CN114663827A - Method and system for judging depth of waterlogging in scene - Google Patents
Method and system for judging depth of waterlogging in scene Download PDFInfo
- Publication number
- CN114663827A CN114663827A CN202210192200.6A CN202210192200A CN114663827A CN 114663827 A CN114663827 A CN 114663827A CN 202210192200 A CN202210192200 A CN 202210192200A CN 114663827 A CN114663827 A CN 114663827A
- Authority
- CN
- China
- Prior art keywords
- scene
- waterlogging
- classification
- depth
- model
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
-
- 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
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Geometry (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method and a system for judging the waterlogging depth of a scene, belongs to the technical field of artificial intelligence image recognition, and solves the problems that the waterlogging depth cannot be effectively recognized according to historical data, and auxiliary waterlogging depth measuring equipment has high requirements on environment, angle of video monitoring equipment, illumination and shielding and is difficult to work normally in complex scenes such as rainstorm and the like. The method comprises the following steps: acquiring image information of each video station in real time, extracting scene classification pre-extraction feature data, inputting the scene classification pre-extraction feature data of each video station image into an optimal scene classification model, and acquiring scene classification information of each video station image; and extracting the ponding depth pre-extraction feature data of each video station image, respectively inputting the ponding depth pre-extraction feature data of each video station image into the waterlogging recognition model of the optimal classification scene consistent with the scene of the image, and acquiring the waterlogging depth information of the scene corresponding to each video station image.
Description
Technical Field
The invention relates to the technical field of artificial intelligence image recognition, in particular to a method and a system for judging the depth of waterlogging in a scene.
Background
Urban waterlogging is a recognized urban management problem all the time, so that timely and effective monitoring of urban waterlogging events and prevention of flood disasters are important work contents for urban flood prevention and guarantee.
At present, the early warning of waterlogging in domestic cities is generally carried out by combining weather forecast and video monitoring, the rainfall intensity is forecasted according to a monitoring site, and scene pictures of a corresponding area are mainly checked in a video monitoring mode, so that partial problems are solved to a certain extent, however, the waterlogging in the monitoring site cannot be found timely due to the fact that the area with poor drainage conditions usually has ponding within 5-10 minutes in heavy rainfall, and because the weather forecast data has certain error and uncertainty, the real-time rainfall data has certain hysteresis, and the finding of the waterlogging in the monitoring site cannot be found timely, meanwhile, because the monitoring pictures are more, the pictures are not clear in a rainstorm scene, details are easy to miss, and when the ponding appears in multiple places, the occurrence of the ponding cannot be quickly judged only by manually watching and checking the scene, the risk of missing report and late report exists, and the method is not beneficial to the timely development of the next flood prevention and drainage work.
The method for judging the scene waterlogging depth by adopting the artificial intelligent image recognition technology can improve the urban waterlogging early warning efficiency to a great extent, improve the waterlogging finding capability, reduce the probability of occurrence of the phenomena of missing report and late report, reduce the workload of flood prevention personnel, enable government human resources to be put into higher-level disaster reduction and prevention work, and reduce the urban flood prevention cost. At present, some manufacturers monitor the depth of urban ponding by using an auxiliary water gauge of hardware equipment through an intelligent chip, although the real-time performance is good, the precision is high, the requirement on the environment is high, the requirements on the angle, illumination and shielding of video monitoring equipment are high, the normal work is difficult to realize particularly under the complex scene conditions such as a rainstorm scene, and the upgrading cost is high and the expandability is poor due to the limitation of the deployment site of hardware.
Therefore, a method and a system for judging the depth of the waterlogging in the scene are lacked in the prior art.
Disclosure of Invention
In view of the foregoing analysis, embodiments of the present invention are directed to providing a method and a system for determining a depth of waterlogging in a scene, so as to solve the problems that in the prior art, an image scene cannot be effectively determined according to historical data and the depth of waterlogging cannot be identified, and an auxiliary measuring device for urban waterlogging depth has high requirements on an environment, a video monitoring device angle, illumination and shielding, and is difficult to work normally in a complex scene such as a rainstorm scene.
In one aspect, an embodiment of the present invention provides a method for determining a depth of waterlogging in a scene, including:
acquiring image information of each video station in real time, extracting scene classification pre-extraction feature data of the image, inputting the scene classification pre-extraction feature data of the image of each video station into an optimal scene classification model, and acquiring scene classification information of the image of each video station;
and extracting the accumulated water depth pre-extraction feature data of each video station image, respectively inputting the accumulated water depth pre-extraction feature data of each video station image into an optimal classification scene waterlogging water identification model consistent with the scene of the image, and acquiring the waterlogging accumulated water depth information of the scene corresponding to each video station image.
Further, the obtaining of the optimal scene classification model and the optimal classification scene waterlogging recognition model comprises:
acquiring historical sample data of each video site, preprocessing the historical sample data, performing scene pre-classification on the preprocessed historical sample data, performing waterlogging depth pre-labeling on the historical sample data contained in each type of scene, and constructing a waterlogging image sample library of each classification scene pre-labeling;
carrying out scene classification pre-extraction feature data extraction on data in the waterlogging water image sample library pre-labeled in each classification scene, and training a scene classification model through the scene classification pre-extraction feature data to obtain an optimal scene classification model;
and carrying out waterlogging depth pre-extraction feature data extraction on data in the waterlogging image sample library pre-labeled in each classification scene, and training a waterlogging identification model in each classification scene through the waterlogging depth pre-extraction feature data to obtain a waterlogging identification model in each optimal classification scene.
Further, the classifying the scene type of the scene includes: level ground road, recessed overpass, wide field and residential periphery.
