CN116758469A - Crowd abnormal condition and single person movement state detection method - Google Patents
Crowd abnormal condition and single person movement state detection method Download PDFInfo
- Publication number
- CN116758469A CN116758469A CN202310587781.8A CN202310587781A CN116758469A CN 116758469 A CN116758469 A CN 116758469A CN 202310587781 A CN202310587781 A CN 202310587781A CN 116758469 A CN116758469 A CN 116758469A
- Authority
- CN
- China
- Prior art keywords
- data
- crowd
- target
- environment
- 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.)
- Pending
Links
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 45
- 238000001514 detection method Methods 0.000 title claims abstract description 35
- 230000004927 fusion Effects 0.000 claims abstract description 57
- 238000011156 evaluation Methods 0.000 claims abstract description 45
- 238000004458 analytical method Methods 0.000 claims abstract description 23
- 238000012544 monitoring process Methods 0.000 claims abstract description 6
- 238000011160 research Methods 0.000 claims abstract description 6
- 238000010276 construction Methods 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 82
- 238000000605 extraction Methods 0.000 claims description 39
- 238000012549 training Methods 0.000 claims description 15
- 238000007781 pre-processing Methods 0.000 claims description 13
- 238000013210 evaluation model Methods 0.000 claims description 12
- 238000003062 neural network model Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000001303 quality assessment method Methods 0.000 claims description 8
- 230000000007 visual effect Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000007613 environmental effect Effects 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 5
- 238000013145 classification model Methods 0.000 claims description 4
- 238000006467 substitution reaction Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 3
- 230000015654 memory Effects 0.000 claims description 3
- 238000012847 principal component analysis method Methods 0.000 claims description 3
- 238000013441 quality evaluation Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000011425 standardization method Methods 0.000 claims description 2
- 230000008901 benefit Effects 0.000 abstract description 3
- 239000000463 material Substances 0.000 abstract description 2
- 238000012806 monitoring device Methods 0.000 abstract 1
- 238000005516 engineering process Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 9
- 238000011176 pooling Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 210000000577 adipose tissue Anatomy 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000036772 blood pressure Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000036387 respiratory rate Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/70—Multimodal biometrics, e.g. combining information from different biometric modalities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Human Computer Interaction (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a crowd abnormal condition and single person movement state detection method, which integrates human-vehicle track and flow data generated by a mobile sensing device and an unmanned aerial vehicle monitoring device carried by a single person, and effectively analyzes and excavates the data so as to monitor the single person movement state and discover the crowd abnormal condition, so that possible danger is prevented, and the outdoor execution task of the single person is ensured to be safely and orderly carried out. The invention is applied to single person movement state evaluation and crowd comprehensive situation awareness system construction, and provides technical support for crowd monitoring, personnel/material scheduling, flow control, emergency early warning and the like. Has good commercial application prospect and social benefit. The invention mainly comprises the following contents: 1) Crowd abnormal situation analysis and research based on multi-source heterogeneous data fusion. 2) And (5) evaluating and analyzing single person comprehensive capacity based on multi-perception data fusion.
Description
Technical Field
The invention relates to a data fusion information processing method, in particular to a crowd abnormal condition and single person movement state detection method.
Background
In the information age, personal information is very important. As equipment evolves, particularly multisource heterogeneous sensors improve in performance, a large amount of multisource heterogeneous personal information data is generated. The crowd environment target characteristic information excavated from the data can be used for discovering and identifying crowd single targets, and can also be used for associating and excavating crowd data to master the target activity rule so as to predict the trend and intention of the crowd data and construct an accurate crowd situation map. The high-level characteristics of the crowd environment targets can improve the accuracy of target rule mining and prediction, and the crowd target data is verified, matched and inferred through multi-source data fusion, so that the accurate prediction and judgment of the crowd target activity trend and intention are realized, and the crowd target multi-source data fusion technology is an important informatization technology for the competitive development of countries around the world.
The application range of the multi-source data fusion is wide, the multi-source data fusion has been developed in different fields such as urban big data fusion, medical big data fusion, track big data fusion and the like, and the crowd multi-source data fusion based on personal information is also urgently needed to be researched and developed. Although the multi-source data fusion method has better effects in some scenes and a certain range, the data fusion references aiming at a single field are obtained. At present, related multi-source data fusion particularly aiming at base single person sports collaborative work does not exist, and a data source simultaneously comprises multi-source data fusion work of my and a monitored target. Because the latter is in the environment with high complexity and uncertainty, various sensor nodes and related sensing monitoring equipment such as unmanned aerial vehicle move in high speed three-dimension, and it is difficult to realize rapid hierarchical dynamic resource scheduling, so the method cannot be well applied.
