CN109543553A - The photoelectricity recognition and tracking method of low small slow target based on machine learning - Google Patents
The photoelectricity recognition and tracking method of low small slow target based on machine learning Download PDFInfo
- Publication number
- CN109543553A CN109543553A CN201811273114.8A CN201811273114A CN109543553A CN 109543553 A CN109543553 A CN 109543553A CN 201811273114 A CN201811273114 A CN 201811273114A CN 109543553 A CN109543553 A CN 109543553A
- Authority
- CN
- China
- Prior art keywords
- target
- frame
- machine learning
- image
- camera
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000010801 machine learning Methods 0.000 title claims abstract description 27
- 230000005622 photoelectricity Effects 0.000 title claims abstract description 15
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 239000000945 filler Substances 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 14
- 125000004122 cyclic group Chemical group 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 7
- 230000004044 response Effects 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 abstract description 3
- 230000004927 fusion Effects 0.000 abstract description 2
- 238000005286 illumination Methods 0.000 abstract description 2
- 238000001228 spectrum Methods 0.000 abstract description 2
- 238000013527 convolutional neural network Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 244000144992 flock Species 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011410 subtraction method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/12—Fingerprints or palmprints
- G06V40/13—Sensors therefor
- G06V40/1318—Sensors therefor using electro-optical elements or layers, e.g. electroluminescent sensing
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention devises a kind of photoelectricity recognition and tracking method of low small slow target based on machine learning, it is first determined then the direction of target adjusts the deflection and pitch angle of camera, so that target is located at camera within sweep of the eye;Then camera reads in image frame by frame, the on-line checking of target identification, using the image of reading as the input of neural network, by machine learning trained network, obtains the output of network, the constraint frame of classification and position including target;If the classification of output belongs to low small slow target, enter in next step, otherwise skip in next step, directly reading next frame image, carries out target following.While guaranteeing real-time, the accuracy of automatic identification is improved, enhances the robustness to influence factors such as illumination, targeted attitudes.The present invention can be used for the imaging device of multi-spectrum fusion, extends the application range of single recognition and tracking algorithm, improves the adaptability of algorithm.
Description
Technical field
The invention belongs to the electromagnetic wave tracking system technical fields in addition to radio wave, more particularly to one kind to be based on engineering
The photoelectricity recognition and tracking method for the low small slow target practised.
Background technique
The low-flying air vehicles such as unmanned plane, cruise missile are the primary challenge modes of current air attack, and unmanned plane
Also have the characteristics that low in cost, light weight size is small, mobility is good, task can be executed under unsafe conditions, so that anti-empty set
System increased pressure seriously threatens the safety guarantee work of occasion, key area.And detect be intercept with strike necessity before
It mentions, therefore early warning for the low small slow target such as unmanned plane and identification are that war master is captured in air defense operation under the conditions of Future Information
The important leverage of dynamic power.Detection for low small slow target, mainly there is the methods of radar detection, acoustic sounding and photodetection.
For radar detection, since the radar area of low small slow target is very small, and flying speed is slow, causes
Doppler effect it is also unobvious, so conventional radar is to the Effect on Detecting of low small slow target and bad, it is blind that there are low-altitude detections
The disadvantages such as Qu great, echo is small and weak, is easy mutually to obscure with meteorological interference, noise jamming or flock of birds, and angular resolution is low.
For acoustic sounding, the noise of low small slow target is mainly that the air generated in engine noise and flight course is disturbed
Moving noise, but the flying power of low small slow target is greatly electric power at present, noise is smaller, and flying speed compared with
Slowly, so that its noise level is very low, it is difficult to detect.
So-called photodetection, the light wave for being reflected using target or directly being radiated implement the technology detected to target.Target exists
It, can be in photoelectric sensor due to the difference of geometry and reflection behavior of surface under sunlight, starlight or artificially lighting irradiation
On be rendered as certain contrast distribution image.Its advantage is that it is small in size, light weight and cost is low, anti-electromagnetic interference capability
By force, concealment is strong, angular resolution is high;Disadvantage is that the identification accuracy of traditional identification and track algorithm is lower, targeted contrast
Degree, target carriage change, lighting angle, the inhomogeneities of background and the bright dark, atmospheric turbulance of background etc. have target acquisition probability
It has a certain impact.
It is main still to use frame-to-frame differences in the photoelectricity recognition and tracking system for low small slow target announced both at home and abroad
The conventional methods such as point-score, background subtraction method, optical flow method, moment invariants.And these methods are to shadows such as lighting angle, targeted attitudes
There are significant improvement spaces in the robustness of the factor of sound.
