CN111723654A - High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization - Google Patents
High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization Download PDFInfo
- Publication number
- CN111723654A CN111723654A CN202010398931.7A CN202010398931A CN111723654A CN 111723654 A CN111723654 A CN 111723654A CN 202010398931 A CN202010398931 A CN 202010398931A CN 111723654 A CN111723654 A CN 111723654A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- training
- building
- picture
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 76
- 238000005457 optimization Methods 0.000 title claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 106
- 238000000034 method Methods 0.000 claims abstract description 50
- 230000008569 process Effects 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims description 38
- 238000012795 verification Methods 0.000 claims description 32
- 238000012545 processing Methods 0.000 claims description 22
- 238000004422 calculation algorithm Methods 0.000 claims description 18
- 230000009466 transformation Effects 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 10
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000012986 modification Methods 0.000 claims description 6
- 230000004048 modification Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 3
- 238000013138 pruning Methods 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims 1
- 238000012216 screening Methods 0.000 claims 1
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000011897 real-time detection Methods 0.000 abstract description 2
- 238000009826 distribution Methods 0.000 description 26
- 239000000203 mixture Substances 0.000 description 7
- 230000006399 behavior Effects 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 230000033001 locomotion Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009435 building construction Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a high-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization. The invention relates to the technical field of deep learning and computer vision, and solves the problem that the accuracy of a model of the existing high-altitude falling object detection method is limited. The invention mainly aims at detecting falling objects, is based on background modeling and real-time detection of YOLOv3, has high model stability and good detection and identification performance, can eliminate the influence of external factors, and achieves the aim of 'precise control'. According to the invention, the building camera is used for shooting and transmitting the video in real time as a data source, so that manual intervention is reduced, and the scene of falling objects can be locked quickly in time. The method can collect the detected object pictures and the results of manual treatment, convert the results of manual treatment into the labels corresponding to the training model, and automatically start the training process of the model after the number of the collected object pictures reaches the specified threshold value, so that the accuracy of the model is gradually improved.
Description
Technical Field
The invention relates to the technical field of deep learning and computer vision, in particular to a high-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization.
Background
In the current urban development and construction, high-rise building construction is more and more common due to various limiting factors such as land, space and the like. The variety of derived problems arising from this is also increasing, and among them, safety issues arising from high altitude parabolas and falls are also receiving increasing attention from all parties. The object thrown or dropped from high altitude is very easy to cause serious accidents of personal injury once the object is suddenly thrown or dropped, and is dangerous and irreplaceable.
At present, the behavior of monitoring the high-altitude parabolic object mainly depends on the methods of unscheduled patrol of spontaneous organization of residents, sticking warning slogans in a high-altitude parabolic disaster area and the like. But the method of irregular patrol cannot completely monitor the behavior of the high-altitude parabola and consumes a great deal of manpower. The method of pasting the warning slogan in the disaster area is also easily ignored by the passerby, and cannot play a role in effectively reminding the passerby. In this case, in order to further prevent an invarious behavior such as high-altitude parabola, which is a very harmful behavior, a high-altitude parabola detection system based on camera video stream image processing needs to be built.
The traditional algorithm is mostly used in the existing high-altitude parabolic detection method: for example, background difference, inter-frame difference, etc., in the conventional algorithm, false detection is easily caused by non-high altitude parabolas such as birds and balloons. And most of the existing high-altitude falling object detection methods are based on the traditional algorithm, and the model cannot be automatically optimized, so that the accuracy of the model is limited in the long run.
Disclosure of Invention
The invention provides a high-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization, and aims to solve the problem that the accuracy of a model of the existing high-altitude falling object detection method is limited.
In a first aspect, the present invention provides a high altitude parabolic detection method based on background modeling, YOLOv3 and self-optimization, wherein the method includes:
acquiring building video data shot by a camera;
intercepting pictures of the building video data according to frames to obtain an initial data set;
preprocessing the initial data set to obtain a building picture data set;
dividing the building picture data set into a training set, a verification set and a test set according to a preset proportion;
marking the building picture data in the training set, the verification set and the test set according to object categories formulated in a preset high-altitude parabolic detection strategy to obtain marking data sets corresponding to the building picture data one by one, wherein the marking data sets comprise object marking frame coordinates and object category information;
performing data enhancement processing on the training set and the corresponding labeled data set;
carrying out model training on the training set subjected to data enhancement processing and the corresponding labeled data set by using a YOLOv3 network to obtain a plurality of high-altitude parabolic detection models with different weights;
inputting the marked data sets of the verification set and the verification set into a YOLOv3 network, verifying the model while training the model, and obtaining the current accuracy of the model so as to adjust the parameters of the model in time and obtain an optimal weight model;
after the model training is finished, testing the optimal weight model by using the test set and the test set label data set;
acquiring real-time building video data shot by a camera;
searching for moving objects in the real-time building video data by using a background modeling algorithm;
inputting the image of the moving object into an optimal weight model to perform high-altitude parabolic detection to obtain a high-altitude parabolic detection result;
returning the high-altitude parabolic detection result to related personnel at the service end of the social management platform in a screenshot form for manual examination;
judging whether the moving object belongs to an overhead falling object or not according to a manual auditing result;
if the moving object belongs to a falling object, sending out falling object warning information to inform related personnel to process;
if the moving object does not belong to the falling object, receiving modification information of the object type information of the moving object by related personnel, and taking a modified picture sample as a difficult sample;
judging whether the number of the difficult samples exceeds a preset threshold value or not;
and if the number of the difficult samples exceeds a preset threshold value, automatically starting a self-optimization process, combining the difficult samples and a training set into a new training set, and performing model training according to the new training set by using a YOLOv3 network to obtain a high-altitude parabolic detection model after training iteration.
With reference to the first aspect, in a first implementation manner of the first aspect, in the step of preprocessing the starting data set to obtain a building picture data set, distorted, deformed and blurred picture data are screened and corrected to obtain the building picture data set.