Disclosure of Invention
The invention aims to: the invention aims to solve the technical problem of providing a crowd abnormal condition and single person movement state detection method aiming at the defects of the prior art.
In order to solve the technical problems, the invention discloses a crowd abnormal condition and single person movement state detection method, which comprises the following steps:
step 1, collecting multi-source heterogeneous data in a crowd environment, wherein the multi-source heterogeneous data comprises: image data, overall environment data, and text information data;
step 2, crowd abnormal situation analysis and research based on multi-source heterogeneous data fusion comprises the following steps:
step 2-1, establishing a crowd environment-oriented multi-target recognition model aiming at image data, and carrying out crowd environment-oriented multi-target recognition;
the crowd environment-oriented multi-target recognition model consists of a crowd target quantity estimation module and a crowd target detection output module;
the crowd target number estimating module comprises the following steps:
firstly, initializing a multi-view module aiming at collected image data, wherein the multi-view module consists of a feature extraction layer, a view coding layer, a view fusion layer and a final classification layer; the multi-view module initialization includes three steps: firstly, establishing a crowd information classification model based on a multi-view image training data set, wherein the model comprises basic quintuple, and the basic quintuple comprises: the method comprises the steps of describing a group of people environment, collecting individual targets in a group of people, collecting individual target positions in a group of people, collecting position information of personnel carrying equipment in a group of people and collecting position information of equipment carried by individuals in a group of people, and performing fine-grained classification on image data collected on site; secondly, independently initializing the image data of multiple visual angles; thirdly, an image standardization module is constructed to correct the crowd environment image data; after the multi-view module initialization task is completed, fusing the data in the manually collected crowd information database with the multi-view module to form multi-view environment information; carrying out character feature extraction and fine granularity classification de-duplication on the multi-view environment information by adopting a Resnet50 deep neural network model, and completing the construction of the crowd target number estimation module;
The crowd target detection output module comprises the following steps:
firstly, extracting crowd information based on a crowd information database acquired manually, preprocessing crowd image information and loading a crowd abnormal state monitoring model according to the crowd information database; performing target feature extraction by using a Darknet53 deep convolutional neural network, and obtaining a final target detection result by using an image feature pyramid method; finally, adopting feature layer decoding regression classification, extracting image features on multiple scales based on an image feature pyramid by using a deep neural network, extracting high-level features of an original image by using a single neural network, and performing layer-by-layer downsampling to obtain feature layers with different scales, thereby completing the output of crowd target detection.
2-2, establishing target recognition based on multi-source heterogeneous data aiming at individual environment data, and adopting an infrared fusion detection method, wherein the specific method is as follows:
the infrared image data is enhanced by utilizing an infrared information base acquired manually, then infrared human feature extraction is carried out, and feature stitching is carried out by adopting a PANet neural network model; and screening the confidence coefficient in the target identification result by adopting a confidence coefficient threshold filtering method, and eliminating the detection result with lower confidence coefficient.
Step 2-3, establishing a crowd detail analysis model based on multiple visual angles according to the target recognition results in the step 2-1 and the step 2-2, and carrying out crowd detail analysis based on multiple visual angles;
step 2-4, aiming at text information data, performing target situation analysis based on multi-mode data, wherein the method specifically comprises the following steps:
aiming at text information data, multi-mode data is adopted for analysis, wherein the multi-mode data comprises image information and text information; establishing a text decoding model, and extracting key information from a text information base; fusing the target characteristic information and text information extracted from the multi-view module; and detecting and outputting the fused information by using confidence threshold filtering to finish text fusion detection, namely analyzing the target situation.