Summary of the invention
The photoelectricity for the low small slow target based on machine learning that the technical problem to be solved by the invention is to provide a kind of identifies
Tracking, the image acquired using electro-optical system are still able to achieve in the case where target effective pixel number is seldom to low small slow
The high-precision of target quickly identifies.
The technical solution adopted by the present invention to solve the technical problems is: providing a kind of low small slow mesh based on machine learning
Target photoelectricity recognition and tracking method, this approach includes the following steps, step 1, it is first determined then the direction of target (utilizes cloud
Platform) adjustment camera deflection and pitch angle so that target is located at camera within sweep of the eye;Step 2, camera are read in frame by frame
Image;Step 3, the on-line checking of target identification, using the image of reading as the input of neural network, by machine learning
Trained good network, obtains the output of network, the constraint frame of classification and position including target;If the classification of output belongs to
Low small slow target then enters in next step, otherwise skips in next step, directly reading next frame image;Step 4, carry out target with
Track.
According to the above technical scheme, in the step 4 the following steps are included:
S1, the constraint frame provided by first frame image, expansion obtain filler constraint frame, and also just having obtained next frame target can
Estimation range existing for energy, using current filler constraint frame as positive sample, using the cyclic shift of current positive sample as negative sample
This, using the method for machine learning, training obtains a target detection classifier;
S2 equally fetters frame cyclic shift to the filler of previous frame, to obtained sample beam to next frame image
The image tied up in frame is classified, and the constraint frame of Response to selection most strong (related coefficient is maximum) is as the benefit where present frame target
White constraint frame;Change in location between frame is fettered according to the filler of present frame and previous frame, obtains the change in location of target;
S3 after obtaining present frame target exact position, obtains negative sample according to current filler constraint frame cyclic shift and updates
Classifier is detected, loops back and forth like this, realizes being continuously tracked for target;
S4 controls the deflection of turntable according to the moving direction of target, so that target is always positioned within the scope of camera fields of view, and
According to the variation of constraint frame size, focus to camera.So that the screen accounting of target is unlikely to too small and does not see, it is also unlikely
Yu Tai great and be easy to escape from field range.
According to the above technical scheme, before the step 1, further include, the off-line training of neural network.This step is to be
It has completed when system building, has repeated before not needing each use.
According to the above technical scheme, in the step 1, the direction of target is determined according to radar information.
According to the above technical scheme, in the step 2, the frame frequency that camera reads in image is 10~50fps.
The beneficial effect comprise that: one, using the method for machine learning, it can be achieved that low small slow mesh under complex background
Target identification and real-time tracking.Under the premise of guaranteeing real-time, the accuracy of identification and tracking is improved.Only have 6 in target
Under the unfavorable conditions of a valid pixel, it is still able to achieve 95% recognition correct rate;Two, it is instructed using the target video stream of acquisition
Practice, the accuracy of automatic recognition and tracking can be improved with the increase for using the time;Three, the space of human intervention is remained,
To ensure that the correctness of system identification and tracking.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples, in attached drawing:
Fig. 1 is convolutional neural networks structural schematic diagram;
Fig. 2 is neural metwork training flow chart;
Fig. 3 is the system flow chart of photoelectricity of embodiment of the present invention recognition and tracking method;
Fig. 4 is target identification flow chart of the embodiment of the present invention based on machine learning;
Fig. 5 is target following flow chart of the embodiment of the present invention based on machine learning;
Fig. 6 is the on-the-spot schematic of unmanned plane of embodiment of the present invention identification.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
In the embodiment of the present invention, a kind of photoelectricity recognition and tracking method of low small slow target based on machine learning is provided, such as
Shown in Fig. 3, Fig. 5, Fig. 6 figure, this approach includes the following steps, step 1, it is first determined then the direction of target utilizes holder tune
The deflection and pitch angle of whole camera, so that target is located at camera within sweep of the eye;Step 2, camera read in image frame by frame;
Step 3, the on-line checking of target identification are trained by machine learning using the image of reading as the input of neural network
Good network, obtains the output of network, the constraint frame of classification and position including target;If the classification of output belongs to low small slow
Target then enters in next step, otherwise skips in next step, directly reading next frame image;Step 4 carries out target following.Step
In one, the direction of target is determined according to radar information.In step 2, the frame frequency that camera reads in image is 30fps.
