CN109102678A - A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique - Google Patents
A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique Download PDFInfo
- Publication number
- CN109102678A CN109102678A CN201811003121.6A CN201811003121A CN109102678A CN 109102678 A CN109102678 A CN 109102678A CN 201811003121 A CN201811003121 A CN 201811003121A CN 109102678 A CN109102678 A CN 109102678A
- Authority
- CN
- China
- Prior art keywords
- drowned
- intelligent
- target
- algorithm
- 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
- 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
- G08B21/088—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 by monitoring a device worn by the person, e.g. a bracelet attached to the swimmer
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
- H04N7/181—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast for receiving images from a plurality of remote sources
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/029—Location-based management or tracking services
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Emergency Management (AREA)
- Multimedia (AREA)
- Business, Economics & Management (AREA)
- Emergency Alarm Devices (AREA)
Abstract
A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique, the equipment being related to includes Intelligent bracelet module, intelligent monitoring message processing module, communication module display alarm module, Intelligent bracelet module is to carry out indoor positioning, and intelligent monitoring message processing module is to handle the video data of intelligent monitoring pick-up head acquisition, the location information and Water Depth Information of Intelligent bracelet；Communication module is to realize the information transmitting between Intelligent bracelet module, intelligent monitoring message processing module and display alarm module, specific detecting step of the invention is carried out as follows: system initialization: establishing the relationship of intelligent monitoring pick-up head image, Intelligent bracelet module positioning and swimming pool coordinate system, in the setting for carrying out drowned condition, drowned judgement is carried out by drowned detection algorithm.Its main structure is simple, and design concept is ingenious, and using effect is good, and detection accuracy is high, and maintenance cost is low, and application environment is friendly, wide market.
Description
Technical field:
The invention belongs to computer vision technique and field of video monitoring, are related to a kind of drowned behavioral value method of swimming pool,
In particular to the drowned behavioral value method of a kind of fusion UWB indoor positioning and video object detection and tracking technique.
Background technique:
Swimming pool is the infrastructure facility in each city, is the main place of people's amusement and recreation.But swimming is also deposited
Drowned danger is occurring, according to data statistics, drowning is the first cause of China teenager unexpected death.It is many to drown
Water event all occurs in public Swimming pool, even if there is the lifesaving personnel supervision of profession, but due to could not enough find to drown in time
Water person results in the generation of death incident.Foreign countries carried out grinding for the automatic context of detection of drowned behavior before more than 20 years
Study carefully.In the prior art, some technical solutions propose the movement speed that swimmer's strokes are detected by installation ultrasonic array
Rate, as the signal that may be found that drowned behavior；It proposes to utilize ultrasonic radar there are also some technical solutions, detects swimming pool bottom
Whether the object that not moves is occurred to determine whether drowning incident has occurred；Furthermore French Poseidon Technology
It has developed first set in the world and is based on computer assisted drowned behavioral value system.The system passes through on the water-bed and water surface
Side's installation camera, tests and analyzes the behavior of crowd in swimming pool, if noting abnormalities involved party, i.e. drowning person just passes through display
It is sounded an alarm with alarm bell.Public Swimming pool of the system in American-European multiple countries and regions installs application for many years, and
And help to find multiple row drowning person, save the life of more people.But there is also many drawbacks for the system, for example, the system can only
The drowning person for the later stage that finds to have drowned is detected, drowning person's life at this time is precarious, meanwhile, which leads to
It crosses and is mounted on water-bed camera realization monitoring, not only maintenance cost is high for such scheme, but also the accuracy rate monitored is easy to
By the interference of other objects.Application No. is the Chinese patent of 201110448257.X disclose it is a kind of entitled based on video
The patent of invention of early stage drowned behavior act detection method, which disclose a kind of drowned rows using Hidden Markov Model
For detection method, maintenance cost is lower, but this method early period is trained and modeler model process is complicated, real in practical implementation
When property is simultaneously bad, while the accuracy detected is not high, and false detection rate is high.Therefore, research and development are a set of can find to detect drowned thing automatically
The system of part has great realistic meaning to find that drowning person saves life in time.Design is researched and developed in a kind of room fusion UWB calmly
The drowned behavioral value method of position and video object detection and tracking technique.
Summary of the invention:
It is an object of the invention to overcome shortcoming and deficiency present on the prior art, seek to provide a kind of room fusion UWB
The drowned behavioral value method of interior positioning and video object detection and tracking technique.
To achieve the goals above, a kind of fusion UWB indoor positioning of the present invention and video object detection and tracking
The drowned behavioral value method of technology is realized by following technical solution: the present invention includes Intelligent bracelet module, intelligence prison
Message processing module, communication module and display alarm module, Intelligent bracelet module is controlled to carry out simultaneously to carry out indoor positioning
The depth of water detects and issues stroboscopic lamp signal and alarms, and Intelligent bracelet module is provided with to carry out information transmitting and processing
MCU (micro-control unit) is additionally provided with and passes through with the UWB positioning label, pressure sensor and stroboscopic lamp, MCU of MCU communication connection
Its pin is also connected with alarm module, i.e. sudden strain of a muscle frequency lamp, to carry out alarming drowning；Pressure sensor becomes to measure water pressure data
Dynamic frequency, and water depth pressure variable signal can be converted to exportable electric signal, and by the electric signal transmission to MCU into
Row data processing, then label is positioned by UWB and is sent to the base station UWB；UWB positions position of the label to obtain bracelet user
Data；Intelligent bracelet module can be realized accurate indoor positioning, while the video detection result with intelligent machine monitoring camera
Matching realizes personnel to have verified whether that personnel do not wear bracelet, while the characteristics of realize using mono- one tag ID of people of UWB
Tracking；Intelligent monitoring message processing module is informix and Intelligent treatment host, is adopted to handle intelligent monitoring pick-up head
The video data of collection, the location information of Intelligent bracelet and Water Depth Information, and the swimming personnel in real-time detection and tracking swimming pool, root
Drowned judgement is carried out according to the feature of extraction and the condition of setting；Intelligent monitoring message processing module also communication link is connected to display,
The Intelligent bracelet that informix and Intelligent treatment host pass over and the data that intelligent monitoring pick-up head transmits can be carried out
Display monitoring and processing, display can show the real-time monitoring data of multiple intelligent monitoring pick-up heads, and show people in swimming pool
The detection and tracking situation of member；Intelligent monitoring message processing module further includes having informix to connect with Intelligent treatment main-machine communication
Intelligent machine monitoring camera, installation swimming pool above, informix is sent for monitoring data by cable or wifi in real time
On Intelligent treatment host, informix and Intelligent treatment host by algorithm software intelligent recognition and track the swimming in swimming pool
Personnel, and judge whether there is drowned movement of struggling；Alarm module includes warning device, and with informix and intelligence at
Main-machine communication connection is managed, informix can send warning message after handling and finding dangerous situation with Intelligent treatment host
Progress and alarm to warning device；Communication module includes multi-layer switches, WiFi equipment and bluetooth equipment, to reality
Information between existing Intelligent bracelet module, intelligent monitoring message processing module and display alarm module is transmitted, wherein Intelligent bracelet
Module and intelligent monitoring message processing module by way of WiFi and Bluetooth communication with informix and Intelligent treatment host
Communication connection.
Further, the best mode of intelligent monitoring pick-up head is as follows, is mounted on the surface wall of swimming pool toward nutation
Depending on camera quantity depends on size, the Aspect Ratio, wall top height, camera FOV parameter of swimming pool, to cover comprehensively
Swimming pool is best, and guarantees that human body is imaged not less than 500 pixels, installation number most preferably 2-3 camera, such as in the picture
Fruit cannot be mounted on whole top, then be installed to the eminence of swimming pool surroundings wall；The quantity of informix and Intelligent treatment host with
Camera quantity is the same, using One-to-one communication, to guarantee enough computing capabilitys.