Step 3, collecting multi-perception data for single person targets in the crowd, wherein the multi-perception data comprises: vital sign data, equipment data, and individual environmental data;
step 4, single person movement state detection based on multi-perception data fusion comprises the following steps:
step 4-1, building a physical quality assessment model according to vital sign data, wherein the specific method comprises the following steps:
collecting single basic physiological data and preprocessing, wherein the preprocessing comprises the steps of carrying out abnormal data processing on physiological features with missing values by adopting a mean value substitution method and carrying out L2 norm feature standardization preprocessing; training to obtain a long-short-period memory neural network model as a physical quality assessment model according to the preprocessed acquired data, and outputting a single body state result, namely a single body quality assessment result;
The mean value replacement method specifically comprises the following steps:
the first step: firstly, sorting physiological characteristic data from small to large in ascending order;
and a second step of: calculating the position: the position of the calibration parameter Q1 is (n+1) ×0.25; the position of the calibration parameter Q2 is (n+1) ×0.5; the position of the calibration parameter Q3 is (n+1) multiplied by 0.75, and n is the total number of data;
and a third step of: calculating the quantile value:
if the calculation result in the second step is an integer, directly taking the numerical value corresponding to the position;
if the result of the second step is a decimal place, the value corresponding to the 2 nd place is (1-decimal part) +the value corresponding to the 3 rd place is the decimal part;
fourth step: calculating a quartile range IQR:
IQR＝Q3-Q1
fifth step: judging whether the physiological characteristic data is an abnormal value or not:
if the physiological characteristic data exceeds the range of [ (Q1-1.5 x IQR) to (Q3 +1.5 x IQR) ] then judging the physiological characteristic data as abnormal values;
sixth step: calculating a physiological characteristic data mean value;
seventh step: replacing the abnormal value screened in the fifth step with the average value;
the L2 norm characteristic standardization pretreatment is carried out, and is concretely as follows:
physiological characteristic data vectors x (x 1, x2, …, x n ) The L2 norm of (2) is defined as:
wherein x is n Representing nth physiological characteristic data; to normalize x to the L2 norm, i.e., build a mapping from x to x ', such that the L2 norm of x' is 1, each dimension of the vector x is divided by II x II 2 Obtaining a new vector X 2 I.e.
Step 4-2, establishing a single equipment state evaluation model according to equipment data, wherein the specific method comprises the following steps:
collecting single equipment data, adopting the maximum normalized characteristic processing, and scaling different characteristic values to the same range to obtain the equipment characteristic data; performing feature extraction on the equipment feature data by using a principal component analysis method, and removing noise and redundant features by mapping high-dimensional data into a low-dimensional space to finish dimension reduction feature extraction; and training based on the processed data to obtain a convolutional neural network model as a single equipment state evaluation model, wherein the convolutional neural network model is used for evaluating the single equipment state to obtain a single equipment state evaluation result.
Step 4-3, establishing a crowd environment state evaluation model according to the environment data, wherein the specific method comprises the following steps:
collecting environmental data in the environment where people are located, preprocessing by adopting a zero-mean value standardization method, ensuring that the mean value of the data is 0 and the standard deviation is 1, and scaling the preprocessed data to the same scale; performing feature extraction on the scaled data by adopting a linear discriminant method, projecting high-dimensional data into a low-order space, and completing maximization of the distance between different categories and minimization of the distance inside the same category, thereby completing dimension reduction and classification of the data; based on the processed data, training to obtain a circulating neural network model as a crowd environment state evaluation model, and outputting crowd environment evaluation results.
Step 4-4, establishing a single comprehensive capacity evaluation model according to the three models, and performing single comprehensive capacity evaluation, wherein the method specifically comprises the following steps:
a single body quality evaluation result, a single equipment state evaluation result and a crowd environment evaluation result are fused through a variation coefficient method and a weighted rank sum ratio method, and a single comprehensive capacity score is obtained through calculation;
the variation coefficient method is used for determining the weight of the evaluation index, and the specific method is as follows:
the method for calculating the variation coefficient of the three indexes of psychological stress, exercise capacity and equipment state comprises the following steps:
CV l ＝S l /X l ,(l＝1,2,3)
in CV l Is the variation coefficient of the first evaluation index, S l Is the standard deviation of the first evaluation index, X l Is the average of the first evaluation index;
then calculate the weight W of each index l The method comprises the following steps:
in which W is l Is the weight of the first evaluation index, m represents the number of the evaluation indexes;
the weighted rank sum ratio method is specifically as follows:
WRSR＝∑RW/m
wherein WRSR represents a weighted rank sum ratio, RSR represents a rank sum ratio, Σr represents a rank sum value of an evaluation target index, W represents a weight of each evaluation index, m represents the number of evaluation indexes, k represents, and l represents; firstly, the high-priority index is ranked from small to large, the low-priority index is ranked from large to small, and the indexes with the same numerical value are ranked evenly.