Further, the acquiring image information of each video site in real time includes: polling each video monitoring device, capturing the video monitoring device at regular time in an automatic frame-cutting mode through video acquisition software, and then naming the video monitoring device according to the national standard number uniqueness of the video monitoring device to obtain the image information of each video site.
Further, the optimal scene classification model and the optimal classification scene waterlogging recognition model of each scene have the same structure, and both the optimal scene classification model and the optimal classification scene waterlogging recognition model comprise:
the number of parameters of the characteristic input layer is the same as the number of scene classification pre-extraction characteristic data/accumulated water depth pre-extraction characteristic data;
the hidden layer includes: the first to third hidden layers are used for pre-extracting feature data/accumulated water depth pre-extracting feature data according to the scene classification and acquiring scene classification/accumulated water depth distributed feature parameters;
the full-connection output layer is used for acquiring the spatial mapping characteristic parameters of the waterlogging water identification of the scene classification/each classification scene in a Dropout characteristic mode according to the scene classification/water depth distribution type characteristic parameters;
and the SoftMax normalization layer is used for classifying the scene classification/various classification scene waterlogging identification space mapping characteristic parameters output by the full-connection output layer, outputting the probability of each scene type/various waterlogging depths, and outputting the scene type/waterlogging depth corresponding to the maximum probability as the output of the scene type/waterlogging depth.
Further, in the process of optimizing the scene classification model and the classification scene waterlogging recognition model, model training process errors are calculated through a cross entropy loss function, and when the errors are stable, the scene classification model/the classification scene waterlogging recognition model is the optimal scene classification model/the optimal classification scene waterlogging recognition model.
Further, in the training process of the scene classification model/classification scene waterlogging ponding identification model, the learning rate is dynamically adjusted, and the scene classification model/classification scene waterlogging ponding identification model is optimized through repeated iteration; the dynamically adjusting learning rate includes: setting an initial learning rate value L0And an attenuation rate α; each training N times, the learning rate decays, Li＝α*Li-1Wherein, LiFor the ith round of training, Li-1For the i-1 st round of training, each round is trained N times.
Further, carrying out water logging depth pre-labeling on the historical sample data, and constructing a pre-labeled waterlogging image sample library of each classification scene, wherein the method comprises the following steps: after the historical sample data is subjected to scene classification according to scene types, carrying out waterlogging grading treatment on the classified images according to the scenes, and grading the depth of the waterlogging according to a preset span grade;
the grading of the depth of the accumulated water comprises the following steps: and prejudging according to the weather state in the scene, the water submerging degree of the conventional reference object and the water surface state.
On the other hand, the embodiment of the invention provides a system for judging the depth of waterlogging in a scene, which comprises the following steps:
the data acquisition module is used for acquiring image information of each video station in real time and sending the image information to the scene classification module and the accumulated water depth identification module;
the scene classification module is used for extracting scene classification pre-extraction feature data of the images, inputting the scene classification pre-extraction feature data of the images of the video stations into an optimal scene classification model, and acquiring scene classification information of the images of the video stations;
and the ponding depth identification module is used for extracting the ponding depth pre-extraction feature data of the images, inputting the ponding depth pre-extraction feature data of the images of the video stations into the ponding identification model of the optimal classification scene consistent with the scene of the images respectively, and acquiring the ponding depth information of the scene corresponding to the images of the video stations.
Further, still include: the modeling module is used for obtaining the optimal scene classification model and the optimal classification scene waterlogging recognition model, and comprises:
the system comprises a sample data acquisition unit, a video processing unit and a data processing unit, wherein the sample data acquisition unit is used for acquiring historical sample data of each video station, preprocessing the historical sample data, performing scene pre-classification on the preprocessed historical sample data, performing water accumulation depth pre-labeling on the historical sample data contained in each class of scenes, and constructing a pre-labeled waterlogging and water accumulation image sample library of each class of scenes;
the scene classification model modeling unit is used for carrying out scene classification pre-extraction feature data extraction on data in the waterlogging water image sample library pre-labeled for each classification scene, and training a scene classification model through the scene classification pre-extraction feature data to obtain an optimal scene classification model;
and the classification scene waterlogging identification model unit is used for extracting waterlogging deep pre-extraction feature data of data in the waterlogging image sample library pre-labeled in each classification scene, and training each classification scene waterlogging identification model according to the waterlogging deep pre-extraction feature data to obtain each optimal classification scene waterlogging identification model.