With reference to the first aspect, in a second implementation manner of the first aspect, the marking building picture data in the training set, the verification set, and the test set according to an object class formulated in a preset high altitude parabolic detection strategy to obtain a marked data set corresponding to the building picture data one to one includes:
and selecting objects in each picture of the training set, the verification set and the test set in the object category by using a rectangular frame, storing the positions of the rectangular frame in the pictures, wherein the positions comprise coordinate information of the upper left corner and the lower right corner of the rectangular frame, marking the category of the objects, generating an XML file from the marked information, and forming a marked data set which corresponds to the marked pictures one by one.
With reference to the second implementable manner of the first aspect, in a third implementable manner of the first aspect, the object categories include moving objects that may often cause misjudgment, such as cars, birds, people, balloons, plastic bags, and the like.
With reference to the first aspect, in a fourth implementable manner of the first aspect, the performing data enhancement processing on the training set and the corresponding labeled data set includes:
performing data enhancement processing by adopting a flip transformation, a random pruning, a color dithering, a translation transformation, a scale transformation, a contrast transformation, a noise disturbance, a rotation transformation or a reflection transformation and a mixup method, wherein the mixup method comprises the following steps:
wherein (x)i,yi),(xj,yj) Is two samples randomly drawn from the training data, x represents the picture matrix, y represents the label information,
participate in model training data after enhancement, and λ ∈ [0,1 ]]。
With reference to the first aspect, in a fifth implementable manner of the first aspect, in the step of performing model training on the training set after data enhancement processing and the corresponding labeled data set by using the YOLOv3 network to obtain the high altitude parabolic detection model with multiple weights, the labeled picture sample and the background picture sample are sent to the YOLOv3 network together, the labeled picture is a positive sample, the background picture does not contain an object in an object class, the unlabeled background picture is a negative sample, the positive and negative samples are trained in the YOLOv3 network together, and the high altitude parabolic detection model with multiple weights is obtained through iterative training.
With reference to the first aspect, in a sixth implementation manner of the first aspect, in the step of acquiring real-time building video data shot by the camera, a data transmission format between the monitoring video and the server is a video stream, and a transmission protocol is an ONVIF protocol.
With reference to the first aspect, in a seventh implementable manner of the first aspect, the background modeling algorithm is an adaptive mixed gaussian background modeling algorithm.
In a second aspect, the present invention provides a high altitude parabolic detection apparatus based on background modeling, YOLOv3 and self-optimization, the apparatus comprising:
the first acquisition unit is used for acquiring building video data shot by the camera;
the intercepting unit is used for intercepting pictures of the building video data according to frames to obtain an initial data set;
the preprocessing unit is used for preprocessing the initial data set to obtain a building picture data set;
the grouping unit is used for dividing the building picture data set into a training set, a verification set and a test set according to a preset proportion;
the marking unit is used for marking the building picture data in the training set, the verification set and the test set according to object types formulated in a preset high-altitude parabolic detection strategy to obtain marking data sets corresponding to the building picture data one by one, and each marking data set comprises object marking frame coordinates and object type information;
the data enhancement unit is used for carrying out data enhancement processing on the training set and the corresponding marked data set;
the model training unit is used for performing model training on the training set subjected to data enhancement processing and the corresponding labeled data set by using a YOLOv3 network to obtain a plurality of high-altitude parabolic detection models with different weights;
the model verification unit is used for inputting the verification set and the marked data set of the verification set into a YOLOv3 network, verifying the model while training the model, and obtaining the current accuracy of the model so as to adjust the model parameters in time and obtain an optimal weight model;
the model testing unit is used for testing the optimal weight model by utilizing the test set and the test set marking data set after the model training is finished;
the second acquisition unit is used for acquiring real-time building video data shot by the camera;
the searching unit is used for searching a moving object in the real-time building video data by utilizing a background modeling algorithm;
the detection unit is used for inputting the image of the moving object into an optimal weight model to perform high-altitude parabolic detection to obtain a high-altitude parabolic detection result;
the return unit is used for returning the high-altitude parabolic detection result to related personnel at the service end of the social management platform in a screenshot form for manual examination;
the first judgment unit is used for judging whether the moving object belongs to an overhead falling object or not according to a manual auditing result;
the notification unit is used for sending out high-altitude falling object warning information to notify related personnel to process under the condition that the moving object belongs to a high-altitude falling object;
the modifying unit is used for receiving modification information of object type information of the moving object from related personnel under the condition that the moving object does not belong to a high falling object, and taking a modified picture sample as a difficult sample;
the second judging unit is used for judging whether the number of the difficult samples exceeds a preset threshold value or not;
and the merging unit is used for automatically starting a self-optimization process under the condition that the number of the difficult samples exceeds a preset threshold value, merging the difficult samples and the training set into a new training set, and performing model training according to the new training set by using a YOLOv3 network to obtain a high-altitude parabolic detection model after training iteration.
The invention has the following beneficial effects: according to the high-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization, a building camera is used for shooting and transmitting a video in real time to serve as a data source, manual intervention is reduced, a scene where high-altitude falling objects occur can be locked timely and quickly, and the purposes of early finding and early treating of high-altitude falling object behaviors are achieved. The invention is applied to the detection of foreign matters in the social treatment platform, and manual treatment can be carried out on the platform after the high-altitude falling object behavior is detected. The method can collect the detected object pictures and the results of manual treatment, convert the results of manual treatment into the labels corresponding to the training models, automatically start the model training process when the number of the collected object pictures reaches the specified threshold value, and obtain more manual feedback and more on-line training iterations as time goes on, so that the effect of the models is gradually optimized, and the accuracy is gradually improved. The invention mainly aims at detecting falling objects, is based on background modeling and real-time detection of YOLOv3, has high model stability and good detection and identification performance, can eliminate the influence of external factors, and achieves the aim of 'precise control'.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained according to the drawings without any inventive exercise.