Further, in the step 4 the following steps are included:
S1, the constraint frame provided by first frame image, expansion obtain filler constraint frame, and also just having obtained next frame target can
Estimation range existing for energy, using current filler constraint frame as positive sample, using the cyclic shift of current positive sample as negative sample
This, using the method for machine learning, training obtains a target detection classifier;
S2 equally fetters frame cyclic shift to the filler of previous frame, to obtained sample beam to next frame image
The image tied up in frame is classified, and the constraint frame of Response to selection most strong (related coefficient is maximum) is as the benefit where present frame target
White constraint frame;Change in location between frame is fettered according to the filler of present frame and previous frame, obtains the change in location of target;
S3 after obtaining present frame target exact position, obtains negative sample according to current filler constraint frame cyclic shift and updates
Classifier is detected, loops back and forth like this, realizes being continuously tracked for target;
S4 controls the deflection of turntable according to the moving direction of target, so that target is always positioned within the scope of camera fields of view, and
According to the variation of constraint frame size, focus to camera.So that the screen accounting of target is unlikely to too small and does not see, it is also unlikely
Yu Tai great and be easy to escape from field range.
According to the above technical scheme, before the step 1, further include, the off-line training of neural network.This step is to be
It has completed when system building, has repeated before not needing each use.
It is specifically divided into following step: firstly, by the size normalization of image to 224*224.Then, utilization is trained
The method of good convolutional neural networks carries out feature extraction to image.Entire neural network includes that 30 convolutional layers and 3 connect entirely
Connect layer.Particularly, input picture is first divided into S*S grid, each grid is responsible for detecting the object of " falling into " grid, i.e.,
The center of object is located in the grid.Each grid exports B constraint frame (matrix area comprising object) information, and
C object belongs to the probabilistic information of certain classification.Fettering frame information includes 5 data values, is x, y, w, h and confidence level respectively.
Wherein x, y refer to the coordinate shift value of the center for the object that current grid is predicted, w, h are the width and height for fettering frame
Degree, confidence level refer to current constraint frame whether include object and object space accuracy, calculation are as follows: confidence level=P
(object)*IOU.Wherein, if including object, P (object)=1 in constraint frame;Otherwise P (object)=0.IOU
(intersection over union) is the intersection area of prediction constraint frame and object real estate.Therefore, neural network is most
The output dimension of whole full articulamentum is S*S* (B*5+C).S=9, B=4, C=15 are used in practice.
Next, calculating the loss function for indicating error:
Wherein, x, y, w, C, p are neural network forecast value, and x, y, w, C, p cap is mark value.Indicate that object falls into grid i's
In j-th of constraint frame,Indicate that object is not fallen in j-th of constraint frame of grid i.
Finally, using the public image database of tape label frame, according to the value of method and error function backwards to transmission, no
The disconnected network parameter w for updating entire neural network, obtains final network parameter, as shown in Fig. 2, wherein α indicates learning rate.
Fig. 4 is the target identification flow chart based on machine learning.It is generally divided into off-line training and two stages of on-line checking.
Off-line training step carries out feature representation with background to the foreground target in training sample respectively, it is established that target or background are apparent
Model, then carry out classifier training and obtain sorter model.The on-line checking stage slides test sample on multiple scales
After dynamic window scanning, apparent model is set up using same feature representation method, is then obtained again with off-line phase training
Sorter model classifies to it, to judge whether each window is foreground target.
The essence of machine learning method is exactly the feature representation of image, that is, original image pixels are mapped to one can
The process for distinguishing dimensional space data, compared with the feature of engineer, although the calculating process of machine learning is relative complex,
Get rid of the dependence to human experience, thus realize it is most essential to image or object module portray, obtained using machine learning
To feature can obtain the recognition result of higher accuracy.Target identification is carried out using the method for machine learning, it is most main at present
The mode of stream is successively to construct a multitiered network by unsupervised deep learning, machine is made automatically to learn to lie in number
According to internal relationship.The input of this network is exactly original image data, and output is exactly position and the classification results of target.Net
The weight of each node is obtained using the data set training by label in network.
According to the difference of neural network Component units, the feature representation method based on deep learning can be divided into limitation Bohr
Hereby graceful machine (Restricted Boltzmann machine, RBM), based on from code machine (Auto encoder, AE) and being based on
Convolutional neural networks (Convolutional neural network, CNN), and various variants on the basis of them.
Wherein, convolutional neural networks are extracted high-level characteristic, improve the ability to express of feature, and the committed step of target detection is melted
It closes in same model, by training end to end, carries out whole function optimization, enhance the separability of feature.One use
It is as shown in Figure 1 in the construction method of target identification and the multilayer convolutional neural networks of classification, wherein convolutional layer is used to extract image
Feature, full articulamentum are used to forecast image position and class probability value.