Compared with the prior art, the invention can realize at least one of the following beneficial effects:
1. the method for judging the depth of the waterlogging in the scene can be quickly applied and integrated in a conventional streaming media video monitoring platform, the waterlogging scene can be quickly identified according to video image information, automatic identification and judgment are carried out on the waterlogging events of all scenes in a video image in real time, grading judgment is accurately carried out on the waterlogging depth in the video scene, scientific and convenient technical means can be provided for urban waterlogging early warning work, and good performance is achieved in identification accuracy and real-time performance;
2. the method has the advantages that the finding capability of the urban waterlogging event is enhanced by utilizing the real-time characteristic of video monitoring, the waterlogging information is timely and effectively provided for flood prevention workers, and a scientific and convenient technical means is provided for urban waterlogging early warning work;
3. the method has low requirement on application environment, realizes the judgment of the depth of the scene waterlogging through software and a waterlogging recognition model, is not limited by the environment in deployment, can be widely applied to the extended application of a conventional video streaming media management platform, has low cost of extension and upgrade, and solves the problems that the auxiliary measuring equipment for the depth of the urban waterlogging has high requirements on environment, angle of video monitoring equipment, illumination and shielding and is difficult to work normally under complex scene conditions such as a rainstorm scene and the like;
4. the method and the system can effectively utilize video monitoring data which are already popularized in cities as data sources, extract the depth information of the accumulated water, and greatly reduce the monitoring cost; common ground features are used as a water accumulation reference object to extract the depth of water accumulation, a specific reference object does not need to be installed and empirical parameters do not need to be set, and the method has the advantages of good popularization, low economic cost, high intelligent degree, high precision and easiness in popularization and application in cities;
5. the method for judging the scene waterlogging depth can accurately judge the depth of the waterlogging in the video image, improve the urban waterlogging early warning efficiency and the waterlogging finding capability, reduce the probability of occurrence of phenomena of missing report and late report, reduce the workload of flood prevention personnel, enable government human resources to be put into higher-level disaster reduction and prevention work, and reduce the urban flood prevention cost;
6. the method is combined with an artificial intelligent image recognition algorithm to realize automatic judgment and recognition of the waterlogging scene in the video, the recognition result has the advantages of timeliness, objectivity, repeatability, consistency and the like, the waterlogging in a new environment scene has certain accuracy, the recognition model can be upgraded and optimized to achieve higher accuracy by adding a new environment sample for training, and compared with the conventional method for manually checking the waterlogging, the method improves the urban waterlogging early warning efficiency and the waterlogging finding capability.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout the figures.
FIG. 1 is a flow chart of a method for determining a depth of waterlogging in a scene according to an embodiment of the present application;
FIG. 2 is a flowchart illustrating steps of a method for determining a depth of waterlogging in a scene according to an embodiment of the present application;
fig. 3 is a flowchart of an obtaining method of an optimal classification scene model and an optimal classification scene waterlogging and ponding recognition model according to an embodiment of the present application;
fig. 4 is a block diagram of a system for determining a scene waterlogging depth according to an embodiment of the present disclosure.
Detailed Description
The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and which together with the embodiments of the invention serve to explain the principles of the invention and not to limit its scope.
Artificial Intelligence (AI) is a scientific technology that studies and develops theories, methods, techniques and application systems for simulating, extending and expanding human Intelligence. The invention discloses an image recognition technology based on a DNN (deep neural network), which is an accurate and effective waterlogging image recognition method formed by fusing various neural network models such as a CNN (convolutional neural network) and the like through a deep learning mode on the basis of a traditional ANN (artificial neural network). The method comprises the steps of fully extracting various potential features of an waterlogging scene image by utilizing a multilayer convolutional neural network system, establishing feature mapping and image relations, carrying out classification learning on the image according to sample image features through inputting a large amount of sample image data and a scientific probability statistics principle, and obtaining an optimal feature model parameter result through continuously optimizing feature model parameters. And then classifying and identifying the input image scene according to the characteristic model parameters and probability statistics, and judging to obtain the depth condition of the waterlogging water in the scene.
As shown in fig. 1 and fig. 2, an embodiment of the present invention discloses a method for determining the depth of a scene waterlogging water deposit, including:
s10, acquiring image information of each video station in real time, extracting scene classification pre-extraction feature data of the images, inputting the scene classification pre-extraction feature data of the images of each video station into an optimal scene classification model, and acquiring scene classification information of the images of each video station; specifically, the classifying the scene type of the scene includes: level ground road, recessed overpass, square and residential periphery.
Specifically, the acquiring image information of each video site in real time includes: polling each video monitoring device, capturing the video monitoring device at regular time in an automatic frame-cutting mode through video acquisition software, and then naming the video monitoring device (namely device ID) uniquely according to the national standard number of the video monitoring device to obtain the image information of each video site. More specifically, for the monitoring equipment supporting the national standard GB28181, the video monitoring management platform obtains the video stream address of the RTSP/RTMP protocol of the specified equipment through the national standard number ID of the video equipment, for other standard monitoring equipment, the SDK provided by the equipment manufacturer is used for obtaining the real-time video stream, the timing image capture is carried out in an automatic frame-cutting mode by adopting a video monitoring equipment polling mode and utilizing video acquisition software through a software decoding background playing mode, each monitoring equipment is continuously captured for 3-5 frames/minute, the monitoring equipment is uniquely named according to the national standard number, the channel and the time of the equipment, and the local storage and the cache are carried out in an image mode; for 100-channel video surveillance equipment, the network bandwidth requirement is usually not less than 100M. The method has low requirement on the bandwidth of the network, and is suitable for application environments with a large number of sites and insufficient network bandwidth; optionally, the image is named in a mode of 'device ID _ channel number _ acquisition time', and local storage and caching are performed in a JPG/PNG image format. The embodiment can effectively utilize video monitoring data which are already popularized in cities as data sources, extract the depth information of the accumulated water and greatly reduce the monitoring cost;
through carrying out scene type division with the image of each video website, in different scenes, through this historical ponding data training model of corresponding scene, carry out ponding degree of depth discernment, can reduce scene ponding discernment's complexity, can reduce the emergence of mistake discernment, improve the recognition rate, effectively guarantee different scene ponding degree of depth discernments's accuracy.
And S20, extracting the waterlogging depth pre-extraction feature data of the images according to the video station images, respectively inputting the waterlogging depth pre-extraction feature data of the video station images into an optimal classification scene waterlogging recognition model consistent with the scene of the images, and acquiring the waterlogging depth information of the scene corresponding to the video station images.
Specifically, when the waterlogging accumulated water in the real-time scene is identified, the video stream of the station is obtained according to the ID of the video through the service provided by the video monitoring streaming media management platform, the video image is captured through the image capture program, the range level where the depth of the waterlogging accumulated water in the current image scene is located and the corresponding confidence rate are identified and output through the parameter calculation of the waterlogging accumulated water identification model of the optimal classification scene, and the range level and the corresponding confidence rate are stored into the application database record as a real-time result.