Fig. 1 is a flowchart of a high altitude parabolic detection method based on background modeling, YOLOv3 and self-optimization provided by the present invention.
Fig. 2 is a schematic diagram of a high altitude parabolic detection apparatus based on background modeling, YOLOv3 and self-optimization provided by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the specific embodiments of the present invention and the accompanying drawings. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Referring to fig. 1, the present invention provides a high altitude parabolic detection method based on background modeling, YOLOv3 and self-optimization, the method includes:
and step S101, acquiring building video data shot by the camera.
Data acquisition is an important basis in the invention, in the YOLOv3 algorithm, data is required to be input into a network to train a model, and the invention can adopt videos or photos of various common objects in a living scene including building video data shot by a camera as a data set.
And S102, intercepting pictures of the building video data according to frames to obtain an initial data set.
And step S103, preprocessing the initial data set to obtain a building picture data set.
Specifically, distorted, deformed and blurred picture data are screened and corrected to obtain the building picture data set.
And step S104, dividing the building picture data set into a training set, a verification set and a test set according to a preset proportion.
And S105, marking the building picture data in the training set, the verification set and the test set according to object types formulated in a preset high-altitude parabolic detection strategy to obtain a marking data set corresponding to the building picture data one by one, wherein the marking data set comprises object marking frame coordinates and object type information.
Specifically, for each object in each picture of the training set, the verification set and the test set in the object category, selecting the object by using a rectangular frame, storing the position of the rectangular frame in the picture, wherein the position comprises coordinate information of the upper left corner and the lower right corner of the rectangular frame, marking the category of the object, generating an XML file from the marked information, and forming a marked data set corresponding to the marked pictures one by one. The object categories may include moving objects such as cars, birds, people, balloons, plastic bags, etc. which are often misjudged.
And S106, performing data enhancement processing on the training set and the corresponding labeled data set.
Because the data volume of this patent is less than the data volume that general deep learning needs far away, therefore this patent can be classified into the study of small sample (few _ shot _ learning), and the method of solving the study of small sample mainly has data enhancement and meta-learning. The invention mainly uses data enhancement to solve the problem of small sample learning. The generalization and the precision of the model can be improved by a large amount of data, so that the expansion of the data volume by technical means is an indispensable step on the premise of limited data.
Specifically, the data enhancement processing may be performed using a flip transform, a random crop, a color dither, a translation transform, a scale transform, a contrast transform, a noise disturbance, a rotation transform, or a reflection transform, and a mixup method, which is as follows:
wherein (x)i,yi),(xj,yj) Is two samples randomly drawn from the training data, x represents the picture matrix, y represents the label information,
participate in model training data after enhancement, and λ ∈ [0,1 ]]. The invention adopts a mixup method to add the obtained picture into the original data set. By expanding the training samples, the overfitting problem caused by too small data volume can be avoided. The mixup method is a simple data enhancement mode irrelevant to data, and a virtual training sample is constructed. The mixup extends the training distribution by incorporating a priori knowledge that linear interpolation of linear vectors should result in linear interpolation of the relevant labels, and introduces minimal computational overhead.
And S107, performing model training on the training set subjected to data enhancement processing and the corresponding labeled data set by using a YOLOv3 network to obtain a plurality of high-altitude parabolic detection models with different weights.
Specifically, the marked picture sample and the background picture sample are sent to a YOLOv3 network together, the marked picture is a positive sample, the background picture does not contain objects in object categories, the unmarked background picture is a negative sample, the positive and negative samples are trained in the YOLOv3 network together, and the high altitude parabolic detection model with multiple weights is obtained through iterative training.
And S108, inputting the marked data sets of the verification set and the verification set into a YOLOv3 network, verifying the model while training the model, and obtaining the current accuracy of the model so as to adjust the parameters of the model in time and obtain an optimal weight model.
And step S109, after the model training is finished, testing the optimal weight model by using the test set and the test set label data set.
And step S110, acquiring real-time building video data shot by the camera.
Specifically, the data transmission format of the monitoring video and the server is a video stream, and the transmission protocol is an ONVIF protocol.
And step S111, searching for a moving object in the real-time building video data by using a background modeling algorithm.
Specifically, the background modeling algorithm is an adaptive Gaussian mixture background modeling algorithm.
The working principle of the self-adaptive mixed Gaussian background modeling is as follows: in the detection and extraction of the moving target, a background target is important for the identification and tracking of the target. Modeling is an important link of background target extraction. Foreground means that any meaningful moving object is the foreground under the assumption that the background is stationary. The problems of moving object detection are mainly divided into two categories, camera fixation and camera motion. For the problem of detecting a moving object moving by a camera, a well-known solution is an optical flow method, and an optical flow field of an image sequence is solved by solving a partial differential equation, so that the motion state of the camera is predicted. In the case of camera fixation, an optical flow method may be used, but due to the complexity of the optical flow method, real-time calculation is often difficult. Whereas hybrid gaussian background modeling is suitable for separating the background and foreground from the image sequence with the camera fixed. Under the condition that a camera is fixed, the change of a background is slow, and is mostly influenced by illumination, wind and the like, a foreground and the background are separated from a given image through modeling the background, and generally, the foreground is a moving object, so that the purpose of detecting the moving object is achieved.
The Gaussian mixture model has been widely applied to robust complex scene background modeling, especially in situations with small repetitive motions, such as swaying leaves, bushes, rotating fans, sea surges, rainy and snowy weather, light reflections, and the like. The pixel-based Gaussian mixture model is effective in modeling the multimodal distribution background, can adapt to the change of the background such as light gradual change, and can basically meet the real-time requirement of the algorithm in practical application.
The mixed Gaussian background modeling is a background representation method based on pixel sample statistical information, the background is represented by using statistical information (such as mode quantity, mean value and standard deviation of each mode) such as probability density of a large number of sample values of a pixel in a long time, and then target pixels are judged by using statistical difference, so that the complex dynamic background can be modeled, and the calculated amount is large.