Specific detecting step of the invention is carried out as follows:
S1, system initialization:
A, the relationship for establishing intelligent monitoring pick-up head image, Intelligent bracelet module positioning and swimming pool coordinate system, completes hardware
The configuration of environment, parameter needed for intelligent monitoring pick-up head are as follows: focal length, FOV, camera heights (with water surface vertical range), pose
Parameter also needs the distortion parameter (radial tangential) for obtaining camera to carry out pattern distortion correction；Calculation is as follows:
Known quantity: camera height H
The corresponding world coordinate point in image coordinate center and camera distance O on the y axis3M
The image coordinate O of optical center point1(ucenter, vcenter)
Measure the image coordinate P of pixel1(u, 0), Q1(u, v)
The length xpix of actual pixels
The width ypix of actual pixels
Camera focal length f
(y-axis direction calculating is identical with a upper model, and x-axis calculating is that y-axis coordinate is calculated by ratio)
β=α-γ
It can be obtained by the coordinate Y=O of vertical direction in this way3P
ByIt obtains
It can be obtained by the coordinate X=PQ of vertical direction in this way
The calibration that intelligent monitoring pick-up head image is completed by above-mentioned calculating, to correct distortion and solve in camera
Ginseng, uses method for existing Zhang Zhengyou calibration method；Intelligent bracelet module positioning and demarcating is carried out again, using swimming pool as establishment of coordinate system
Positioning system, the image coordinate that intelligent monitoring pick-up head obtains then use monocular triangle analogue method ranging to swimming pool coordinate, swim
Swimming personnel detection and contours extract use Mask-RCNN algorithm, and multiple target tracking uses Kalman filter algorithm, position matching
Using Hungary Algorithm, using the drowned Feature extraction~+ threshold decision of image, UWB no signal is drowned to be sentenced the drowned judgement of struggle movement
Disconnected to use UWB certain time no signal+video detection less than algorithm, hydraulic pressure is drowned to judge that certain hydraulic pressure is more than certain time+sudden strain of a muscle
Lamp inspection method of determining and calculating, above-mentioned algorithm pass through host and are judged；
Table 1
| Process project | Using algorithm |
| Camera calibration | Zhang Zhengyou calibration method |
| Image coordinate is to swimming pool coordinate | Monocular triangle analogue method ranging |
| Swimming personnel detection and contours extract | Mask-RCNN |
| Multiple target tracking | Kalman filter |
| Position matching | Hungary Algorithm |
| The drowned judgement of struggle movement | Image is drowned Feature extraction~+ threshold decision |
| The drowned judgement of UWB no signal | UWB certain time no signal+video detection less than |
| The drowned judgement of hydraulic pressure | Certain hydraulic pressure is more than certain time+flashing light detection |
The setting of S2, drowned condition:
As shown in table 2, by intelligent monitoring pick-up head obtain swimming pool in and periphery personnel following behavior and location information
And record, specifically include: uprightly, speed is slow, area change is larger, red light: 1 meter of the depth of water or less is more than threshold time, amber light:
0.5-1 meters of the depth of water, more than threshold time, into profundal zone, video detection less than people；It is obtained and is recorded as follows by Intelligent bracelet
Information: it is more than threshold time, UWB signal frequency in certain threshold range that UWB signal, which disappears, and above- mentioned information are transmitted to host
Place is judged according to whether following condition is drowned to someone by host, respectively reaches following several conditions and be then judged to drowning
Water, host control warning device and issue drowned signal, condition 1 of drowning: while satisfaction is upright, speed is slow, area change is larger,
Into profundal zone, UWB signal frequency in certain threshold range；Drowned condition 2:UWB blackout is more than threshold time, enters
Profundal zone, video detection are less than people；Drowned condition 3:UWB blackout is more than threshold time, red light: 1 meter of the depth of water or less is more than
Threshold time, into profundal zone, video detection less than people；Respectively reaching following several conditions, then host control warning device carries out
Warning, warning condition 1:UWB blackout are more than threshold time, amber light: 0.5-1 meters of the depth of water, be more than threshold time；Warning condition
2: amber light: 0.5-1 meters of the depth of water are more than threshold time, video detection less than people；
Table 1 is drowned condition
UWB signal disappears in Intelligent bracelet then carries out the setting of extinction time threshold value by MCU, is then more than time threshold
One of the condition that drowning danger occurs or needs to alert；
Wherein, red light: 1 meter of the depth of water or less is more than threshold time and amber light: 0.5-1 meters of the depth of water, being passed through more than threshold time
If under type judges, stroboscopic lamp threshold value is set by MCU, and sets corresponding according to the depth of water measured and opens stroboscopic lamp,
Detect when the depth of water is more than 1 meter then blinking red lamp, then amber light flashes at 0.5-1.0 meters of the depth of water, then passes through intelligent monitoring camera shooting
Head captures the color for determining lamp；
After occurring to drown, drowning person usually has following characteristics: first is that movement speed is slow, second is that area change is big, and three
It is in upright state, features described above passes through proprietary front and back state in the video data that obtains to intelligent monitoring pick-up head
Variation determines that slow-paced judgement is then the pixel distance for calculating each target centroid before and after frames after being calculated；Body hangs down
Absolute value that is straight then being number of pixels before and after frames difference shared by calculating target before and after frames, area change is larger, is to calculate target most
The ratio of good matching elliptical long axis and short axle；
Table 3 is drowned judgment method and condition
S3, drowned judgement is carried out by drowned detection algorithm:
Drowned detection algorithm combining target detection algorithm, target tracking algorism and it is drowned judge algorithm for an entirety, one
It is and to position its profile for detecting and tracking the personnel in swimming pool；Second is that judge automatically detection and tracking personnel whether have it is excessive
Water or potential drowned behavior, detailed process as figure 5 illustrates, the specific process steps are as follows:
One, swimming personnel are effectively extracted by target detection and partitioning algorithm, removes spray, light, isolation side by side
Band etc. interference, swimming personnel can effectively be extracted using the method for moving object detection, but simultaneously also can by spray and every
Movement mistake from band is considered swimming personnel, surveys although can reduce this false retrieval by the methods of color and Complexion filter,
It is the easy influence by background spray color of this method, and the profile extracted is also not accurate enough, for effective solution back
Scape interference problem is taken and carries out the method for human testing frame by frame to extract target, because subsequent drowned behavior judgement needs to mention
The profile of human body is taken, therefore needs to carry out image segmentation to target after target detection；Taking into account the above, it is calculated using Mask- RCNN
Method, the algorithm will test the network integration of convolutional neural networks and image segmentation, realize and train end to end, in Mask R-
The basis of CNN algorithm: it improves and carries out target detection with FPN, and carry out semantic segmentation (additional segmentation point by addition additional branches
Branch and former detection branches not shared parameter), i.e. there are three output branch (classification, coordinate return and segmentation), tools by MaskR-CNN
There is following benefit: improving RoIpooling, makes the alignment of candidate region and convolution feature not because of quantization by bilinearity difference
And lose information；First is that MaskR-CNN will judge that classification and output template (mask) the two tasks are decoupling in segmentation；
Second is that being lost with sigmoid cooperation to rate (logistic)
Function individually handles the template of each classification, allows all categories with softmax compared to classical dividing method
The accuracy rate competed together is high；Specific step is as follows:
(1) whole picture is sent into CNN, carries out feature extraction；
(2) on the last layer convolution featuremap, ROI, about 300 suggestion windows of every picture are generated by RPN
Mouthful；
(3) the feature map (ROIAlign for making each suggestion window generate fixed size by RoIAlign layers
It is the key that generate mask prediction)
(4) obtain three output vectors, first is softmax classification, second be every one kind bounding box
It returns, third is the binary mask Mask (FCN generation) of each ROI, and experimental result is as shown in Figure 6 and Figure 7；
Two, multiple target tracking can determine relevant parameter between each frame, include target in sequence of video images
Position, speed, shape, color obtain their matching relationship, target trajectory are believed by matching between before and after frames