Specifically, as shown in fig. 2 and fig. 3, the obtaining of the optimal classification model and the optimal classification scene waterlogging and ponding identification model includes:
s201, acquiring historical sample data of each video site, preprocessing the historical sample data, performing scene pre-classification on the preprocessed historical sample data, performing water accumulation depth pre-labeling on the historical sample data contained in each type of scene, and constructing a pre-labeled waterlogging image sample library of each classification scene;
specifically, the method for pre-labeling the depth of the waterlogging of the historical sample data and constructing the pre-labeled waterlogging image sample library of each classification scene comprises the following steps: after the historical sample data is subjected to scene classification according to scene types, carrying out waterlogging grading treatment on the classified images according to the scenes, and grading the depth of the waterlogging according to a preset span grade; grading the depth of the accumulated water, comprising: and prejudging according to the weather state in the scene, the water submerging degree of the conventional reference object and the water surface state.
Specifically, after historical sample image data are obtained, preprocessing is carried out according to the brightness and the shielding condition of an image, and classification is carried out according to scene types; more specifically, the historical sample image data is preprocessed, the image is preprocessed after being stored, a severely blocked image (for example, a scene effective area is discarded by < 60%) in the image is discarded, and brightness adjustment compensation is performed on a photo with too high or too low brightness of the image by using software, wherein optionally, the brightness adjustment compensation range is 75% -125%; after the preprocessing is finished, classifying the images according to scenes, namely classifying the preprocessed images according to different scene types, and optionally classifying the scenes according to the flat road, the sunken overpass, the square and the periphery of the house;
secondly, constructing scene-by-scene waterlogging image sample libraries according to the preprocessed and classified scene images, performing waterlogging depth pre-labeling according to waterlogging reference objects, and constructing pre-labeled waterlogging image sample libraries of various classified scenes; specifically, the construction of the pre-labeled inland inundation waterlogging image sample library of each classification scene through historical sample data comprises the following steps: carrying out waterlogging grading treatment on the images classified according to the scenes, grading the depth of waterlogging according to a set span grade, if urban waterlogging can be graded according to 0, 0-1cm, 1-5cm, 5-10cm, 10-20cm,20-30cm and >30cm, manually marking the images, then adopting a depth grade construction file subdirectory mode (for example, constructing file subdirectory according to '00 cm,01cm,05cm,10cm,20cm,30cm and 30+ cm'), and recording and storing the manually marked images into the subdirectory of the corresponding depth grade; the method comprises the steps of (1) prejudging the depth of the ponding according to the submergence degree, the water surface state and the weather state of a conventional reference object in a scene, and extracting the depth of the ponding by taking a common ground object as a ponding reference object as shown in a table 1; the rough range of the water depth can be judged according to the water surface state, for example, a ponding automobile with the water depth of 5-10cm splashes large water flowers, but ponding with the water depth of more than 20cm only has deeper and larger ripples generally, and cannot splash too large water flowers; in addition, the prejudgment result is corrected according to the current weather state, if the current image weather state is heavy rain or rainstorm, a judgment level is usually increased on the level of extracting the depth of the accumulated water by referring to the figure, so that the situation is closer to the actual field situation, and the accumulated effect of the accumulated water in rainfall is better met.
The method provided by the application has the advantages of good popularization, low economic cost, high intelligent degree, high precision and easiness in popularization and application in cities because the image information is acquired without installing specific reference objects and setting experience parameters. Alternatively, conventional references such as: the road surface shoulder (road tooth), the vehicle wheel of going (motor vehicle, bicycle, motorcycle etc.), the bumper, the door handle, preceding bonnet, roof, the peripheral afforestation vegetation of road, pedestrian, road rail etc. the surface of water state mainly with the splash size of vehicle running state, ponding surface of water reflection of light state etc.. Table 1 shows the height range of a reference object in a conventional waterlogging scene, and the estimation and marking are performed according to the height range of the reference object in table 1 according to the relative position of the object and the water surface in the scene image.
TABLE 1
S202, carrying out scene classification pre-extraction feature data extraction on data in the waterlogging image sample library pre-labeled in each classification scene, and training a scene classification model through the scene classification pre-extraction feature data to obtain an optimal scene classification model;
s203, extracting waterlogging depth pre-extraction feature data of the data in the waterlogging image sample library pre-labeled in each classification scene, and training a waterlogging identification model in each classification scene according to the waterlogging depth pre-extraction feature data to obtain a waterlogging identification model in each optimal classification scene.
Specifically, the optimal scene classification model and the optimal classification scene waterlogging recognition model of each scene are deep neural network models with the same structure, and both the models comprise: the device comprises a characteristic input layer, a hidden layer, a full connection output layer and a SoftMax normalization layer;
the number of parameters of the feature input layer is the same as the number of scene classification pre-extraction feature data/accumulated water depth pre-extraction feature data;
the hidden layer includes: the first to third hidden layers are used for pre-extracting feature data/accumulated water depth pre-extracting feature data according to the scene classification and acquiring scene classification/accumulated water depth distributed feature parameters;
the full-connection output layer is used for acquiring the spatial mapping characteristic parameters of the scene classification/waterlogging recognition of each classified scene in a Dropout characteristic mode according to the scene classification/waterlogging depth distribution type characteristic parameters;
and the SoftMax normalization layer is used for classifying the scene classification/various classification scene waterlogging identification space mapping characteristic parameters output by the full-connection output layer, outputting the probability of each scene type/various waterlogging depths, and outputting the scene type/waterlogging depth corresponding to the maximum probability as the output of the scene type/waterlogging depth.