In the Gaussian mixture background model, the color information among the pixels is considered to be irrelevant, and the processing of each pixel point is independent. For each pixel point in the video image, the change of the value in the sequence image can be regarded as a random process which continuously generates the pixel value, i.e. the color rendering rule of each pixel point is described by Gaussian distribution.
For a multi-peak Gaussian distribution model, each pixel point of an image is modeled according to superposition of a plurality of Gaussian distributions with different weights, each Gaussian distribution corresponds to a state which can possibly generate the color presented by the pixel point, and the weight and distribution parameters of each Gaussian distribution are updated along with time. When processing color images, it is assumed that the image pixels R, G, B have three color channels that are independent of each other and have the same variance. Observation data set { X for random variable X1,x2,…,xN}，xt＝(rt,gt,bt) For a sample of the pixel at time t, then a single sample point xtIts obeyed mixture gaussian distribution probability density function:
where k is the total number of distribution patterns, η (x)t,μi,t,τi,t) Is the ith Gaussian distribution at time ti,tIs the mean value ofi,tFor the purpose of its covariance matrix,
is variance, I is three-dimensional identity matrix, wi,tThe weight of the ith gaussian distribution at time t.
The maximum number of Gaussian distributions of each pixel point is set to be k by the self-adaptive Gaussian mixture modelmax4. Assuming that the initial number of the Gaussian distributions of each pixel point is k equal to 1, taking the pixel value of each point of the first frame as the initial mean value u of the Gaussian distributions0Variance is σ0Weight is taken as 300＝0.2。
If the current k Gaussian distributions are not matched with the target pixel and k exist<kmaxThen k is k +1 and a new gaussian distribution is added to the background model, which takes the current pixel value as the mean, with a standard deviation and weight of 30, 0.2, respectively. But if k is equal to k at this timemaxA new gaussian distribution will be generated with its mean initialized with the current pixel value, standard deviation and weight of 30, 0.01 respectively. The new Gaussian distribution will immediately replace the original kmaxThe one with the smallest weight in the distribution.
In the traditional mixed Gaussian background modeling algorithm, the first successful matching is taken as a matching result. In fact, the new pixel may be successfully matched to the polynomial gaussian distribution, and the first match is not necessarily the best match. In the adaptive Gaussian mixture model, each Gaussian distribution is matched with a new pixel, and the optimal distribution of the matching result is found out, wherein the optimal distribution is obtained by the following formula:
if the optimal matching of the continuous 10 frames is the same distribution and k is greater than 1, k is k-1, and the distribution with the minimum weight value is directly removed.
The working principle of the YOLOv3 network is explained again:
the Yolov3 adjusts the network structure on the basis of Yolov1 and Yolov2, utilizes multi-scale features to detect objects, and replaces softmax with Logistic in object classification. YOLOv3 has no full connection layer and pooling layer, can correspond to input images of any size, mainly comprises 75 convolutional layers, and is additionally provided with a resnet residual module in the network, so that the gradient problem of the deep network is solved.
The Resnet residual error network is equivalent to adding a shortcut path in an original CNN network structure, and the learning process is changed from directly learning features to adding certain features on the basis of the previously learned features so as to obtain better features. Thus, a complex feature h (x), which was previously learned independently layer by layer, now becomes a model h (x) ═ f (x) + x, where x is the feature at the beginning of the short, and f (x) is the padding and addition of x, which becomes the residual. Therefore, the learning target changes from learning complete information to learning residual. The difficulty in learning the high-quality features is thus greatly reduced.
An image typically contains a variety of different objects and has a size. It is desirable to be able to detect all sizes of objects at once. Therefore, the network must have the ability to "see" objects of different sizes. And the deeper the network, the smaller the signature, so that smaller objects are more difficult to detect later. Aiming at the problem, the YOLOv3 uses 3 feature maps with different scales to detect the object, so as to detect features with finer granularity, the YOLOv3 adopts an fpn (feature Pyramid network) structure to correspond to different accuracies of multiple scales, and respectively performs target detection on feature maps with different depths, and the feature map of the current layer performs up-sampling on the feature map of the future layer and utilizes the feature map to fuse the low-order features and the high-order features, thereby improving the detection accuracy.
The Softmax layer is replaced with a 1x1 convolutional layer + logistic activation function structure. May correspond to a multi-tagged object. When YOLOv3 predicts the preselected boxes bbox, a logistic regression is adopted, each preselected box contains five elements bbox (bx, y, w, h, c), wherein the first four elements represent the size and coordinate position of the preselected box, and the last value is confidence.
Pr (object) IOU (bbox, object) is σ (t0), where pr (object) IOU (bbox, object) is a confidence level.
The logistic regression will score the bbox envelope for an objective score to find the one with the highest score for the likelihood of existence of the objective.
And step S112, inputting the image of the moving object into the optimal weight model to perform high-altitude parabolic detection, so as to obtain a high-altitude parabolic detection result.
Specifically, by comparing the image of the moving object with the object identified by the YOLOv3 target detection model, if the identification result is an object such as a bird, a balloon, a plastic bag, or the like, it is not determined as a falling object. And the strategy can be flexibly configured according to the actual scene, and the interference of the flying object in the air can be effectively reduced.
And S113, returning the high-altitude parabolic detection result to relevant personnel of the social management platform server side in a screenshot form for manual examination.
And step S114, judging whether the moving object belongs to a falling object or not according to the manual checking result.
And step S115, if the moving object belongs to a falling object, sending out falling object warning information to inform related personnel to process.
And step S116, if the moving object does not belong to the high-altitude falling object, receiving modification information of the object type information of the moving object by related personnel, and taking the modified picture sample as a difficult sample.
In step S117, it is determined whether the number of difficult samples exceeds a preset threshold.
And S118, if the number of the difficult samples exceeds a preset threshold value, automatically starting a self-optimization process, combining the difficult samples and a training set into a new training set, and performing model training according to the new training set by using a YOLOv3 network to obtain a high-altitude parabolic detection model after training iteration.