The present invention establishes a kind of photoelectricity recognition and tracking method for low small slow target based on machine learning, so that by
This electro-optical system built can small slow target progress automatic identification low to unmanned plane, cruise missile etc. and real-time tracking.It is low small slow
Target is in flight course, and for flying height generally at 1000 meters hereinafter, speed is slower, flying speed is generally less than 200km/h, thunder
Small up to reflective surface area, less than 2 square metres, hardly possible discovery, difficult capture, difficulty is set, hardly possible is coped with, and the air defence of counterweight syllabus target is formed safely
Great threat.Compared with traditional recognition and tracking algorithm, the method for foundation improves automatic identification while guaranteeing real-time
Accuracy, enhance the robustness to influence factors such as illumination, targeted attitudes.This method can be not only used for the light of visible light
Electrical imaging equipment can be also used for infrared and multi-spectrum fusion imaging device, extend answering for single recognition and tracking algorithm
With range, the adaptability of algorithm is improved.
It should be understood that for those of ordinary skills, it can be modified or changed according to the above description,
And all these modifications and variations should all belong to the protection domain of appended claims of the present invention.
Claims (5)
1. a kind of photoelectricity recognition and tracking method of the low small slow target based on machine learning, which is characterized in that this method include with
Lower step, step 1, it is first determined the then direction of target adjusts the deflection and pitch angle of camera, so that target is located at phase
Machine is within sweep of the eye;
Step 2, camera read in image frame by frame;
Step 3, the on-line checking of target identification, using the image of reading as the input of neural network, by machine learning
Trained network obtains the output of network, the constraint frame of classification and position including target;If the classification of output belongs to low
Small slow target then enters in next step, otherwise skips in next step, directly reading next frame image;
Step 4 carries out target following.
2. the photoelectricity recognition and tracking method of the low small slow target according to claim 1 based on machine learning, feature exist
In, include the following steps in the step 4,
S1, the constraint frame provided by first frame image, expansion obtain filler constraint frame, and also just having obtained next frame target may deposit
Estimation range, it is sharp using the cyclic shift of current positive sample as negative sample using current filler constraint frame as positive sample
With the method for machine learning, training obtains a target detection classifier;
S2 equally fetters frame cyclic shift to the filler of previous frame to next frame image, fetters frame to obtained sample
Interior image is classified, and the strongest constraint frame of Response to selection fetters frame as the filler where present frame target;According to current
The filler of frame and previous frame fetters change in location between frame, obtains the change in location of target;
S3 after obtaining present frame target exact position, obtains negative sample according to current filler constraint frame cyclic shift and updates detection
Classifier loops back and forth like this, and realizes being continuously tracked for target;
S4 controls the deflection of turntable according to the moving direction of target, so that target is always positioned within the scope of camera fields of view, and according to
The variation for fettering frame size, focuses to camera.
3. the photoelectricity recognition and tracking method of the low small slow target according to claim 1 or 2 based on machine learning, feature
It is, before the step 1, further includes, the off-line training of neural network.
4. the photoelectricity recognition and tracking method of the low small slow target according to claim 1 or 2 based on machine learning, feature
It is, in the step 1, the direction of target is determined according to radar information.