Specifically, in the process of optimizing the scene classification model and the classification scene waterlogging recognition model, the model training process error is calculated through a cross entropy loss function, and when the error is stable, the scene classification model/the classification scene waterlogging recognition model is the optimal scene classification model/the optimal classification scene waterlogging recognition model.
Specifically, in the training process of the optimal scene classification model/optimal classification scene waterlogging recognition model, the learning rate is dynamically adjusted, and the scene classification model/classification scene waterlogging recognition model is optimized through repeated iteration; the dynamically adjusting learning rate includes: setting an initial learning rate value L0And an attenuation rate α; every N training sessions, the learning rate decays once, Li＝α*Li-1Wherein L isiFor the ith round of training, Li-1For the i-1 st round, each round was trained N times. The model can be quickly converged by dynamically adjusting the learning rate, and the precision of model training is improved by gradually reducing the attenuation of the learning rate.
Specifically, the acquisition process of the waterlogging water identification model in the optimal classification scene is described in detail below:
(1) carrying out feature pre-extraction on the ponding depth pre-extraction feature data, comprising the following steps: the programming language adopts Python language, the language is an object-oriented, interpreted and weak script language, an open source artificial intelligent platform Tensorflow is used as a basic model building, training and operating platform of the invention, a conventional image feature extraction model is used for pre-extracting waterlogging water depth feature data of a waterlogging water image sample pre-labeled in each classified scene, for each image in a sample library, all feature data of image features extracted from the image are stored in a feature parameter file, and each image corresponds to a feature parameter file for model training; optionally, the reference model adopts Google inclusion Net-V3/Mobile Net V1, and both models can perform feature pre-extraction operation on the image sample, wherein the Google inclusion Net-V3 has the advantages of acquiring more feature parameters (2048), but has a large model and a slow training speed, and is suitable for a more accurate recognition model; the Mobile Net V1 has the advantages of small volume, moderate characteristic parameters (1001), high model training speed and suitability for rapid identification of application scenes; preferably, the method adopts a Mobile Net V1 network model to extract the water depth pre-extraction feature data of the image sample.
(2) And establishing a deep neural network recognition model, namely establishing a classification scene waterlogging water recognition model for the extracted image features according to an MLP (multi-layer perceptron model structure). The model structure is 1 characteristic input layer, the number of input layer parameters is determined by the number of the pre-extracted characteristic data stored in (1), 3 hidden layers, 1 fully-connected output layer, a Drop characteristic mode (randomly discarding 50% of parameters and reserving 50% of parameters) is adopted in the last hidden layer, all characteristic parameters (number 1024) are output by the fully-connected output layer, and the Drop characteristic mode can be adopted to avoid the occurrence of overfitting, so that the model achieves better effect in an actual scene. The formula of the calculation process of the neural network layer is as follows:
y＝f(WiXi+bi)
wherein XiPre-extracting feature data for the input; y is an output characteristic parameter; wiIs a characteristic weight value; biIs a characteristic offset; i is a feature index;
and finally, performing final classification processing by using a SoftMax normalized index function of a SoftMax normalization layer, outputting the probability of each classification result, and taking the classification with the maximum probability as a final identification result. The output result is set according to actual requirements, the sample and classification stage in the early stage is determined, and optionally, urban ponding can be classified according to 0, 0-1cm, 1-5cm, 5-10cm, 10-20cm,20-30cm and more than 30 cm.
(3) Deep learning, parameter optimization and parameter solidification storage of image features of the classified scene waterlogging recognition model are carried out according to different scenes, specifically, classification is carried out according to scene types, the same training steps are repeated, and the training process comprises 5 sub-steps:
step 1, reading the characteristic parameter files extracted from the waterlogging image sample library pre-labeled by using each classification scene in the step (1), and adding the cached characteristic files with the minimum quantity not less than 5000 required by the minimum quantity of the characteristic files into a training set, a verification set and a test set according to the proportion of 80%, 10% and 10% according to the pre-classification category and the sample quantity;
and 2, inputting the characteristic sample data into the classified scene waterlogging recognition models with consistent scenes of the images for deep learning training by using the classified scene waterlogging recognition models constructed in the step 2. And in the training, the recognition result and the cross entropy result of the predefined label are used for measurement, and the cross entropy calculation formula is as follows:
H(p,q)＝-i＝1∑np(xi)log(q(xi))
p(xi): representing the true distribution of the sample; q (x)i): representing the distribution predicted by the model; i: the number of model classifications; h (p, q): represents the cross entropy of the true distribution p and the predicted distribution q;
true probability distribution p (x) in training optimization processi) And the predicted probability distribution q (x)i) Smaller difference values (cross entropy) between the two represent better predicted results;
step 3, the weight and the offset of the input characteristic parameters are continuously adjusted to minimize the cross entropy H (p, q) so as to achieve the optimal model parameter result,
step 4, in order to enable the model to obtain a convergence result quickly, dynamically adjusting the learning rate LR (Learn rate), and improving the precision of model training by gradually reducing the learning rate to obtain an optimal waterlogging image recognition model for the current training; the initial value of the learning rate is set to be in a range (for example, the range is 0.02-0.08), the attenuation rate is 0.9, the learning rate is attenuated once every 2000 training times, LR is LR 0.9, through repeated iterative calculation, the weight parameters of the deep neural network are optimized, and the precision of model training is gradually improved;
and 5, outputting relevant parameters of the waterlogging recognition model of the optimal classification scene of the training when the specified training times or target accuracy are reached, and solidifying and storing the parameters of the waterlogging recognition model of the optimal classification scene for later-stage model service.