The data source of the invention mainly comprises three parts: and detecting the wrong picture sample and the background picture sample which does not contain the target object type. The marked picture sample is a positive sample, the background picture is a negative sample, the positive sample and the negative sample form a data set X, and the X is sent to a YOLOv3 network for training; the samples that detect errors are called difficult samples whose labels are manually altered and then iteratively trained in conjunction with the data set X. The method adds the detected result into the data set to enter a self-optimization process, collects the detected object picture and the result of manual treatment, and converts the result of manual treatment into a label corresponding to the training model. When the number of the collected object pictures reaches a specified threshold value, the method can automatically start the process of model training. With the lapse of time, the artificial feedback obtained by the method is increased, the number of times of on-line training iteration is increased, the effect of the model is gradually optimized, and the accuracy is gradually improved.
As shown in fig. 2, the present invention provides a high altitude parabolic detection apparatus based on background modeling, YOLOv3 and self-optimization, the apparatus includes:
a first obtaining unit 201, configured to obtain building video data captured by a camera.
And the intercepting unit 202 is used for intercepting pictures of the building video data according to frames to obtain an initial data set.
And the preprocessing unit 203 is configured to preprocess the initial data set to obtain a building picture data set.
And the grouping unit 204 is configured to divide the building picture data set into a training set, a verification set and a test set according to a preset ratio.
And the marking unit 205 is configured to mark the building picture data in the training set, the verification set and the test set according to an object category formulated in a preset high-altitude parabolic detection strategy to obtain a marking data set corresponding to the building picture data one to one, where the marking data set includes object marking frame coordinates and object category information.
And the data enhancement unit 206 is configured to perform data enhancement processing on the training set and the corresponding labeled data set.
And the model training unit 207 is configured to perform model training on the training set subjected to the data enhancement processing and the corresponding labeled data set by using the YOLOv3 network, so as to obtain a plurality of high-altitude parabolic detection models with different weights.
And the model verification unit 208 is configured to input the verification set and the labeled data set of the verification set into a YOLOv3 network, verify the model while training the model, and obtain the current accuracy of the model, so as to adjust model parameters in time and obtain an optimal weight model.
And the model testing unit 209 is configured to test the optimal weight model by using the test set and the test set label data set after the model training is completed.
And a second obtaining unit 210, configured to obtain real-time building video data captured by the camera.
The finding unit 211 is configured to find a moving object in the real-time building video data by using a background modeling algorithm.
And the detection unit 212 is configured to input the image of the moving object into the optimal weight model to perform high-altitude parabolic detection, so as to obtain a high-altitude parabolic detection result.
And the returning unit 213 is configured to return the high-altitude parabolic detection result to the relevant personnel of the social governance platform server in the form of screenshot for manual review.
And a first judging unit 214, configured to judge whether the moving object belongs to an overhead falling object according to the manual review result.
And a notification unit 215, configured to send out falling object warning information to notify relevant persons to perform processing in case that the moving object belongs to a falling object.
And the modifying unit 216 is configured to receive modification information of the object category information of the moving object from a related person under the condition that the moving object does not belong to a high falling object, and use the modified picture sample as a difficult sample.
A second judging unit 217, configured to judge whether the number of difficult samples exceeds a preset threshold.
A merging unit 218, configured to automatically start a self-optimization process when the number of the difficult samples exceeds a preset threshold, merge the difficult samples and the training set into a new training set, and perform model training according to the new training set by using a YOLOv3 network to obtain a high-altitude parabolic detection model after training iteration.
An embodiment of the present invention further provides a storage medium, and a storage medium, where a computer program is stored in the storage medium, and when the computer program is executed by a processor, the computer program implements part or all of the steps in each embodiment of the high altitude parabola detection method based on background modeling, YOLOv3 and self-optimization provided by the present invention. The storage medium may be a magnetic disk, an optical disk, a Read-only memory (ROM) or a Random Access Memory (RAM).
Those skilled in the art will readily appreciate that the techniques of the embodiments of the present invention may be implemented as software plus a required general purpose hardware platform. Based on such understanding, the technical solutions in the embodiments of the present invention may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments.
The same and similar parts in the various embodiments in this specification may be referred to each other. In particular, for the embodiments of the high altitude parabolic detection apparatus based on background modeling, YOLOv3 and self optimization, since they are substantially similar to the embodiments of the method, the description is simple, and the relevant points can be referred to the description in the embodiments of the method.
The above-described embodiments of the present invention should not be construed as limiting the scope of the present invention.
Claims (9)
1. A high altitude parabolic detection method based on background modeling, YOLOv3 and self optimization is characterized by comprising the following steps:
acquiring building video data shot by a camera;
intercepting pictures of the building video data according to frames to obtain an initial data set;
preprocessing the initial data set to obtain a building picture data set;
dividing the building picture data set into a training set, a verification set and a test set according to a preset proportion;
marking the building picture data in the training set, the verification set and the test set according to object categories formulated in a preset high-altitude parabolic detection strategy to obtain marking data sets corresponding to the building picture data one by one, wherein the marking data sets comprise object marking frame coordinates and object category information;
performing data enhancement processing on the training set and the corresponding labeled data set;
carrying out model training on the training set subjected to data enhancement processing and the corresponding labeled data set by using a YOLOv3 network to obtain a plurality of high-altitude parabolic detection models with different weights;
inputting the marked data sets of the verification set and the verification set into a YOLOv3 network, verifying the model while training the model, and obtaining the current accuracy of the model so as to adjust the parameters of the model in time and obtain an optimal weight model;
after the model training is finished, testing the optimal weight model by using the test set and the test set label data set;
acquiring real-time building video data shot by a camera;
searching for moving objects in the real-time building video data by using a background modeling algorithm;
inputting the image of the moving object into an optimal weight model to perform high-altitude parabolic detection to obtain a high-altitude parabolic detection result;
returning the high-altitude parabolic detection result to related personnel at the service end of the social management platform in a screenshot form for manual examination;
judging whether the moving object belongs to an overhead falling object or not according to a manual auditing result;
if the moving object belongs to a falling object, sending out falling object warning information to inform related personnel to process;
if the moving object does not belong to the falling object, receiving modification information of the object type information of the moving object by related personnel, and taking a modified picture sample as a difficult sample;
judging whether the number of the difficult samples exceeds a preset threshold value or not;
and if the number of the difficult samples exceeds a preset threshold value, automatically starting a self-optimization process, combining the difficult samples and a training set into a new training set, and performing model training according to the new training set by using a YOLOv3 network to obtain a high-altitude parabolic detection model after training iteration.