Breath is recorded, to realize target following；Tracking section, is Kalman filtering algorithm, and Kalman is one and linearly estimates
Calculating method can establish the relationship of interframe bboxs, and tracking is divided into 5 kinds of states: 1, fresh target occurs；2, object matching；3, target
It blocks；4, target separates；5, target disappears；
Kalman filter algorithm is that a dynamical system is described by state equation and observational equation.If linear system
State equation and observational equation be respectively as follows:
State equation:
xk=Axk-1+B·uk+wk
Observational equation:
zk=Hxk+vk
Here, xkIt is tkThe state vector that the n at moment × 1 is tieed up, zkIt is tkThe observation vector that the m at moment × 1 is tieed up, A is tk-1
Moment is to tkThe state-transition matrix at moment is tieed up for n × n, and B is that system controls matrix, ukFor the control amount of system
State equation: X (k+1)=A (K+1, K) X (K)+w (K),
Wherein X (k)=[x (k), y (k), w (k), h (k), v (k)], x, y, w, h respectively indicate the transverse and longitudinal coordinate of bboxs,
Length and width；
Observational equation: Z (k)=H (k) X (k)+v (k)
Wherein, w (k), v (k), incoherent white Gaussian noise；
The tracking of Kalman filter realization moving target can be used by having defined observational equation and state equation later,
Steps are as follows:
1) characteristic information of moving target, including movement mass center and boundary rectangle are calculated)；
2) with obtained characteristic information initialized card Thalmann filter, when beginning, can be initially 0；
3) target area corresponding in next frame is predicted with Kalman filter, when next frame arrives, pre-
It surveys and carries out object matching in region；
4) if successful match updates Kalman filter；
Hungary matching algorithm is that the moving object detected in new frame picture is matched to corresponding track here,
Matching process is realized by minimizing the sum of Euclidean distance between the mass center that Kalman Prediction obtains and the mass center detected
's；
Two steps can be specifically divided into:
A) loss matrix is calculated, size is [M N], wherein M is trace number, and N is the moving object number detected；
B) loss matrix is solved
Specific matching algorithm is as follows:
1) characteristic information (mass center, bounding box) of moving target is calculated；
2) remember existing track number M, initial value 0；
3) target area corresponding in next frame is predicted, when next frame arrives, mesh is carried out in estimation range
Mark matching；
If 4) successful match updates track, during matched, Kuhn-Munkres (KM) is used to calculate
The suspected target detected in new frame picture is matched to corresponding track by method, and matching process is measured in advance by minimizing
To mass center and the mass center that detects between the sum of Euclidean distance realize, can specifically be divided into two steps:
1) loss matrix is calculated, size is [M N], wherein M is trace number, and N is the moving object number detected；
2) loss matrix is solved
KM algorithm can be summarized in following steps:
A) feasible mark post is initialized；
B) Perfect matching is found with Hungary Algorithm；
C) feasible mark post is modified if not finding Perfect matching；
D) Perfect matching of (b), (c) until finding equal subgraph is repeated.
Compared with prior art, the present invention what is obtained has the beneficial effect that:
1, strong anti-interference performance: the present invention is not by high-voltage line, mobile phone signal, and the interference of the electromagnetic signals such as radio station will not be done
Above-mentioned signal is disturbed, suitable environment is good；
2, power system capacity is big: single region can be compatible with 5000 positioning labels, and maximum can support more than 10,000；
3, power is very small: positioning label 1Hz refresh rate power consumption 0.6mW, locating base station power consumption 5W, energy consumption pole
It is low；
4, positioning accuracy is high: positioning accuracy reaches as high as 2 centimetres, 10-20 centimetres of usual precision；
5, penetration capacity is strong: plank, and glass etc., which blocks, does not influence positioning accuracy, wears solid wall and sees 50% left side of signal decaying
The right side, positioning accuracy are deteriorated.
6, its main structure is simple, and design concept is ingenious, and using effect is good, and detection accuracy is high, and maintenance cost is low, application
It is environmental-friendly, wide market.
Figure of description:
Fig. 1 is main structure schematic illustration of the present invention.
Fig. 2 is placement scheme design diagram of the present invention.
Fig. 3 is drowned decision flow chart of the present invention.
Fig. 4 is network simplified diagram of the present invention.
Fig. 5 is video processing procedure schematic illustration of the present invention.
Fig. 6 is the effect schematic illustration that the present invention handles video.
Fig. 7 is the effect schematic illustration that the present invention handles video.
Fig. 8 is Hungary Algorithm flow chart of the present invention.
Specific embodiment:
The present invention is further described by way of example and in conjunction with the accompanying drawings.
Embodiment 1:
The drowned behavior of the detection of a kind of fusion UWB indoor positioning and video object and tracking technique that the present embodiment is related to is examined
Survey method, is achieved through the following technical solutions:
The present embodiment includes Intelligent bracelet module, intelligent monitoring message processing module, communication module and display alarm mould
Block, Intelligent bracelet module carry out depth of water detection and issue stroboscopic lamp signal alarming to carry out indoor positioning, intelligence
Bracelet module is provided with the MCU (micro-control unit) to carry out information transmitting and processing, is additionally provided with and MCU communication connection
UWB positioning label, pressure sensor and stroboscopic lamp, MCU are also connected with alarm module by its pin, that is, dodge frequency lamp, into
Row alarming drowning；Pressure sensor can convert water depth pressure variable signal to measure the frequency of water pressure data variation
For exportable electric signal, and the electric signal transmission to MCU is subjected to data processing, then UWB is sent to by UWB positioning label
Base station；UWB positions position data of the label to obtain bracelet user；Intelligent bracelet module can be realized accurate indoor fixed
Position, while being matched with the video detection result of intelligent machine monitoring camera, to have verified whether that personnel do not wear bracelet, simultaneously
The characteristics of realizing using mono- one tag ID of people of UWB, realizes the tracking of personnel；Intelligent monitoring message processing module is informix
With Intelligent treatment host, to handle intelligent monitoring pick-up head acquisition video data, Intelligent bracelet location information and the depth of water
Information, and the swimming personnel in real-time detection and tracking swimming pool carry out drowned judgement according to the feature of extraction and the condition of setting；
Intelligent monitoring message processing module also communication link is connected to display, can pass over informix and Intelligent treatment host
The data that Intelligent bracelet and intelligent monitoring pick-up head transmit carry out display monitoring and processing, and display can show multiple intelligence prisons
The real-time monitoring data of camera is controlled, and shows the detection and tracking situation of personnel in swimming pool；Intelligent monitoring message processing module
Further include the intelligent machine monitoring camera for thering is informix to connect with Intelligent treatment main-machine communication, above installation swimming pool, passes through net
Line or wifi send monitoring data in real time on informix and Intelligent treatment host, informix and Intelligent treatment host
By the swimming personnel in algorithm software intelligent recognition and tracking swimming pool, and judge whether there is drowned movement of struggling；Alarm mould
Block includes warning device, and is connect with informix with Intelligent treatment main-machine communication, informix and Intelligent treatment host
Warning message can be sent to progress and alarm at warning device after handling and finding dangerous situation；Communication module includes
There are multi-layer switches, WiFi equipment and bluetooth equipment, to realize Intelligent bracelet module, intelligent monitoring message processing module and show
Show the information transmitting between alarm module, wherein Intelligent bracelet module and intelligent monitoring message processing module pass through WiFi and indigo plant
The mode of tooth communication is connect with informix with Intelligent treatment main-machine communication.
Further, the best mode of intelligent monitoring pick-up head is as follows, is mounted on the surface wall of swimming pool toward nutation
Depending on camera quantity depends on size, the Aspect Ratio, wall top height, camera FOV parameter of swimming pool, to cover comprehensively
Swimming pool is best, and guarantees that human body is imaged not less than 500 pixels, installation number most preferably 2-3 camera, such as in the picture
Fruit cannot be mounted on whole top, then be installed to the eminence of swimming pool surroundings wall；The quantity of informix and Intelligent treatment host with
Camera quantity is the same, using One-to-one communication, to guarantee enough computing capabilitys.