In the embodiment, a large number of waterlogging scene images are collected, a waterlogging recognition model is constructed and trained in an artificial intelligence deep learning mode, recognition results of waterlogging depths are given out in an actual monitoring system by utilizing the waterlogging recognition models of various classified scenes, and the depth of the waterlogging in the actual scene is basically consistent with the depth of the waterlogging in the actual scene through testing, so that the actual waterlogging condition is met, and the requirement of general waterlogging early warning can be met.
Compared with the prior art, the method for judging the depth of the waterlogging in the scene can be quickly applied and integrated in a conventional streaming media video monitoring platform, the waterlogging scene can be quickly identified according to the video image information, the automatic identification and judgment can be carried out on the waterlogging event of each scene in the video image in real time, the graded judgment can be accurately carried out on the waterlogging depth in the video scene, a scientific and convenient technical means can be provided for the early warning work of the urban waterlogging, and the method has good performance in identification accuracy and real-time performance; the method has the advantages that the detection capability of urban waterlogging events is enhanced by utilizing the real-time characteristic of video monitoring, waterlogging information is timely and effectively provided for flood prevention workers, and a scientific and convenient technical means is provided for urban waterlogging early warning work;
the method for judging the depth of the scene waterlogging through the artificial intelligence image recognition technology is provided to overcome the problems that the auxiliary measuring equipment for the depth of the urban waterlogging has high requirements on environment, angle, illumination and shielding of video monitoring equipment and is difficult to work normally under complex scene conditions such as rainstorm scenes and the like; the method and the system can effectively utilize video monitoring data which are already popularized in cities as data sources, extract the depth information of the accumulated water, and greatly reduce the monitoring cost; common ground features are used as ponding reference objects to extract the ponding depth, no specific reference object is required to be installed and empirical parameters are required to be set, the method has good universality, low economic cost, high intelligent degree and high precision, and is easy to popularize and apply in cities; the method for judging the scene waterlogging depth can accurately judge the depth of the waterlogging in the video image, improve the efficiency of urban waterlogging early warning and the waterlogging finding capability, reduce the probability of missing reports and late reports, reduce the workload of flood prevention personnel, enable government human resources to be put into higher-level disaster reduction and prevention work, and reduce the cost of urban flood prevention; the method is combined with an artificial intelligent image recognition algorithm to realize automatic judgment and recognition of the waterlogging scene in the video, the recognition result has the advantages of timeliness, objectivity, repeatability, consistency and the like, the waterlogging in a new environment scene has certain accuracy, the recognition model can be upgraded and optimized to achieve higher accuracy by adding a new environment sample for training, and compared with the traditional method for manually checking the waterlogging, the method improves the urban waterlogging early warning efficiency and the waterlogging finding capability.
As shown in fig. 4, another embodiment of the present invention discloses a system for determining a depth of waterlogging in a scene, including:
the data acquisition module 10 is used for acquiring image information of each video station in real time and sending the image information to the scene classification module and the accumulated water depth identification module;
a scene classification module 20, configured to extract scene classification pre-extraction feature data of the images, input the scene classification pre-extraction feature data of the images at the video sites into an optimal scene classification model, and acquire scene classification information of the images at the video sites;
and the ponding depth identification module 30 is used for extracting the ponding depth pre-extraction feature data of the images, inputting the ponding depth pre-extraction feature data of the images of the video stations into the ponding identification model of the optimal classification scene consistent with the scene of the images respectively, and acquiring the ponding depth information of the scene corresponding to the images of the video stations.
Specifically, the method further comprises the following steps: the modeling module is used for obtaining the optimal scene classification model and the optimal classification scene waterlogging recognition model, and comprises:
the system comprises a sample data acquisition unit, a video processing unit and a data processing unit, wherein the sample data acquisition unit is used for acquiring historical sample data of each video station, preprocessing the historical sample data, performing scene pre-classification on the preprocessed historical sample data, performing water accumulation depth pre-labeling on the historical sample data contained in each class of scenes, and constructing a pre-labeled waterlogging and water accumulation image sample library of each class of scenes;
the scene classification model modeling unit is used for carrying out scene classification pre-extraction feature data extraction on data in the waterlogging water image sample library pre-labeled for each classification scene, and training a scene classification model through the scene classification pre-extraction feature data to obtain an optimal scene classification model;
and the classification scene waterlogging identification model unit is used for extracting waterlogging deep pre-extraction feature data of data in the waterlogging image sample library pre-labeled in each classification scene, and training each classification scene waterlogging identification model according to the waterlogging deep pre-extraction feature data to obtain each optimal classification scene waterlogging identification model.
The specific implementation process of the system embodiment of the present invention may be implemented by referring to the method embodiment described above, and this embodiment is not described herein again. Since the principle of the present embodiment is the same as that of the above method embodiment, the present system also has the corresponding technical effects of the above method embodiment.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program instructing associated hardware, and the program may be stored in a computer readable storage medium. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (10)
1. A method for judging the depth of waterlogging in a scene is characterized by comprising the following steps:
acquiring image information of each video station in real time, extracting scene classification pre-extraction feature data of the image, inputting the scene classification pre-extraction feature data of each video station image into an optimal scene classification model, and acquiring scene classification information of each video station image;
and extracting the ponding depth pre-extraction feature data of each video station image, respectively inputting the ponding depth pre-extraction feature data of each video station image into an optimal classification scene ponding identification model consistent with the scene of the image, and acquiring the ponding depth information of the scene corresponding to each video station image.