2. The method as claimed in claim 1, wherein the step of preprocessing said starting data set to obtain a building picture data set comprises screening and correcting distorted, distorted or blurred picture data to obtain said building picture data set.
3. The method as claimed in claim 1, wherein the step of marking the building picture data in the training set, the verification set and the test set according to the object categories formulated in the preset high altitude parabolic detection strategy to obtain a marked data set corresponding to the building picture data one by one comprises the steps of:
and selecting objects in each picture of the training set, the verification set and the test set in the object category by using a rectangular frame, storing the positions of the rectangular frame in the pictures, wherein the positions comprise coordinate information of the upper left corner and the lower right corner of the rectangular frame, marking the category of the objects, generating an XML file from the marked information, and forming a marked data set which corresponds to the marked pictures one by one.
4. The method of claim 3, wherein the object categories include moving objects that are commonly found to cause false positives, such as cars, birds, people, balloons, plastic bags, and the like.
5. The method of claim 1, wherein data enhancing the training set and corresponding labeled data set comprises:
performing data enhancement processing by adopting a flip transformation, a random pruning, a color dithering, a translation transformation, a scale transformation, a contrast transformation, a noise disturbance, a rotation transformation or a reflection transformation and a mixup method, wherein the mixup method comprises the following steps:
wherein (x)i,yi),(xj,yj) Is two samples randomly drawn from the training data, x represents the picture matrix, y represents the label information,
participate in model training data after enhancement, and λ ∈ [0,1 ]]。
6. The method as claimed in claim 1, wherein in the step of performing model training on the training set after data enhancement processing and the corresponding labeled data set by using a YOLOv3 network to obtain the multi-weight high altitude parabolic detection model, the labeled picture samples and the background picture samples are sent into a YOLOv3 network together, the labeled picture samples are positive samples, the background picture does not contain objects in the object class, the unlabeled background picture samples are negative samples, the positive and negative samples are trained in the YOLOv3 network together, and the multi-weight high altitude parabolic detection model is obtained through iterative training.
7. The method as claimed in claim 1, wherein in the step of acquiring real-time building video data shot by the camera, the data transmission format of the monitoring video and the server is video stream, and the transmission protocol is ONVIF protocol.
8. The method of claim 1, wherein the background modeling algorithm is an adaptive mixed gaussian background modeling algorithm.
9. An apparatus for high altitude parabolic detection based on background modeling, YOLOv3 and self optimization, the apparatus comprising:
the first acquisition unit is used for acquiring building video data shot by the camera;
the intercepting unit is used for intercepting pictures of the building video data according to frames to obtain an initial data set;
the preprocessing unit is used for preprocessing the initial data set to obtain a building picture data set;
the grouping unit is used for dividing the building picture data set into a training set, a verification set and a test set according to a preset proportion;
the marking unit is used for marking the building picture data in the training set, the verification set and the test set according to object types formulated in a preset high-altitude parabolic detection strategy to obtain marking data sets corresponding to the building picture data one by one, and each marking data set comprises object marking frame coordinates and object type information;
the data enhancement unit is used for carrying out data enhancement processing on the training set and the corresponding marked data set;
the model training unit is used for performing model training on the training set subjected to data enhancement processing and the corresponding labeled data set by using a YOLOv3 network to obtain a plurality of high-altitude parabolic detection models with different weights;
the model verification unit is used for inputting the verification set and the marked data set of the verification set into a YOLOv3 network, verifying the model while training the model, and obtaining the current accuracy of the model so as to adjust the model parameters in time and obtain an optimal weight model;
the model testing unit is used for testing the optimal weight model by utilizing the test set and the test set marking data set after the model training is finished;
the second acquisition unit is used for acquiring real-time building video data shot by the camera;
the searching unit is used for searching a moving object in the real-time building video data by utilizing a background modeling algorithm;
the detection unit is used for inputting the image of the moving object into an optimal weight model to perform high-altitude parabolic detection to obtain a high-altitude parabolic detection result;
the return unit is used for returning the high-altitude parabolic detection result to related personnel at the service end of the social management platform in a screenshot form for manual examination;
the first judgment unit is used for judging whether the moving object belongs to an overhead falling object or not according to a manual auditing result;
the notification unit is used for sending out high-altitude falling object warning information to notify related personnel to process under the condition that the moving object belongs to a high-altitude falling object;