The specific detecting step of the present embodiment is carried out as follows:
S1, system initialization:
A, the relationship for establishing intelligent monitoring pick-up head image, Intelligent bracelet module positioning and swimming pool coordinate system, completes hardware
The configuration of environment, required parameter are as follows:
Table 1
| Parameter | Effect |
| Focal length | Calculate swimming pool coordinate |
| FOV | Calculate swimming pool coordinate |
| Camera heights (with water surface vertical range) | Calculate swimming pool coordinate |
| Pose parameter | Calculate swimming pool coordinate |
| Distortion parameter (radial tangential) | Pattern distortion correction |
Known heavy: camera height H
The corresponding world coordinate point in image coordinate center and camera distance O on the y axis3M
The image coordinate O of optical center point1(ucenter, vcenter)
Measure the image coordinate P of pixel1(u, 0), Q1(u, v)
The length xpix of actual pixels
The width ypix of actual pixels
Camera focal length f
(y-axis direction calculating is identical with a upper model, and x-axis calculating is that y-axis coordinate is calculated by ratio)
β=α-γ
It can be obtained by the coordinate Y=O of vertical direction in this way3P
ByIt obtains
It can be obtained by the coordinate X=PQ of vertical direction in this way
The calibration that intelligent monitoring pick-up head image is completed by above-mentioned calculating, to correct distortion and solve in camera
Ginseng, uses method for existing Zhang Zhengyou calibration method；Intelligent bracelet module positioning and demarcating is being carried out, using swimming pool as establishment of coordinate system
Positioning system, the image coordinate that intelligent monitoring pick-up head obtains then use monocular triangle analogue method ranging to swimming pool coordinate, swim
Swimming personnel detection and contours extract use Mask-RCNN algorithm, and multiple target tracking uses Kalman filter algorithm, position matching
Using Hungary Algorithm, using the drowned Feature extraction~+ threshold decision of image, UWB no signal is drowned to be sentenced the drowned judgement of struggle movement
Disconnected to use UWB certain time no signal+video detection less than algorithm, hydraulic pressure is drowned to judge that certain hydraulic pressure is more than certain time+sudden strain of a muscle
Lamp inspection method of determining and calculating, above-mentioned algorithm pass through host and are judged；
Table 2
| Process project | Using algorithm |
| Camera calibration | Zhang Zhengyou calibration method |
| Image coordinate is to swimming pool coordinate | Monocular triangle analogue method ranging |
| Swimming personnel detection and contours extract | Mask-RCNN |
| Multiple target tracking | Kalman filter |
| Position matching | Hungary Algorithm |
| The drowned judgement of struggle movement | Image is drowned Feature extraction~+ threshold decision |
| The drowned judgement of UWB no signal | UWB certain time no signal+video detection less than |
| The drowned judgement of hydraulic pressure | Certain hydraulic pressure is more than certain time+flashing light detection |
The setting of S2, drowned condition:
As shown in table 2, by intelligent monitoring pick-up head obtain swimming pool in and periphery personnel following behavior and location information
And record, specifically include: uprightly, speed is slow, area change is larger, red light: 1 meter of the depth of water or less is more than threshold time, amber light:
0.5-1 meters of the depth of water, more than threshold time, into profundal zone, video detection less than people；It is obtained and is recorded as follows by Intelligent bracelet
Information: it is more than threshold time, UWB signal frequency in certain threshold range that UWB signal, which disappears, and above- mentioned information are transmitted to host
Place is judged according to whether following condition is drowned to someone by host, respectively reaches following several conditions and be then judged to drowning
Water, host control warning device and issue drowned signal, condition 1 of drowning: while satisfaction is upright, speed is slow, area change is larger,
Into profundal zone, UWB signal frequency in certain threshold range；Drowned condition 2:UWB blackout is more than threshold time, enters
Profundal zone, video detection are less than people；Drowned condition 3:UWB blackout is more than threshold time, red light: 1 meter of the depth of water or less is more than
Threshold time, into profundal zone, video detection less than people；Respectively reaching following several conditions, then host control warning device carries out
Warning, warning condition 1:UWB blackout are more than threshold time, amber light: 0.5-1 meters of the depth of water, be more than threshold time；Warning condition
2: amber light: 0.5-1 meters of the depth of water are more than threshold time, video detection less than people；
Table 2 is drowned condition
UWB signal disappears in Intelligent bracelet then carries out the setting of extinction time threshold value by MCU, is then more than time threshold
One of the condition that drowning danger occurs or needs to alert；
Wherein, red light: 1 meter of the depth of water or less is more than threshold time and amber light: 0.5-1 meters of the depth of water, being passed through more than threshold time
If under type judges, stroboscopic lamp threshold value is set by MCU, and sets corresponding according to the depth of water measured and opens stroboscopic lamp,
Detect when the depth of water is more than 1 meter then blinking red lamp, then amber light flashes at 0.5-1.0 meters of the depth of water, then passes through intelligent monitoring camera shooting
Head captures the color for determining lamp；
After occurring to drown, drowning person usually has following characteristics: first is that movement speed is slow, second is that area change is big, and three
It is in upright state, features described above passes through proprietary front and back state in the video data that obtains to intelligent monitoring pick-up head
Variation determines that slow-paced judgement is then the pixel distance for calculating each target centroid before and after frames after being calculated；Body hangs down
Absolute value that is straight then being number of pixels before and after frames difference shared by calculating target before and after frames, area change is larger, is to calculate target most
The ratio of good matching elliptical long axis and short axle；
Table 3 is drowned judgment method and condition
S3, drowned judgement is carried out by drowned detection algorithm:
Drowned detection algorithm combining target detection algorithm, target tracking algorism and it is drowned judge algorithm for an entirety, one
It is and to position its profile for detecting and tracking the personnel in swimming pool；Second is that judge automatically detection and tracking personnel whether have it is excessive
Water or potential drowned behavior, detailed process as figure 5 illustrates, the specific process steps are as follows:
One, swimming personnel are effectively extracted by target detection and partitioning algorithm, removes spray, light, isolation side by side
Band etc. interference, swimming personnel can effectively be extracted using the method for moving object detection, but simultaneously also can by spray and every
Movement mistake from band is considered swimming personnel, surveys although can reduce this false retrieval by the methods of color and Complexion filter,
It is the easy influence by background spray color of this method, and the profile extracted is also not accurate enough, for effective solution back
Scape interference problem is taken and carries out the method for human testing frame by frame to extract target, because subsequent drowned behavior judgement needs to mention
The profile of human body is taken, therefore needs to carry out image segmentation to target after target detection；Taking into account the above, it is calculated using Mask- RCNN
Method, the algorithm will test the network integration of convolutional neural networks and image segmentation, realize and train end to end, as shown in figure 4,
On the basis of Mask R-CNN algorithm: improving and carry out target detection with FPN, and carry out semantic segmentation by addition additional branches
(additional segmentation branch and former detection branches not shared parameter), i.e. MaskR-CNN there are three output branch (classification, coordinate return,
And segmentation), it has the advantage that: improving RoIpooling, pair of candidate region and convolution feature is made by bilinearity difference
Together information is not lost because of quantization；First is that in segmentation, MaskR-CNN will judge classification and output template (mask) the two
It is engaged in decoupling；Second is that individually being handled with sigmoid cooperation template of rate (logistic) loss function to each classification, compare