2. The method for determining the depth of a scene of waterlogging according to claim 1,
the obtaining of the optimal scene classification model and the optimal classification scene waterlogging recognition model comprises the following steps:
acquiring historical sample data of each video site, preprocessing the historical sample data, performing scene pre-classification on the preprocessed historical sample data, performing waterlogging depth pre-labeling on the historical sample data contained in each class of scene, and constructing a waterlogging image sample library of each classification scene pre-labeling;
carrying out scene classification pre-extraction feature data extraction on data in the waterlogging water image sample library pre-labeled in each classification scene, and training a scene classification model through the scene classification pre-extraction feature data to obtain an optimal scene classification model;
and carrying out water logging depth pre-extraction feature data extraction on data in the pre-labeled waterlogging image sample library of each classification scene, and training a waterlogging recognition model of each classification scene through the water logging depth pre-extraction feature data to obtain a waterlogging recognition model of each optimal classification scene.
3. The method for determining the depth of a scene of waterlogging according to claim 2,
the classifying the scene type of the scene comprises: level ground road, recessed overpass, square and residential periphery.
4. The method for determining the depth of waterlogging water in a scene according to claim 1,
the acquiring image information of each video site in real time includes: polling each video monitoring device, capturing in an automatic frame-cutting mode at regular time through video acquisition software, and then naming according to the national standard number uniqueness of the video monitoring device to obtain the image information of each video station.
5. The method for determining the depth of a scene of waterlogging according to claim 2,
the optimal scene classification model and the optimal classification scene waterlogging recognition model of each scene have the same structure and respectively comprise:
the number of parameters of the feature input layer is the same as the number of scene classification pre-extraction feature data/accumulated water depth pre-extraction feature data;
the hidden layer includes: the first to third hidden layers are used for pre-extracting feature data/accumulated water depth pre-extracting feature data according to the scene classification and acquiring scene classification/accumulated water depth distributed feature parameters;
the full-connection output layer is used for acquiring the spatial mapping characteristic parameters of the scene classification/waterlogging recognition of each classified scene in a Dropout characteristic mode according to the scene classification/waterlogging depth distribution type characteristic parameters;
and the SoftMax normalization layer is used for classifying the scene classification/various classification scene waterlogging identification space mapping characteristic parameters output by the full-connection output layer, outputting the probability of each scene type/various waterlogging depths, and outputting the scene type/waterlogging depth corresponding to the maximum probability as the scene type/waterlogging depth.
6. The method for determining the depth of a scene of waterlogging according to claim 5,
in the process of optimizing the scene classification model and the classification scene waterlogging recognition model, calculating model training process errors through a cross entropy loss function, and when the errors are stable, the scene classification model/the classification scene waterlogging recognition model is an optimal scene classification model/an optimal classification scene waterlogging recognition model.
7. The method for determining the depth of a scene of waterlogging according to claim 6,
dynamically adjusting the learning rate in the training process of the scene classification model/classification scene waterlogging recognition model, and optimizing the scene classification model/classification scene waterlogging recognition model through repeated iteration; the dynamically adjusting the learning rate comprises: setting an initial learning rate value L0And an attenuation rate α; every N training sessions, the learning rate decays once, Li＝α*Li-1Wherein L isiFor the ith round of training, Li-1For the i-1 st round of training, each round is trained N times.
8. The method for determining the depth of a scene of waterlogging according to claim 2,
carrying out waterlogging depth pre-labeling on the historical sample data, and constructing a waterlogging image sample library pre-labeled in each classification scene, wherein the waterlogging depth pre-labeling method comprises the following steps: after the historical sample data is subjected to scene classification according to scene types, carrying out waterlogging grading treatment on the classified images according to the scenes, and grading the depth of the waterlogging according to a preset span grade;
grading the depth of the accumulated water, comprising: and prejudging according to the weather state in the scene, the water submerging degree of the conventional reference object and the water surface state.
9. A system for judging the depth of waterlogging in a scene, comprising:
the data acquisition module is used for acquiring image information of each video station in real time and sending the image information to the scene classification module and the accumulated water depth identification module;
the scene classification module is used for extracting scene classification pre-extraction feature data of the images, inputting the scene classification pre-extraction feature data of the images of the video stations into an optimal scene classification model, and acquiring scene classification information of the images of the video stations;
and the accumulated water depth identification module is used for extracting accumulated water depth pre-extraction feature data of the images, inputting the accumulated water depth pre-extraction feature data of the images of the video stations into an optimal classification scene accumulated water identification model consistent with the scene of the images respectively, and acquiring the accumulated water depth information of the corresponding scene of the images of the video stations.
