship detection method integrated with local structure information | |
| CN113159300A (en) | Image detection neural network model, training method thereof and image detection method | |
| CN106705849A (en) | Calibration method of linear-structure optical sensor | |
| CN107221006A (en) | A kind of communication single pipe tower slant detection method based on unmanned plane imaging platform | |
| CN101739690A (en) | Method for detecting motion targets by cooperating multi-camera | |
| CN115346177A (en) | Novel system and method for detecting target under road side view angle | |
| CN113359782A (en) | Unmanned aerial vehicle autonomous addressing landing method integrating LIDAR point cloud and image data | |
| WO2023109664A1 (en) | Monitoring method and related product | |
| CN110298281A (en) | Video structural method, apparatus, electronic equipment and storage medium | |
| CN112580558A (en) | Infrared image target detection model construction method, detection method, device and system | |
| CN113033315A (en) | Rare earth mining high-resolution image identification and positioning method | |
| CN113724259A (en) | Well lid abnormity detection method and device and application thereof | |
| KR102162065B1 (en) | Method for predicting water leisure safety and computer readable record-medium on which program for executing method therefor | |
| CN117437201A (en) | Road crack detection method based on improved YOLOv7 | |
| CN113902792A (en) | Building height detection method and system based on improved RetinaNet network and electronic equipment | |
| CN112700568B (en) | Identity authentication method, equipment and computer readable storage medium | |
| CN112818964A (en) | Unmanned aerial vehicle detection method based on FoveaBox anchor-free neural network | |
| CN116433580A (en) | Dam crack detection method based on intelligent unmanned aerial vehicle |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190813 |
|
| RJ01 | Rejection of invention patent application after publication |