5. the photoelectricity recognition and tracking method of the low small slow target according to claim 1 or 2 based on machine learning, feature
It is, in the step 2, the frame frequency that camera reads in image is 10~50fps.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811273114.8A CN109543553A (en) | 2018-10-30 | 2018-10-30 | The photoelectricity recognition and tracking method of low small slow target based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811273114.8A CN109543553A (en) | 2018-10-30 | 2018-10-30 | The photoelectricity recognition and tracking method of low small slow target based on machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109543553A true CN109543553A (en) | 2019-03-29 |
Family
ID=65845893
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811273114.8A Pending CN109543553A (en) | 2018-10-30 | 2018-10-30 | The photoelectricity recognition and tracking method of low small slow target based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109543553A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110533691A (en) * | 2019-08-15 | 2019-12-03 | 合肥工业大学 | Method for tracking target, equipment and storage medium based on multi-categorizer |
CN110991531A (en) * | 2019-12-02 | 2020-04-10 | 中电科特种飞机***工程有限公司 | Training sample library construction method, device and medium based on air-to-ground small and slow target |
RU2799078C1 (en) * | 2022-03-24 | 2023-07-03 | Акционерное общество "Лаборатория Касперского" | Method for detecting and recognizing small-sized objects in images using machine learning algorithm and device for its implementation |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101676744A (en) * | 2007-10-31 | 2010-03-24 | 北京航空航天大学 | Method for tracking small target with high precision under complex background and low signal-to-noise ratio |
CN106707296A (en) * | 2017-01-09 | 2017-05-24 | 华中科技大学 | Dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition method |
CN106815859A (en) * | 2017-01-13 | 2017-06-09 | 大连理工大学 | Target tracking algorism based on dimension self-adaption correlation filtering and Feature Points Matching |
CN107846258A (en) * | 2017-09-07 | 2018-03-27 | 新疆美特智能安全工程股份有限公司 | A kind of unmanned plane system of defense |
-
2018
- 2018-10-30 CN CN201811273114.8A patent/CN109543553A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101676744A (en) * | 2007-10-31 | 2010-03-24 | 北京航空航天大学 | Method for tracking small target with high precision under complex background and low signal-to-noise ratio |
CN106707296A (en) * | 2017-01-09 | 2017-05-24 | 华中科技大学 | Dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition method |
CN106815859A (en) * | 2017-01-13 | 2017-06-09 | 大连理工大学 | Target tracking algorism based on dimension self-adaption correlation filtering and Feature Points Matching |
CN107846258A (en) * | 2017-09-07 | 2018-03-27 | 新疆美特智能安全工程股份有限公司 | A kind of unmanned plane system of defense |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110533691A (en) * | 2019-08-15 | 2019-12-03 | 合肥工业大学 | Method for tracking target, equipment and storage medium based on multi-categorizer |
CN110533691B (en) * | 2019-08-15 | 2021-10-22 | 合肥工业大学 | Target tracking method, device and storage medium based on multiple classifiers |
CN110991531A (en) * | 2019-12-02 | 2020-04-10 | 中电科特种飞机***工程有限公司 | Training sample library construction method, device and medium based on air-to-ground small and slow target |
RU2799078C1 (en) * | 2022-03-24 | 2023-07-03 | Акционерное общество "Лаборатория Касперского" | Method for detecting and recognizing small-sized objects in images using machine learning algorithm and device for its implementation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | Data-driven based tiny-YOLOv3 method for front vehicle detection inducing SPP-net | |
CN105373135B (en) | A kind of method and system of aircraft docking guidance and plane type recognition based on machine vision | |
CN111222574B (en) | Ship and civil ship target detection and classification method based on multi-model decision-level fusion | |
CN114419825B (en) | High-speed rail perimeter intrusion monitoring device and method based on millimeter wave radar and camera | |
CN111814621A (en) | Multi-scale vehicle and pedestrian detection method and device based on attention mechanism | |
CN110263706A (en) | A kind of haze weather Vehicular video Detection dynamic target and know method for distinguishing | |
CN115439424A (en) | Intelligent detection method for aerial video image of unmanned aerial vehicle | |
CN103049909B (en) | A kind of be focus with car plate exposure method | |
Zhu et al. | Extreme weather recognition using convolutional neural networks | |
WO2009102001A1 (en) | Bird search system, bird search method, and computer program | |
CN112541403B (en) | Indoor personnel falling detection method by utilizing infrared camera | |
CN109543553A (en) | The photoelectricity recognition and tracking method of low small slow target based on machine learning | |
CN113253289A (en) | Unmanned aerial vehicle detection tracking system implementation method based on combination of laser radar and vision | |
Zhang et al. | Surface defect detection of wind turbine based on lightweight YOLOv5s model | |
CN117572885B (en) | Night tracking method, system and related device based on thermal infrared camera of unmanned aerial vehicle | |
CN105447431A (en) | Docking airplane tracking and positioning method and system based on machine vision | |
Shafie et al. | Smart video surveillance system for vehicle detection and traffic flow control | |
Liu et al. | Background filtering and object detection with roadside lidar data | |
CN109815773A (en) | A kind of low slow small aircraft detection method of view-based access control model | |
CN111898427A (en) | Multispectral pedestrian detection method based on feature fusion deep neural network | |
CN105551017A (en) | Transmission line forest fire target extraction method on the basis of spatio-temporal union | |
CN116189136A (en) | Deep learning-based traffic signal lamp detection method in rainy and snowy weather | |
CN114912536A (en) | Target identification method based on radar and double photoelectricity | |
CN113095181A (en) | Traffic sign identification method based on Defense-GAN | |
CN113158800A (en) | Enclosure intrusion hybrid detection method and enclosure intrusion hybrid detection system |
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: 20190329 |