In the accuracy rate height that classical dividing method allows all categories to compete together with softmax；Specific step is as follows:
(1) whole picture is sent into CNN, carries out feature extraction；
(2) on the last layer convolution featuremap, ROI, about 300 suggestion windows of every picture are generated by RPN
Mouthful；
(3) the feature map (ROIAlign for making each suggestion window generate fixed size by RoIAlign layers
It is the key that generate mask prediction)
(4) obtain three output vectors, first is softmax classification, second be every one kind bounding box
It returns, third is the binary mask Mask (FCN generation) of each ROI, and experimental result is as shown in Figure 6 and Figure 7；
Two, multiple target tracking can determine relevant parameter between each frame, include target in sequence of video images
Position, speed, shape, color obtain their matching relationship, target trajectory are believed by matching between before and after frames
Breath is recorded, to realize target following；Tracking section, is Kalman filtering algorithm, and Kalman is one and linearly estimates
Calculating method can establish the relationship of interframe bboxs, and tracking is divided into 5 kinds of states:
1, fresh target occurs；2, object matching；3, target occlusion；4, target separates；5, target disappears；
Kalman filter algorithm is that a dynamical system is described by state equation and observational equation.If linear system
State equation and observational equation be respectively as follows:
State equation:
xk=Axk-1+B·uk+wk
Observational equation:
zk=Hxk+vk
Here, xkIt is tkThe state vector that the n at moment × 1 is tieed up, zkIt is tkThe observation vector that the m at moment × 1 is tieed up, A is tk-1
Moment is to tkThe state-transition matrix at moment is tieed up for n × n, and B is that system controls matrix, ukFor the control amount of system
State equation: X (k+1)=A (K+1, K) X (K)+w (K),
Wherein X (k)=[x (k), y (k), w (k), h (k), v (k)], x, y, w, h respectively indicate the transverse and longitudinal coordinate of bboxs,
Length and width；
Observational equation: Z (k)=H (k) X (k)+v (k)
Wherein, w (k), v (k), incoherent white Gaussian noise；
The tracking of Kalman filter realization moving target can be used by having defined observational equation and state equation later,
Steps are as follows:
1) characteristic information of moving target, including movement mass center and boundary rectangle are calculated)；
2) with obtained characteristic information initialized card Thalmann filter, when beginning, can be initially 0；
3) target area corresponding in next frame is predicted with Kalman filter, when next frame arrives, pre-
It surveys and carries out object matching in region；
4) if successful match updates Kalman filter；
Hungary matching algorithm is that the moving object detected in new frame picture is matched to corresponding track here,
Matching process is realized by minimizing the sum of Euclidean distance between the mass center that Kalman Prediction obtains and the mass center detected
's；
Two steps can be specifically divided into:
A) loss matrix is calculated, size is [M N], wherein M is trace number, and N is the moving object number detected；
B) loss matrix is solved
Specific matching algorithm is as follows:
1) characteristic information (mass center, bounding box) of moving target is calculated；
2) remember existing track number M, initial value 0；
3) target area corresponding in next frame is predicted, when next frame arrives, mesh is carried out in estimation range
Mark matching；
If 4) successful match updates track, during matched, Kuhn-Munkres (KM) is used to calculate
The suspected target detected in new frame picture is matched to corresponding track by method, and matching process is measured in advance by minimizing
To mass center and the mass center that detects between the sum of Euclidean distance realize, can specifically be divided into two steps:
1) loss matrix is calculated, size is [M N], wherein M is trace number, and N is the moving object number detected；
2) loss matrix is solved
KM algorithm can be summarized in following steps:
A) feasible mark post is initialized；
B) Perfect matching is found with Hungary Algorithm；
C) feasible mark post is modified if not finding Perfect matching；
D) Perfect matching of (b), (c) until finding equal subgraph is repeated.
Claims (3)
1. a kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique, feature exist
It is realized in by following equipment, includes Intelligent bracelet module, intelligent monitoring message processing module, communication module and display report
Alert module, Intelligent bracelet module carry out depth of water detection and issue stroboscopic lamp signal alarming to carry out indoor positioning,
Intelligent bracelet module is provided with the MCU to carry out information transmitting and processing, is additionally provided with and positions with the UWB of MCU communication connection
Label, pressure sensor and stroboscopic lamp, MCU are also connected with alarm module by its pin, i.e. sudden strain of a muscle frequency lamp, to carry out drowned report
It is alert；Pressure sensor can convert water depth pressure variable signal to exportable to measure the frequency of water pressure data variation
Electric signal, and by the electric signal transmission to MCU carry out data processing, then by UWB positioning label be sent to the base station UWB；UWB
Position position data of the label to obtain bracelet user；Intelligent bracelet module can be realized accurate indoor positioning, simultaneously
It is matched with the video detection result of intelligent machine monitoring camera, to have verified whether that personnel do not wear bracelet, while realizing benefit
The characteristics of with mono- one tag ID of people of UWB, realize the tracking of personnel；Intelligent monitoring message processing module is at informix and intelligence
Host is managed, to handle the video data of intelligent monitoring pick-up head acquisition, the location information and Water Depth Information of Intelligent bracelet, and it is real
When detection and tracking swimming pool in swimming personnel, drowned judgement is carried out according to the feature of extraction and the condition of setting；Intelligent monitoring
Message processing module also communication link is connected to display, the Intelligent bracelet that can pass over informix and Intelligent treatment host
Multiple intelligent monitoring pick-up heads can be shown by carrying out display monitoring and processing, display with the data that intelligent monitoring pick-up head transmits
Real-time monitoring data, and show the detection and tracking situation of personnel in swimming pool；Intelligent monitoring message processing module further includes having
The intelligent machine monitoring camera that informix is connect with Intelligent treatment main-machine communication, above installation swimming pool, by cable or
Wifi sends monitoring data in real time on informix and Intelligent treatment host, and informix and Intelligent treatment host pass through calculation
Swimming personnel in method software intelligent recognition and tracking swimming pool, and judge whether there is drowned movement of struggling；Alarm module includes
There is warning device, and connect with informix with Intelligent treatment main-machine communication, informix is being handled with Intelligent treatment host
And warning message can be sent to progress and alarm at warning device after finding dangerous situation；Communication module includes routing
Interchanger, WiFi equipment and bluetooth equipment, to realize Intelligent bracelet module, intelligent monitoring message processing module and display alarm
Information transmitting between module, wherein Intelligent bracelet module and intelligent monitoring message processing module pass through WiFi and Bluetooth communication
Mode connect with informix with Intelligent treatment main-machine communication.
2. the drowned row of a kind of fusion UWB indoor positioning and video object detection and tracking technique according to claim 1
For detection method, it is characterised in that the best mode of intelligent monitoring pick-up head is as follows, is mounted on past on the surface wall of swimming pool
Lower vertical view, camera quantity depends on size, the Aspect Ratio, wall top height, camera FOV parameter of swimming pool, with comprehensive
It is best for covering swimming pool, and guarantees that human body is imaged in the picture not less than 500 pixels, installation number most preferably 2-3 camera shooting
Head is installed to the eminence of swimming pool surroundings wall if whole top cannot be mounted on；The number of informix and Intelligent treatment host
Amount is as camera quantity, using One-to-one communication, to guarantee enough computing capabilitys.