The beneficial effects are that:
the method provided by the invention is applied to single person movement state evaluation and crowd comprehensive situation awareness system construction, and provides technical support for crowd monitoring, personnel/material scheduling, flow control, emergency early warning and the like. Has good commercial application prospect and social benefit.
Drawings
The foregoing and/or other advantages of the invention will become more apparent from the following detailed description of the invention when taken in conjunction with the accompanying drawings and detailed description.
FIG. 1 is a diagram of a crowd abnormal situation analysis and study framework based on multi-source heterogeneous data fusion.
FIG. 2 is a diagram of a single person comprehensive ability assessment framework for multi-perception data fusion.
FIG. 3 is a diagram of a crowd abnormal situation system architecture for multi-source heterogeneous data fusion.
Fig. 4 is a schematic structural diagram of a multi-target recognition model facing crowd environment.
FIG. 5 is a diagram of a single integrated capability system architecture for multi-sensor data fusion.
FIG. 6 is a block diagram of a single quality assessment subsystem.
Fig. 7 is a block diagram of an equipment status evaluation subsystem.
FIG. 8 is a block diagram of an environmental status sub-assessment system.
FIG. 9 is a diagram of a single person comprehensive ability assessment model.
Detailed Description
The invention is further described below with reference to the drawings and the detailed description.
The invention mainly solves the problems of diversity and complexity of sensing data such as images, texts, environments, single person states and the like acquired in crowd environments, integrates the integral characteristics of multi-view data based on the intelligent algorithm high-efficiency data fusion technology, digs the correlation of the multi-mode data, and respectively carries out deep analysis on single person information and my single person information based on the intelligent data fusion technology. The crowd abnormal situation analysis and research scheme based on multi-source heterogeneous data fusion is shown in fig. 1, and the single person comprehensive ability evaluation scheme based on multi-perception data fusion is shown in fig. 2.
The invention is mainly divided into two parts: crowd abnormal situation analysis and research based on multi-source heterogeneous data fusion and single person comprehensive capacity assessment based on multi-perception data fusion.
1) Crowd abnormal situation analysis and research based on multi-source heterogeneous data fusion
As shown in fig. 3, the present invention is composed of 4 parts: the method comprises the steps of (1) constructing a multi-target recognition model for crowd environment, (2) constructing a target recognition model based on multi-source heterogeneous data, (3) analyzing crowd details based on multi-view angles, and (4) analyzing a target situation based on multi-mode data.
Firstly, establishing a crowd environment-oriented multi-target recognition model aiming at image data, and carrying out crowd environment-oriented multi-target recognition, wherein the crowd environment-oriented multi-target recognition model mainly comprises a crowd target number estimation module and a crowd target detection output module;
The implementation process of the crowd target number estimation module comprises the following key steps:
firstly, initializing a multi-view module according to collected multi-angle crowd environment image data. The task includes three steps:
firstly, establishing a crowd information classification model based on a multi-view image training data set, wherein the crowd information classification model is used for carrying out fine granularity classification on image data acquired on site; specifically, the model established by the invention comprises basic quintuple:
{E,T,TP,PP,P}
e is a group environment description set:
E＝{Temperature,Humidity,Visibility,Others}
temperature represents the crowd environment Temperature condition, humidi represents the crowd environment Humidity condition, visiability represents the crowd environment Visibility condition, and other conditions in the crowd environment are represented by other.
T represents a target set of individuals in the population:
T(Target)＝{T i |i＝1,2...NT}
wherein NT is the target number of individuals in the population;
TP represents a set of individual target locations in a population:
TP(Target)＝{TP i |i＝1,2...NT}
PP represents a set of location information of carrying equipment personnel in a crowd:
PP(Person)＝{PP i |i＝1,2...NT}
wherein NT is the number of people carrying equipment;
p represents a set of location information for individuals in the population carrying equipment:
P(Equipment)＝{PE i |i＝1,2...NT}
wherein NT is the number of individuals in the population carrying equipment;
secondly, independently initializing the multi-view data by adopting a micro-service technology so as to construct complete multi-view crowd environment data later;
Thirdly, an image standardization module is constructed to correct the crowd environment image data. Firstly, collecting crowd environment image data from image data sets of different sources, and preprocessing; then, performing color space conversion, and converting the image data from RGB color space to LAB color space, so as to reduce color deviation and illumination variation in the image; and finally, processing the image by using a morphological filter and performing scale transformation to complete the image data correction task.