the modifying unit is used for receiving modification information of object type information of the moving object from related personnel under the condition that the moving object does not belong to a high falling object, and taking a modified picture sample as a difficult sample;
the second judging unit is used for judging whether the number of the difficult samples exceeds a preset threshold value or not;
and the merging unit is used for automatically starting a self-optimization process under the condition that the number of the difficult samples exceeds a preset threshold value, merging the difficult samples and the training set into a new training set, and performing model training according to the new training set by using a YOLOv3 network to obtain a high-altitude parabolic detection model after training iteration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010398931.7A CN111723654B (en) | 2020-05-12 | 2020-05-12 | High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010398931.7A CN111723654B (en) | 2020-05-12 | 2020-05-12 | High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111723654A true CN111723654A (en) | 2020-09-29 |
| CN111723654B CN111723654B (en) | 2023-04-07 |
Family
ID=72564368
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010398931.7A Active CN111723654B (en) | 2020-05-12 | 2020-05-12 | High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111723654B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308000A (en) * | 2020-11-06 | 2021-02-02 | 安徽清新互联信息科技有限公司 | High-altitude parabolic detection method based on space-time information |
| CN112329627A (en) * | 2020-11-05 | 2021-02-05 | 重庆览辉信息技术有限公司 | High-altitude throwing object distinguishing method |
| CN112330743A (en) * | 2020-11-06 | 2021-02-05 | 安徽清新互联信息科技有限公司 | High-altitude parabolic detection method based on deep learning |
| CN112347915A (en) * | 2020-11-05 | 2021-02-09 | 重庆览辉信息技术有限公司 | High-altitude throwing object distinguishing system |
| CN112686186A (en) * | 2021-01-05 | 2021-04-20 | 润联软件***(深圳)有限公司 | High-altitude parabolic recognition method based on deep learning and related components thereof |
| CN112926445A (en) * | 2021-02-24 | 2021-06-08 | 北京爱笔科技有限公司 | Parabolic behavior recognition method, model training method and related device |
| CN113076809A (en) * | 2021-03-10 | 2021-07-06 | 青岛海纳云科技控股有限公司 | High-altitude falling object detection method based on visual Transformer |
| CN113112777A (en) * | 2021-04-07 | 2021-07-13 | 浙江磐至科技有限公司 | Wireless transmission security system based on system technology service under monitoring |
| CN113159159A (en) * | 2021-04-15 | 2021-07-23 | 东北大学 | Small sample image classification method based on improved CNN |
| CN113516042A (en) * | 2021-05-17 | 2021-10-19 | 江苏奥易克斯汽车电子科技股份有限公司 | High-altitude parabolic detection method, device and equipment |
| CN113837087A (en) * | 2021-09-24 | 2021-12-24 | 上海交通大学宁波人工智能研究院 | Animal target detection system and method based on YOLOv3 |
| WO2022142414A1 (en) * | 2020-12-30 | 2022-07-07 | 深圳云天励飞技术股份有限公司 | High-rise littering monitoring method and apparatus, electronic device, and storage medium |
| CN114863370A (en) * | 2022-07-08 | 2022-08-05 | 合肥中科类脑智能技术有限公司 | Complex scene high altitude parabolic identification method and system |
| CN117079095A (en) * | 2023-06-25 | 2023-11-17 | 江南大学 | Deep learning-based high-altitude parabolic detection method, system, medium and equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018095082A1 (en) * | 2016-11-28 | 2018-05-31 | 江苏东大金智信息***有限公司 | Rapid detection method for moving target in video monitoring |
| CN109309811A (en) * | 2018-08-31 | 2019-02-05 | 中建三局智能技术有限公司 | A kind of throwing object in high sky detection system based on computer vision and method |
| CN109711319A (en) * | 2018-12-24 | 2019-05-03 | 安徽高哲信息技术有限公司 | A kind of method and system that grain unsound grain image recognition sample database is established |
| CN109872341A (en) * | 2019-01-14 | 2019-06-11 | 中建三局智能技术有限公司 | A kind of throwing object in high sky detection method based on computer vision and system |
| CN110275042A (en) * | 2019-05-07 | 2019-09-24 | 深圳市零壹移动互联***有限公司 | A kind of throwing object in high sky detection method based on computer vision and radio signal analysis |
| CN110853295A (en) * | 2019-11-12 | 2020-02-28 | 江西赣鄱云新型智慧城市技术研究有限公司 | High-altitude parabolic early warning method and device |
| CN110969604A (en) * | 2019-11-26 | 2020-04-07 | 北京工业大学 | Intelligent security real-time windowing detection alarm system and method based on deep learning |
| CN111079663A (en) * | 2019-12-19 | 2020-04-28 | 深圳云天励飞技术有限公司 | High-altitude parabolic monitoring method and device, electronic equipment and storage medium |
-
2020
- 2020-05-12 CN CN202010398931.7A patent/CN111723654B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018095082A1 (en) * | 2016-11-28 | 2018-05-31 | 江苏东大金智信息***有限公司 | Rapid detection method for moving target in video monitoring |
| CN109309811A (en) * | 2018-08-31 | 2019-02-05 | 中建三局智能技术有限公司 | A kind of throwing object in high sky detection system based on computer vision and method |
| CN109711319A (en) * | 2018-12-24 | 2019-05-03 | 安徽高哲信息技术有限公司 | A kind of method and system that grain unsound grain image recognition sample database is established |
| CN109872341A (en) * | 2019-01-14 | 2019-06-11 | 中建三局智能技术有限公司 | A kind of throwing object in high sky detection method based on computer vision and system |
| CN110275042A (en) * | 2019-05-07 | 2019-09-24 | 深圳市零壹移动互联***有限公司 | A kind of throwing object in high sky detection method based on computer vision and radio signal analysis |
| CN110853295A (en) * | 2019-11-12 | 2020-02-28 | 江西赣鄱云新型智慧城市技术研究有限公司 | High-altitude parabolic early warning method and device |
| CN110969604A (en) * | 2019-11-26 | 2020-04-07 | 北京工业大学 | Intelligent security real-time windowing detection alarm system and method based on deep learning |
| CN111079663A (en) * | 2019-12-19 | 2020-04-28 | 深圳云天励飞技术有限公司 | High-altitude parabolic monitoring method and device, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| JING LI 等: "Visual Detail Augmented Mapping for Small Aerial Target Detection", 《REMOTE SENSING》 * |