3. the drowned row of a kind of fusion UWB indoor positioning and video object detection and tracking technique according to claim 1
For detection method, it is characterised in that specific detecting step is carried out as follows:
S1, system initialization:
A, the relationship for establishing intelligent monitoring pick-up head image, Intelligent bracelet module positioning and swimming pool coordinate system, completes hardware environment
Configuration, parameter needed for intelligent monitoring pick-up head is as follows: focal length, FOV, camera heights (with water surface vertical range), pose parameter,
Also need the distortion parameter (radial tangential) for obtaining camera to carry out pattern distortion correction；It is completed by existing calculation
The calibration of intelligent monitoring pick-up head image, to correct distortion and solve camera internal reference, use method for existing just
Friendly standardization；Intelligent bracelet module positioning and demarcating is carried out again, using swimming pool as establishment of coordinate system positioning system, intelligent monitoring pick-up head
The image coordinate of acquisition then uses monocular triangle analogue method ranging to swimming pool coordinate, and swimming personnel detection and contours extract use
Mask-RCNN algorithm, multiple target tracking use Kalman filter algorithm, and position matching uses Hungary Algorithm, and struggle movement is drowned
Water judgement uses UWB certain time no signal+view using the drowned Feature extraction~+ threshold decision of image, the drowned judgement of UWB no signal
Frequency can't detect algorithm, and hydraulic pressure is drowned to judge that certain hydraulic pressure is more than certain time+flashing light detection algorithm, and above-mentioned algorithm passes through master
Machine is judged；
The setting of S2, drowned condition:
It is obtained in swimming pool and the following behavior of periphery personnel and location information and is recorded by intelligent monitoring pick-up head, it is specific to wrap
Include: uprightly, speed is slow, area change is larger, red light: 1 meter of the depth of water or less is more than threshold time, amber light: 0.5-1 meters of the depth of water,
More than threshold time, into profundal zone, video detection less than people；It is obtained by Intelligent bracelet and records following information: UWB signal
Disappearing is more than threshold time, UWB signal frequency in certain threshold range, and above- mentioned information are transmitted at host, are pressed by host
Judge according to whether following condition is drowned to someone, respectively reaches following several conditions and be then judged to drowning, host control
Warning device issues drowned signal, condition 1 of drowning: at the same meet upright, speed is slow, area change is larger, into profundal zone,
UWB signal frequency is in certain threshold range；Drowned condition 2:UWB blackout is more than threshold time, into profundal zone, video
It can't detect people；Drowned condition 3:UWB blackout is more than threshold time, red light: 1 meter of the depth of water or less more than threshold time, into
Enter profundal zone, video detection less than people；Respectively reaching following several conditions, then host control warning device is alerted, and alerts item
Part 1:UWB blackout is more than threshold time, amber light: 0.5-1 meters of the depth of water, be more than threshold time；Warning condition 2: amber light: the depth of water
0.5-1 meters are more than threshold time, video detection less than people；
Table 1 is drowned condition
UWB signal disappears in Intelligent bracelet then carries out the setting of extinction time threshold value by MCU, is then to occur more than time threshold
One of the condition that drowning danger or needs alert；
Wherein, red light: 1 meter of the depth of water or less is more than threshold time and amber light: 0.5-1 meters of the depth of water, being passed through more than threshold time as follows
Mode judges, sets stroboscopic lamp threshold value by MCU, and sets corresponding according to the depth of water measured and open stroboscopic lamp, is detecting
Then blinking red lamp when to the depth of water being more than 1 meter, then amber light flashes at 0.5-1.0 meters of the depth of water, is then caught by intelligent monitoring pick-up head
Catch the color of determining lamp；
After occurring to drown, drowning person usually has following characteristics: first is that movement speed is slow, second is that area change is big, third is that place
In upright state, variation that features described above passes through proprietary front and back state in the video data that obtains to intelligent monitoring pick-up head
Determine after being calculated, slow-paced judgement is then the pixel distance for calculating each target centroid before and after frames；Vertical body is then
It is the absolute value for calculating number of pixels before and after frames difference shared by target before and after frames, area change is larger, is to calculate target best
Ratio with elliptical long axis and short axle；
S3, drowned judgement is carried out by drowned detection algorithm:
Drowned detection algorithm combining target detection algorithm, target tracking algorism and it is drowned judge algorithm for an entirety, first is that with
In detecting and track the personnel in swimming pool, and position its profile；Second is that judge automatically detection and tracking personnel whether have it is drowned or
Potential drowned behavior, detailed process as figure 5 illustrates, the specific process steps are as follows:
One, swimming personnel are effectively extracted by target detection and partitioning algorithm, side by side except spray, light, isolation strip etc.
Interference, can effectively extract swimming personnel using the method for moving object detection, but simultaneously also can be by spray and isolation strip
Movement mistake be considered swimming personnel, surveyed although this false retrieval can be reduced by the methods of color and Complexion filter, should
Method is easy to be influenced by background spray color, and the profile extracted is also not accurate enough, in order to which effective solution background is dry
Problem is disturbed, takes and carries out the method for human testing frame by frame to extract target, because subsequent drowned behavior judgement needs to extract people
The profile of body, therefore need to carry out image segmentation to target after target detection；It taking into account the above, should using Mask-RCNN algorithm
Algorithm will test the network integration of convolutional neural networks and image segmentation, realizes and trains end to end, calculate in Mask R-CNN
The basis of method: improving and carry out target detection with FPN, and by addition additional branches carry out semantic segmentation (additional segmentation branch with
Former detection branches not shared parameter), i.e. MaskR-CNN has such as there are three output branch (classification, coordinate return and segmentation)
Lower benefit: improving RoIpooling, makes the alignment of candidate region and convolution feature not because quantization is damaged by bilinearity difference
It breaks one's promise breath；First is that MaskR-CNN will judge that classification and output template (mask) the two tasks are decoupling in segmentation；Second is that
Rate (logistic) is lost with sigmoid cooperation
Function individually handles the template of each classification, makes all categories competing together with softmax compared to classical dividing method
The accuracy rate striven is high；Specific step is as follows:
(1) whole picture is sent into CNN, carries out feature extraction；
(2) on the last layer convolution featuremap, ROI, about 300 suggestion windows of every picture are generated by RPN；
(3) by the RoIAlign layers of feature map for making each suggestion window generate fixed size, (ROIAlign is to generate
The key of mask prediction)
(4) obtain three output vectors, first is softmax classification, second be every one kind bounding box recurrence,
Third is the binary mask Mask (FCN generation) of each ROI, and experimental result is as shown in Figure 6 and Figure 7；
Two, multiple target tracking can determine relevant parameter in sequence of video images between each frame, include target position,
Speed, shape, color obtain their matching relationship, target trajectory information are recorded by matching between before and after frames