After the multi-view module initialization task is completed, the invention fuses the data in the manually collected crowd information database with the multi-view module to form multi-view environment information [1] . Then, a Resnet50 deep neural network model is adopted [2] And extracting the character features and carrying out fine-granularity classification and duplication removal on the multi-view environment information so as to realize the construction task of the crowd target number estimation module.
The implementation of the crowd target detection output module firstly needs to carry out crowd information extraction work based on a crowd information database which is acquired manually, and carries out crowd image information preprocessing and loading tasks of a crowd abnormal state monitoring model according to the database, wherein the characteristic extraction task of the module adopts a Darknet53 deep convolution neural network which is a part of a YOLOv3 target detection algorithm. In the feature extraction task, targets may appear in images in different scales and sizes, so the invention adopts the image feature pyramid technology to solve the problem. Specifically, at the bottom of the pyramid, the original input image is processed to the smallest size, while also being the highest resolution. Then, the feature map of the layer is extracted through the convolutional neural network, while the image size is reduced, and the feature map of the layer is also extracted through the convolutional neural network, and the process is repeated until the top of the pyramid is reached. Targets can be detected on each scale of the pyramid, and final target detection results can be obtained by integrating and screening the detection results. Finally, the invention adopts a feature layer decoding regression classification technology, extracts image features on a plurality of scales by using a depth neural network based on an image feature pyramid, extracts high-level features of an original image by using a single neural network, and performs layer-by-layer downsampling to obtain feature layers with different scales, so that the model of the invention can sense target features with different scales at the same time, thereby realizing the output of crowd target detection;
In summary, as shown in fig. 4, the multi-objective recognition model for crowd environment designed by the present invention mainly includes two phases, namely a feature extraction phase and a metric learning phase, wherein the feature extraction phase is divided into three parts, namely single-view feature extraction, multi-view feature extraction and overall view feature extraction, and the metric learning carries out similarity metric learning on the AC image features obtained in the feature extraction phase, so as to complete multi-class classification and multi-class quantity estimation of individual targets in crowd. The single-view feature extraction inputs the images shot by the crowd into a feature extraction network, the low-order general feature extraction is completed through a convolution layer and a pooling layer, then the ROI pooling is carried out, and the high-order special feature extraction is carried out on the images shot by the crowd, so that the features of the images shot by the crowd with the single view are obtained; after the input multiple crowd shooting images are subjected to primary feature extraction through a convolution layer, the images are respectively subjected to horizontal pooling to finish local feature extraction and global feature extraction, and when global feature extraction is carried out, the number of channels of the feature images is required to be properly reduced, so that global features of the crowd shooting images are obtained. Then fusing the global features and the horizontal local features to obtain multi-view features of the crowd input images; the overall view angle feature extraction inputs the overall individual images of the crowd into a feature extraction network, the low-order general feature extraction is completed through a convolution layer and a pooling layer, then the ROI pooling is carried out, and the high-order special feature extraction is carried out on the overall view angle images to obtain the overall view angle features. After three kinds of features are extracted, the three kinds of features are sent to a measurement learning network for multiple image similarity learning, and the image features after measurement learning are subjected to multi-class classifier and multi-class number prediction processing to complete the multi-target recognition model building task of the crowd shooting images.
Then, the second part is to establish target recognition based on multi-source heterogeneous data aiming at the whole environment data, and the step is mainly realized by adopting an infrared fusion detection technology;
the invention utilizes the infrared information base collected manually to enhance the infrared image data, thereby realizing the extraction of infrared human features. In addition, the PANet neural network model is adopted to carry out feature splicing, so that the problems of information loss and scale mismatch between different layers of features in the feature pyramid network are solved, and global feature representation is generated to better identify and position the target. In the crowd target detection process, crowd characteristic fusion based on a concentration mechanism is used, and weighting processing of information is achieved by calculating weights of different parts in input data, so that a model concentrates more attention on a data part related to a target task, and the performance of the model is improved. Finally, the confidence coefficient in the detection result is screened by adopting a confidence coefficient threshold filtering technology, and the detection result with lower confidence coefficient is eliminated, so that the detection precision and the robustness of the model are improved.