10. The system for determining scene waterlogging water depth according to claim 9, further comprising: the modeling module is used for obtaining the optimal scene classification model and the optimal classification scene waterlogging recognition model, and comprises:
the system comprises a sample data acquisition unit, a scene pre-classification unit and a data processing unit, wherein the sample data acquisition unit is used for acquiring historical sample data of each video site, pre-processing the historical sample data, performing scene pre-classification on the pre-processed historical sample data, performing ponding depth pre-labeling on the historical sample data contained in each class of scenes, and constructing a ponding image sample library pre-labeled in each class of scenes;
the scene classification model modeling unit is used for carrying out scene classification pre-extraction feature data extraction on data in the waterlogging image sample library pre-labeled for each classification scene, and training a scene classification model through the scene classification pre-extraction feature data to obtain an optimal scene classification model;
and the classification scene waterlogging identification model unit is used for carrying out waterlogging deep pre-extraction feature data extraction on data in the waterlogging image sample library pre-labeled in each classification scene, and training each classification scene waterlogging identification model through the waterlogging deep pre-extraction feature data to obtain each optimal classification scene waterlogging identification model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210192200.6A CN114663827B (en) | 2022-02-28 | 2022-02-28 | Method and system for judging waterlogging and ponding depth of scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210192200.6A CN114663827B (en) | 2022-02-28 | 2022-02-28 | Method and system for judging waterlogging and ponding depth of scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114663827A true CN114663827A (en) | 2022-06-24 |
| CN114663827B CN114663827B (en) | 2023-05-23 |
Family
ID=82027879
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210192200.6A Active CN114663827B (en) | 2022-02-28 | 2022-02-28 | Method and system for judging waterlogging and ponding depth of scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114663827B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115375683A (en) * | 2022-10-24 | 2022-11-22 | 江西省大气探测技术中心 | Image processing-based waterlogging point detection method, system, storage medium and equipment |
| CN116778696A (en) * | 2023-08-14 | 2023-09-19 | 易启科技(吉林省)有限公司 | Visual-based intelligent urban waterlogging early warning method and system |
| CN117012006A (en) * | 2023-08-28 | 2023-11-07 | 浪潮智慧科技有限公司 | Flood disaster early warning method, equipment and medium for urban road |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120087572A1 (en) * | 2010-10-11 | 2012-04-12 | Goksel Dedeoglu | Use of Three-Dimensional Top-Down Views for Business Analytics |
| CN111046773A (en) * | 2019-12-05 | 2020-04-21 | 北京城市排水集团有限责任公司 | Method for judging water retention in pavement based on image technology |
| CN113793069A (en) * | 2021-09-30 | 2021-12-14 | 天津市气象台 | Urban waterlogging intelligent identification method of deep residual error network |
| CN113887532A (en) * | 2021-11-17 | 2022-01-04 | 安徽省公共气象服务中心 | Method for identifying and correcting accumulated snow images of expressway based on scene classification |
-
2022
- 2022-02-28 CN CN202210192200.6A patent/CN114663827B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120087572A1 (en) * | 2010-10-11 | 2012-04-12 | Goksel Dedeoglu | Use of Three-Dimensional Top-Down Views for Business Analytics |
| CN111046773A (en) * | 2019-12-05 | 2020-04-21 | 北京城市排水集团有限责任公司 | Method for judging water retention in pavement based on image technology |
| CN113793069A (en) * | 2021-09-30 | 2021-12-14 | 天津市气象台 | Urban waterlogging intelligent identification method of deep residual error network |
| CN113887532A (en) * | 2021-11-17 | 2022-01-04 | 安徽省公共气象服务中心 | Method for identifying and correcting accumulated snow images of expressway based on scene classification |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115375683A (en) * | 2022-10-24 | 2022-11-22 | 江西省大气探测技术中心 | Image processing-based waterlogging point detection method, system, storage medium and equipment |
| CN116778696A (en) * | 2023-08-14 | 2023-09-19 | 易启科技(吉林省)有限公司 | Visual-based intelligent urban waterlogging early warning method and system |
| CN116778696B (en) * | 2023-08-14 | 2023-11-14 | 易启科技(吉林省)有限公司 | Visual-based intelligent urban waterlogging early warning method and system |
| CN117012006A (en) * | 2023-08-28 | 2023-11-07 | 浪潮智慧科技有限公司 | Flood disaster early warning method, equipment and medium for urban road |
| CN117012006B (en) * | 2023-08-28 | 2024-03-08 | 浪潮智慧科技有限公司 | Flood disaster early warning method, equipment and medium for urban road |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114663827B (en) | 2023-05-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114663827B (en) | Method and system for judging waterlogging and ponding depth of scene | |
| CN110660222B (en) | Intelligent environment-friendly electronic snapshot system for black-smoke road vehicle | |
| CN110929577A (en) | Improved target identification method based on YOLOv3 lightweight framework | |
| CN109816024A (en) | A kind of real-time automobile logo detection method based on multi-scale feature fusion and DCNN | |
| CN111458721B (en) | Exposed garbage identification and positioning method, device and system | |
| CN112668375B (en) | Tourist distribution analysis system and method in scenic spot | |
| CN110956146B (en) | Road background modeling method and device, electronic equipment and storage medium | |
| CN116761049B (en) | Household intelligent security monitoring method and system | |
| US11580659B2 (en) | Method for size estimation by image recognition of specific target using given scale | |
| CN117152513A (en) | Vehicle boundary positioning method for night scene | |
| CN115546742A (en) | Rail foreign matter identification method and system based on monocular thermal infrared camera | |
| CN112785610B (en) | Lane line semantic segmentation method integrating low-level features | |
| CN110765900B (en) | Automatic detection illegal building method and system based on DSSD | |
| KR102525491B1 (en) | Method of providing structure damage detection report | |
| CN113450385B (en) | Night work engineering machine vision tracking method, device and storage medium | |
| CN110163081A (en) | Regional invasion real-time detection method, system and storage medium based on SSD | |
| CN115187880A (en) | Communication optical cable defect detection method and system based on image recognition and storage medium | |
| CN116453033A (en) | Crowd density estimation method with high precision and low calculation amount in video monitoring scene | |
| CN117269011B (en) | Meteorological visibility judging method based on laser light path measuring system | |
| CN117116065B (en) | Intelligent road traffic flow control method and system | |
| CN116665016B (en) | Single-frame infrared dim target detection method based on improved YOLOv5 | |
| CN117953445B (en) | Road visibility measuring method, system and medium based on traffic monitoring camera in rainy days | |
| CN118135505A (en) | Vehicle tracking method, device and medium in fuzzy environment | |
| CN116778696B (en) | Visual-based intelligent urban waterlogging early warning method and system | |
| CN114462315A (en) | Method and device for acquiring confidence of geo-fence and storage medium |
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 |