Get off, to realize target following；Tracking section is Kalman filtering algorithm, and Kalman is a linear estimation algorithm,
It can establish the relationship of interframe bboxs, tracking is divided into 5 kinds of states: 1, fresh target occurs；2, object matching；3, target occlusion；4,
Target separation；5, target disappears；It can be used after having defined observational equation and state equation by Kalman filtering algorithm
Kalman filter realizes the tracking of moving target, and steps are as follows:
1) characteristic information of moving target, including movement mass center and boundary rectangle are calculated；
2) with obtained characteristic information initialized card Thalmann filter, when beginning, can be initially 0；
3) target area corresponding in next frame is predicted with Kalman filter, when next frame arrives, in Target area
Object matching is carried out in domain；
4) if successful match updates Kalman filter；
Hungary matching algorithm is that the moving object detected in new frame picture is matched to corresponding track here, matching
Process is realized by minimizing the sum of Euclidean distance between the mass center that Kalman Prediction obtains and the mass center detected；
Two steps can be specifically divided into:
A) loss matrix is calculated, size is [M N], wherein M is trace number, and N is the moving object number detected；
B) loss matrix is solved
Specific matching algorithm is as follows:
1) characteristic information (mass center, bounding box) of moving target is calculated；
2) remember existing track number M, initial value 0；
3) target area corresponding in next frame is predicted, when next frame arrives, target is carried out in estimation range
Match；
If 4) successful match updates track, during matched, Kuhn-Munkres (KM) algorithm is used, it will
The suspected target detected in new frame picture is matched to corresponding track, and matching process is the matter obtained by minimizing prediction
What the sum of Euclidean distance between the heart and the mass center detected was realized, it can specifically be divided into two steps:
1) loss matrix is calculated, size is [M N], wherein M is trace number, and N is the moving object number detected；
2) loss matrix is solved
KM algorithm can be summarized in following steps:
A) feasible mark post is initialized；
B) Perfect matching is found with Hungary Algorithm；
C) feasible mark post is modified if not finding Perfect matching；
D) Perfect matching of (b), (c) until finding equal subgraph is repeated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811003121.6A CN109102678A (en) | 2018-08-30 | 2018-08-30 | A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811003121.6A CN109102678A (en) | 2018-08-30 | 2018-08-30 | A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109102678A true CN109102678A (en) | 2018-12-28 |
Family
ID=64864342
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811003121.6A Pending CN109102678A (en) | 2018-08-30 | 2018-08-30 | A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109102678A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785363A (en) * | 2018-12-29 | 2019-05-21 | 中国电子科技集团公司第五十二研究所 | A kind of unmanned plane video motion Small object real-time detection and tracking |
| CN109902676A (en) * | 2019-01-12 | 2019-06-18 | 浙江工业大学 | A kind of separated based on dynamic background stops detection algorithm |
| CN109996182A (en) * | 2019-04-19 | 2019-07-09 | 无锡艾森汇智科技有限公司 | A kind of localization method, apparatus and system combined based on UWB positioning with monitoring |
| CN110363153A (en) * | 2019-07-16 | 2019-10-22 | 广州图普网络科技有限公司 | It paddles detection method, device, server and computer readable storage medium |
| CN110569772A (en) * | 2019-08-30 | 2019-12-13 | 北京科技大学 | Method for detecting state of personnel in swimming pool |
| CN110696003A (en) * | 2019-09-17 | 2020-01-17 | 五邑大学 | Water side rescue robot based on SLAM technology and deep learning |
| CN111222421A (en) * | 2019-12-26 | 2020-06-02 | 杭州宇泛智能科技有限公司 | Method and device for detecting personnel state in water area and electronic equipment |
| CN111462488A (en) * | 2020-04-01 | 2020-07-28 | 北京工业大学 | Intersection safety risk assessment method based on deep convolutional neural network and intersection behavior characteristic model |
| CN111667512A (en) * | 2020-05-28 | 2020-09-15 | 浙江树人学院（浙江树人大学） | Multi-target vehicle track prediction method based on improved Kalman filtering |
| CN112165600A (en) * | 2020-08-26 | 2021-01-01 | 苏宁云计算有限公司 | Drowning identification method and device, camera and computer system |
| CN112309077A (en) * | 2020-09-23 | 2021-02-02 | 苏宁云计算有限公司 | Drowning alarm method and drowning alarm device for swimming pool |
| CN112329688A (en) * | 2020-11-16 | 2021-02-05 | 广州顶新时代电子系统科技有限公司 | Intelligent risk management and control system for public places |
| CN112489372A (en) * | 2020-12-28 | 2021-03-12 | 南京大学 | Swimming pool monitoring and alarming system |
| CN112529851A (en) * | 2020-11-27 | 2021-03-19 | 中冶赛迪重庆信息技术有限公司 | Method, system, terminal and medium for determining state of hydraulic pipe |
| CN112799009A (en) * | 2021-04-13 | 2021-05-14 | 国能信控互联技术有限公司 | Personnel positioning system and positioning method based on UWB |
| CN112860946A (en) * | 2021-01-18 | 2021-05-28 | 四川弘和通讯有限公司 | Method and system for converting video image information into geographic information |
| CN113345205A (en) * | 2021-03-24 | 2021-09-03 | 蒋勇武 | Drowning detection method and device |
| CN114363949A (en) * | 2022-01-21 | 2022-04-15 | 杭州北斗时空研究院 | Abnormal data detection method for UWB positioning system |
| CN114783147A (en) * | 2022-04-19 | 2022-07-22 | 珠海市杰理科技股份有限公司 | Intelligent monitoring method and device, wearable device and readable storage medium |
| TWI780378B (en) * | 2019-11-26 | 2022-10-11 | 臺中榮民總醫院 | System and method for detecting and classifying animal behavior |
| CN115938073A (en) * | 2022-11-08 | 2023-04-07 | 湛江市钜晟科技有限公司 | Prevent drowned monitored control system with automatic alarm function |
| CN115979250A (en) * | 2023-03-20 | 2023-04-18 | 山东上水环境科技集团有限公司 | Positioning method based on UWB module, semantic map and visual information |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103021136A (en) * | 2012-12-24 | 2013-04-03 | 电子科技大学 | System and method for monitoring swimming pool drowning based on Internet of Things |
| CN104517410A (en) * | 2015-01-21 | 2015-04-15 | 苏州大学 | Intelligent drowning-prevention system and method |
| CN104524747A (en) * | 2014-12-19 | 2015-04-22 | 苏州大学 | Intelligent swimming system based on Internet of Things and artificial intelligence |
| CN104851243A (en) * | 2015-05-29 | 2015-08-19 | 付建文 | Swimming pool drowning monitoring system and method |
| CN105243792A (en) * | 2015-11-16 | 2016-01-13 | 常州机电职业技术学院 | Underwater alarm bracelet, working method thereof and underwater alarm system |
| CN102693606B (en) * | 2011-03-24 | 2016-07-06 | 中国人民解放军国防科学技术大学 | A kind of swimming pool anti-drowning method for early warning based on shooting network and system |
| CN105765637A (en) * | 2015-09-23 | 2016-07-13 | 深圳还是威健康科技有限公司 | Underwater alarming method and intelligent hand ring |
| CN205594842U (en) * | 2016-05-13 | 2016-09-21 | 西南大学 | Swimming danger souding and alarm device |
| CN106022230A (en) * | 2016-05-11 | 2016-10-12 | 太原理工大学 | Video-based detection method for drowning event in swimming pool |
| CN106912006A (en) * | 2017-04-01 | 2017-06-30 | 成都精位科技有限公司 | A kind of wisdom swimming exercise system and the method for realizing Sport Administration |
| CN107320092A (en) * | 2017-08-04 | 2017-11-07 | 山西大学 | A kind of outdoor swimming state monitoring bracelet and swimming state method for monitoring abnormality |
| CN107566797A (en) * | 2017-09-07 | 2018-01-09 | 青岛博晶微电子科技有限公司 | A kind of drowned monitor and detection device of swimming pool |
| CN107808495A (en) * | 2017-12-04 | 2018-03-16 | 电子科技大学 | A kind of swimming intelligent security monitoring system based on cordless communication network |
-
2018
- 2018-08-30 CN CN201811003121.6A patent/CN109102678A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102693606B (en) * | 2011-03-24 | 2016-07-06 | 中国人民解放军国防科学技术大学 | A kind of swimming pool anti-drowning method for early warning based on shooting network and system |
| CN103021136A (en) * | 2012-12-24 | 2013-04-03 | 电子科技大学 | System and method for monitoring swimming pool drowning based on Internet of Things |
| CN104524747A (en) * | 2014-12-19 | 2015-04-22 | 苏州大学 | Intelligent swimming system based on Internet of Things and artificial intelligence |
| CN104517410A (en) * | 2015-01-21 | 2015-04-15 | 苏州大学 | Intelligent drowning-prevention system and method |
| CN104851243A (en) * | 2015-05-29 | 2015-08-19 | 付建文 | Swimming pool drowning monitoring system and method |