| 崔艳鹏 等: "一种改进YOLOv3的动态小目标检测方法", 《西安电子科技大学学报》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112329627A (en) * | 2020-11-05 | 2021-02-05 | 重庆览辉信息技术有限公司 | High-altitude throwing object distinguishing method |
| CN112347915A (en) * | 2020-11-05 | 2021-02-09 | 重庆览辉信息技术有限公司 | High-altitude throwing object distinguishing system |
| CN112347915B (en) * | 2020-11-05 | 2024-03-12 | 重庆览辉信息技术有限公司 | High-altitude throwing object distinguishing system |
| CN112329627B (en) * | 2020-11-05 | 2024-02-09 | 重庆览辉信息技术有限公司 | High-altitude throwing object distinguishing method |
| CN112330743A (en) * | 2020-11-06 | 2021-02-05 | 安徽清新互联信息科技有限公司 | High-altitude parabolic detection method based on deep learning |
| CN112308000A (en) * | 2020-11-06 | 2021-02-02 | 安徽清新互联信息科技有限公司 | High-altitude parabolic detection method based on space-time information |
| CN112330743B (en) * | 2020-11-06 | 2023-03-10 | 安徽清新互联信息科技有限公司 | High-altitude parabolic detection method based on deep learning |
| CN112308000B (en) * | 2020-11-06 | 2023-03-07 | 安徽清新互联信息科技有限公司 | High-altitude parabolic detection method based on space-time information |
| WO2022142414A1 (en) * | 2020-12-30 | 2022-07-07 | 深圳云天励飞技术股份有限公司 | High-rise littering monitoring method and apparatus, electronic device, and storage medium |
| CN112686186A (en) * | 2021-01-05 | 2021-04-20 | 润联软件***(深圳)有限公司 | High-altitude parabolic recognition method based on deep learning and related components thereof |
| CN112926445A (en) * | 2021-02-24 | 2021-06-08 | 北京爱笔科技有限公司 | Parabolic behavior recognition method, model training method and related device |
| CN113076809B (en) * | 2021-03-10 | 2023-07-21 | 海纳云物联科技有限公司 | Visual transducer-based high-altitude object detection method |
| CN113076809A (en) * | 2021-03-10 | 2021-07-06 | 青岛海纳云科技控股有限公司 | High-altitude falling object detection method based on visual Transformer |
| CN113112777A (en) * | 2021-04-07 | 2021-07-13 | 浙江磐至科技有限公司 | Wireless transmission security system based on system technology service under monitoring |
| CN113159159A (en) * | 2021-04-15 | 2021-07-23 | 东北大学 | Small sample image classification method based on improved CNN |
| CN113159159B (en) * | 2021-04-15 | 2023-09-29 | 东北大学 | Small sample image classification method based on improved CNN |
| CN113516042A (en) * | 2021-05-17 | 2021-10-19 | 江苏奥易克斯汽车电子科技股份有限公司 | High-altitude parabolic detection method, device and equipment |
| CN113837087A (en) * | 2021-09-24 | 2021-12-24 | 上海交通大学宁波人工智能研究院 | Animal target detection system and method based on YOLOv3 |
| CN113837087B (en) * | 2021-09-24 | 2023-08-29 | 上海交通大学宁波人工智能研究院 | Animal target detection system and method based on YOLOv3 |
| CN114863370A (en) * | 2022-07-08 | 2022-08-05 | 合肥中科类脑智能技术有限公司 | Complex scene high altitude parabolic identification method and system |
| CN117079095A (en) * | 2023-06-25 | 2023-11-17 | 江南大学 | Deep learning-based high-altitude parabolic detection method, system, medium and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111723654B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111723654B (en) | High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization | |
| CN110046631B (en) | System and method for automatically inferring changes in spatiotemporal images | |
| CN107563372B (en) | License plate positioning method based on deep learning SSD frame | |
| US20200065968A1 (en) | Joint Deep Learning for Land Cover and Land Use Classification | |
| CN106845621B (en) | Dense population number method of counting and system based on depth convolutional neural networks | |
| CN108615226A (en) | A kind of image defogging method fighting network based on production | |
| CN111723657B (en) | River foreign matter detection method and device based on YOLOv3 and self-optimization | |
| CN115731164A (en) | Insulator defect detection method based on improved YOLOv7 | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN113920107A (en) | Insulator damage detection method based on improved yolov5 algorithm | |
| CN114022432A (en) | Improved yolov 5-based insulator defect detection method | |
| CN107563299B (en) | Pedestrian detection method using RecNN to fuse context information | |
| CN116343330A (en) | Abnormal behavior identification method for infrared-visible light image fusion | |
| CN113408584B (en) | RGB-D multi-modal feature fusion 3D target detection method | |
| CN111488911B (en) | Image entity extraction method based on Mask R-CNN and GAN | |
| Gleason et al. | A Fusion Approach for Tree Crown Delineation from Lidar Data. | |
| CN111539422B (en) | Flight target cooperative identification method based on fast RCNN | |
| Shen et al. | Biomimetic vision for zoom object detection based on improved vertical grid number YOLO algorithm | |
| CN116385958A (en) | Edge intelligent detection method for power grid inspection and monitoring | |
| CN112991362B (en) | Color sorter adhesion material image segmentation method based on Gaussian mixture model | |
| CN108734200A (en) | Human body target visible detection method and device based on BING features | |
| CN114170144A (en) | Power transmission line pin defect detection method, equipment and medium | |
| Chang et al. | Locating waterfowl farms from satellite images with parallel residual u-net architecture | |
| CN105160285A (en) | Method and system for recognizing human body tumble automatically based on stereoscopic vision | |
| Zhou et al. | Information distribution based defense against physical attacks on object detection |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240122 Address after: No. N3013, 3rd Floor, R&D Building N, Artificial Intelligence Science and Technology Park, Wuhan Economic and Technological Development Zone, Wuhan City, Hubei Province, 430058 Patentee after: Zhongdian Cloud Computing Technology Co.,Ltd. Country or region after: China Patentee after: CHINA ELECTRONIC SYSTEM TECHNOLOGY Co.,Ltd. Address before: No.49 Fuxing Road, Haidian District, Beijing 100036 Patentee before: CHINA ELECTRONIC SYSTEM TECHNOLOGY Co.,Ltd. Country or region before: China |
|
| TR01 | Transfer of patent right |