The third part is to establish a crowd detail analysis model based on multiple visual angles according to the target recognition results realized by the two parts, and to perform crowd detail analysis based on multiple visual angles; firstly, the invention utilizes the information obtained in the target recognition result, such as the number of people, the position, the behavior and the like, and combines the multi-view data source to carry out comprehensive analysis. The multi-view data source may include image data, sensor data, portable device social media information, etc. to obtain richer crowd details. Secondly, aiming at the whole environment data, the invention adopts the infrared image data and the image enhancement technology to carry out data enhancement and correction on the single source data so as to obtain more accurate crowd detail analysis.
Finally, the fourth section aims at providing a target situation analysis method based on multi-modal data. Specifically, for text information data, the invention adopts multi-modal data for analysis, including image information and text information. In order to fully utilize the text information, the invention establishes a text decoding model so as to extract key information from a text information base. On the basis, the method and the device realize the loading work of the multi-mode fusion model, and fuse the target characteristic information and the text information. And finally, detecting and outputting the fused information by using confidence threshold filtering, thereby achieving the aim of text fusion detection.
For a text decoding model, the invention refers to a model for extracting key information from a text information base. The model may convert the text information into a computer-processable representation, such as a vector or matrix. The structure of the text decoding model consists of two parts: an encoder and a decoder. The task of the encoder is to encode the input representation of the features into a fixed length vector, which is processed using convolutional neural network structures. The encoder typically considers various time steps and hierarchies in processing the input features in order to better understand the semantics of the input. The task of the decoder is to decode the fixed length vector output by the encoder into a piece of natural language text. The decoder adopts a transducer structure, and the word generated in the last step, the characteristics of the current time step and the text generated before are considered in the generation process so as to generate smoother and more accurate natural language text. The structure of the entire text decoding model is an end-to-end model, the input of which is a set of text input features and the output of which is a piece of natural language text. The training process of the model usually adopts methods such as maximum likelihood estimation, and the like, and optimizes model parameters by minimizing the difference between the text generated by the model and the real text. This representation may be used by other models, such as a multimodal fusion model.
Aiming at the multi-mode fusion model, the method disclosed by the invention is used for fusing information from different modes together so as to obtain more complete and accurate information. In the present invention, the multimodal information includes text information and image information. The fusion model adopts deep neural network and other technologies to fuse the information. In the invention, the multi-mode fusion model structure consists of a plurality of mode feature extractors, fusion layers and classifiers, wherein the feature extractor of each mode adopts different deep neural networks to extract the features of input data. The fusion layer combines the features of different modes and uses the modes of weighted summation, splicing, dot product and the like. Finally, classifying the input according to the fusion result of the features by a classifier; the multi-mode fusion model is divided into two modes of early fusion and late fusion, wherein the early fusion refers to the process that the input data of a plurality of modes are firstly subjected to feature extraction and then fused. Late fusion refers to adding a fusion layer between the feature extractor and the classifier of each mode, and fusing the features of different modes before the classifier. The multi-mode fusion model is used for fusing the target characteristic information and the text information so as to realize text fusion detection output. By fusing the information together, the accuracy and efficiency of the target situation analysis can be improved.
2) Design and implementation of single comprehensive capacity assessment method based on multi-perception data fusion
As shown in fig. 4, the present invention is composed of 4 parts: the method comprises the steps of (1) building a physical quality assessment model, (2) building a single equipment state assessment model based on equipment data, and (3) building a crowd environment state assessment model based on multi-source data, and (4) designing and realizing a single comprehensive capacity scoring algorithm.
The first part of the invention aims at building a physical fitness assessment model from vital sign data; firstly, basic physiological data such as height, weight, body fat rate, heart rate, blood pressure and respiratory rate of an individual are collected, abnormal data processing is carried out by adopting a mean value substitution method aiming at physiological features with missing values, and specifically, the invention calculates the mean value of the features in the existing data and replaces the missing feature values for the features containing the missing values. For the acquired data, the method performs L2 norm characteristic standardization preprocessing so as to normalize the data characteristics according to different scaling proportions, so that the characteristic values are consistent in order of magnitude, and the influence of differences among the data characteristics on model training is avoided. The invention adopts a long-term and short-term memory network with time sequence memory capability to learn the change of the body state of the single person, analyzes the current vital sign of the single person, and evaluates the physical quality of the single person in five aspects of pressure, endurance, speed, strength and flexibility. The learning training process of the physical fitness assessment model is shown in fig. 5.