| CN105765637A (en) * | 2015-09-23 | 2016-07-13 | 深圳还是威健康科技有限公司 | Underwater alarming method and intelligent hand ring |
| CN105243792A (en) * | 2015-11-16 | 2016-01-13 | 常州机电职业技术学院 | Underwater alarm bracelet, working method thereof and underwater alarm system |
| CN106022230A (en) * | 2016-05-11 | 2016-10-12 | 太原理工大学 | Video-based detection method for drowning event in swimming pool |
| CN205594842U (en) * | 2016-05-13 | 2016-09-21 | 西南大学 | Swimming danger souding and alarm device |
| CN106912006A (en) * | 2017-04-01 | 2017-06-30 | 成都精位科技有限公司 | A kind of wisdom swimming exercise system and the method for realizing Sport Administration |
| CN107320092A (en) * | 2017-08-04 | 2017-11-07 | 山西大学 | A kind of outdoor swimming state monitoring bracelet and swimming state method for monitoring abnormality |
| CN107566797A (en) * | 2017-09-07 | 2018-01-09 | 青岛博晶微电子科技有限公司 | A kind of drowned monitor and detection device of swimming pool |
| CN107808495A (en) * | 2017-12-04 | 2018-03-16 | 电子科技大学 | A kind of swimming intelligent security monitoring system based on cordless communication network |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785363A (en) * | 2018-12-29 | 2019-05-21 | 中国电子科技集团公司第五十二研究所 | A kind of unmanned plane video motion Small object real-time detection and tracking |
| CN109902676A (en) * | 2019-01-12 | 2019-06-18 | 浙江工业大学 | A kind of separated based on dynamic background stops detection algorithm |
| CN109902676B (en) * | 2019-01-12 | 2021-04-09 | 浙江工业大学 | Dynamic background-based violation detection algorithm |
| CN109996182B (en) * | 2019-04-19 | 2021-01-19 | 无锡艾森汇智科技有限公司 | Positioning method, device and system based on combination of UWB positioning and monitoring |
| CN109996182A (en) * | 2019-04-19 | 2019-07-09 | 无锡艾森汇智科技有限公司 | A kind of localization method, apparatus and system combined based on UWB positioning with monitoring |
| CN110363153A (en) * | 2019-07-16 | 2019-10-22 | 广州图普网络科技有限公司 | It paddles detection method, device, server and computer readable storage medium |
| CN110363153B (en) * | 2019-07-16 | 2022-08-02 | 广州图普网络科技有限公司 | Wading detection method and device, server and computer readable storage medium |
| CN110569772A (en) * | 2019-08-30 | 2019-12-13 | 北京科技大学 | Method for detecting state of personnel in swimming pool |
| CN110569772B (en) * | 2019-08-30 | 2022-03-08 | 北京科技大学 | Method for detecting state of personnel in swimming pool |
| CN110696003A (en) * | 2019-09-17 | 2020-01-17 | 五邑大学 | Water side rescue robot based on SLAM technology and deep learning |
| TWI780378B (en) * | 2019-11-26 | 2022-10-11 | 臺中榮民總醫院 | System and method for detecting and classifying animal behavior |
| CN111222421A (en) * | 2019-12-26 | 2020-06-02 | 杭州宇泛智能科技有限公司 | Method and device for detecting personnel state in water area and electronic equipment |
| CN111462488A (en) * | 2020-04-01 | 2020-07-28 | 北京工业大学 | Intersection safety risk assessment method based on deep convolutional neural network and intersection behavior characteristic model |
| CN111462488B (en) * | 2020-04-01 | 2021-09-10 | 北京工业大学 | Intersection safety risk assessment method based on deep convolutional neural network and intersection behavior characteristic model |
| CN111667512A (en) * | 2020-05-28 | 2020-09-15 | 浙江树人学院（浙江树人大学） | Multi-target vehicle track prediction method based on improved Kalman filtering |
| CN111667512B (en) * | 2020-05-28 | 2024-04-09 | 浙江树人学院（浙江树人大学） | Multi-target vehicle track prediction method based on improved Kalman filtering |
| CN112165600A (en) * | 2020-08-26 | 2021-01-01 | 苏宁云计算有限公司 | Drowning identification method and device, camera and computer system |
| CN112309077A (en) * | 2020-09-23 | 2021-02-02 | 苏宁云计算有限公司 | Drowning alarm method and drowning alarm device for swimming pool |
| CN112329688B (en) * | 2020-11-16 | 2021-07-06 | 广州顶新时代电子系统科技有限公司 | Intelligent risk management and control system for public places |
| CN112329688A (en) * | 2020-11-16 | 2021-02-05 | 广州顶新时代电子系统科技有限公司 | Intelligent risk management and control system for public places |
| CN112529851A (en) * | 2020-11-27 | 2021-03-19 | 中冶赛迪重庆信息技术有限公司 | Method, system, terminal and medium for determining state of hydraulic pipe |
| CN112529851B (en) * | 2020-11-27 | 2023-07-18 | 中冶赛迪信息技术(重庆)有限公司 | Hydraulic pipe state determining method, system, terminal and medium |
| CN112489372A (en) * | 2020-12-28 | 2021-03-12 | 南京大学 | Swimming pool monitoring and alarming system |
| CN112860946A (en) * | 2021-01-18 | 2021-05-28 | 四川弘和通讯有限公司 | Method and system for converting video image information into geographic information |
| CN112860946B (en) * | 2021-01-18 | 2023-04-07 | 四川弘和通讯集团有限公司 | Method and system for converting video image information into geographic information |
| CN113345205A (en) * | 2021-03-24 | 2021-09-03 | 蒋勇武 | Drowning detection method and device |
| CN113345205B (en) * | 2021-03-24 | 2024-03-08 | 蒋勇武 | Drowning detection method and device |
| CN112799009A (en) * | 2021-04-13 | 2021-05-14 | 国能信控互联技术有限公司 | Personnel positioning system and positioning method based on UWB |
| CN114363949B (en) * | 2022-01-21 | 2023-06-02 | 杭州北斗时空研究院 | Abnormal data detection method for UWB positioning system |
| CN114363949A (en) * | 2022-01-21 | 2022-04-15 | 杭州北斗时空研究院 | Abnormal data detection method for UWB positioning system |
| CN114783147A (en) * | 2022-04-19 | 2022-07-22 | 珠海市杰理科技股份有限公司 | Intelligent monitoring method and device, wearable device and readable storage medium |
| CN114783147B (en) * | 2022-04-19 | 2023-10-27 | 珠海市杰理科技股份有限公司 | Intelligent monitoring method and device, wearable device and readable storage medium |
| CN115938073A (en) * | 2022-11-08 | 2023-04-07 | 湛江市钜晟科技有限公司 | Prevent drowned monitored control system with automatic alarm function |
| CN115979250A (en) * | 2023-03-20 | 2023-04-18 | 山东上水环境科技集团有限公司 | Positioning method based on UWB module, semantic map and visual information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109102678A (en) | A kind of drowned behavioral value method of fusion UWB indoor positioning and video object detection and tracking technique | |
| CN110889350B (en) | Line obstacle monitoring and alarming system and method based on three-dimensional imaging | |
| CN106707296B (en) | It is a kind of based on the unmanned machine testing of Based on Dual-Aperture photo electric imaging system and recognition methods | |
| CN106022230B (en) | A kind of swimming pool drowning event detecting method based on video | |
| CN101493980B (en) | Rapid video flame detection method based on multi-characteristic fusion | |
| CN108062349A (en) | Video frequency monitoring method and system based on video structural data and deep learning | |
| CN106384106A (en) | Anti-fraud face recognition system based on 3D scanning | |
| CN109255298A (en) | Safety helmet detection method and system in dynamic background | |
| CN102693606A (en) | Shooting network-based method and system for warning of drowning prevention in swimming pool | |
| CN115603466B (en) | Ship shore power system based on artificial intelligence visual identification | |
| CN108986158A (en) | A kind of across the scene method for tracing identified again based on target and device and Computer Vision Platform | |
| CN112084963B (en) | Monitoring early warning method, system and storage medium | |
| CN110007366A (en) | A kind of life searching method and system based on Multi-sensor Fusion | |
| CN112200841B (en) | Cross-domain multi-camera tracking method and device based on pedestrian posture | |
| CN113963315A (en) | Real-time video multi-user behavior recognition method and system in complex scene | |
| CN114202646A (en) | Infrared image smoking detection method and system based on deep learning | |
| CN115761537B (en) | Power transmission line foreign matter intrusion identification method oriented to dynamic feature supplementing mechanism | |
| CN112270807A (en) | Old man early warning system that tumbles | |
| CN113269091A (en) | Personnel trajectory analysis method, equipment and medium for intelligent park | |
| CN110246306A (en) | Underwater exercise safety monitoring system design method based on wireless signal transmission | |
| CN113920461A (en) | Power grid operation and maintenance process image monitoring system and monitoring method | |
| CN112270381A (en) | People flow detection method based on deep learning | |
| CN109583339A (en) | A kind of ATM video brainpower watch and control method based on image procossing | |
| CN103456123A (en) | Video smoke detection method based on flowing and diffusion characters | |
| CN202033875U (en) | Swimming pool anti-drowning early warning system based on shooting network |
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: 20181228 |