The invention adopts a mean value substitution method to process abnormal data aiming at physiological characteristics with missing values, and particularly, the invention calculates the mean value of the characteristics containing the missing values in the existing data and replaces the missing characteristic values. Because the sampling frequency of single human sign data is higher, the instability of the data is increased, and the body sensor is easily influenced by factors such as the distance between the sensor and the human body, the external environment and the like when the body sensor collects data such as heart rate, temperature and the like, the collected data mostly have abnormal values. In order to compensate the prediction error caused by the abnormal value, the invention uses a quartile method to screen out the possible abnormal value, and adopts a mean value replacement method to process the abnormal value so as to better preserve the real situation of the data.
Quartiles have very good effects when used to explore data distributions, which compensates for the defect that average values are "susceptible to outliers". The quartile mean value replacement method comprises the following steps:
the first step: the data is sorted from small to large in ascending order.
And a second step of: calculate position (n is the total number of data): position of Q1= (n+1) ×0.25; position of Q2= (n+1) ×0.5; position of Q3= (n+1) ×0.75.
And a third step of: calculating the quantile value:
1) If the second step is that the calculation result is an integer, the numerical value corresponding to the position is directly taken.
2) If the second step is a decimal, such as 2.25: the number corresponding to position 2 is (1-0.25) +the number corresponding to position 3 is 0.25.
Fourth step: calculating a quartile range IQR:
IQR＝Q3-Q1
fifth step: judging whether the data is abnormal value
If the data is outside the [ (Q1-1.5×iqr) to (q3+1.5×iqr) ] range, it is considered an outlier.
Sixth step: calculating data mean
Seventh step: by usingReplacing the outlier x screened in the fifth step
Then, for the data after exception processing, the invention performs L2 norm characteristic standardization pretreatment so as to normalize the data characteristics according to different scaling proportions, thereby keeping the characteristic values consistent in order of magnitude and avoiding the influence of the difference between the data characteristics on model training.
The L2 norm characteristic standardized pretreatment method adopted by the invention comprises the following steps:
the L2 norm of vector x (x 1, x2, …, xn) is defined as
To normalize x to the L2 norm, a mapping from x to x 'is established such that the L2 norm of x' is 1. The specific processing operation is that each dimension data X1, X2, …, xn of the vector X is divided by X2 to obtain a new vector, namely
After L2 norm normalization, the euclidean distance of single sign data and their cosine similarity can be equivalent.
According to the preprocessed acquired data, the long-term and short-term memory neural network model is obtained through training. The model can output a single body state result, and the single body state result is presented in the form of a single body quality five-dimensional radar chart and a single body quality evaluation text result;
as shown in fig. 7, a second aspect of the present invention is directed to a method of establishing a single person equipment status assessment model based on equipment data; therefore, the invention firstly collects the multi-source data of single person communication state, peripheral state, computer, tablet, mobile phone, unmanned plane equipment and the like.
For the data, the invention adopts the maximum normalized characteristic processing to scale different characteristic values to the same range, thereby avoiding dimension problems among different characteristic values and ensuring that the contribution degree of each characteristic to the model is relatively balanced. The conversion formula of the maximum normalization adopted by the invention is as follows:
next, the present invention performs feature extraction on the feature data using a principal component analysis method, which performs dimension reduction and feature extraction of the data by mapping high-dimensional data into a low-dimensional space, retaining main information in the data, and removing noise and redundant features. The invention carries out aggregation operation on equipment states based on the thought of weighted summation, and comprises the following basic steps:
(1) Determining at which particular stage each maturity capability attribute element is, and determining the relative importance between them;
(2) Determining the relative importance between the maturity capabilities according to the maturity levels of the capability attributes based on the weighted summation rules;
(3) The performance maturity level of the equipment is evaluated according to a weighted summation based rule.
Wherein the maturity level of the physical domain action capability, the information domain utilization capability, the cognitive domain understanding capability and the social domain management capability is Cap respectively i (i=1, 2,3, 4) and the weight set between them is w= { W i (i=1, 2,3, 4) }, the maturity levels of the individual maturity capability attribute elements are Cap, respectively ij (j＝1,2,3,...,k i ) The weight set between them is W i ＝{w ij }，(k 1 ，k 2 ，k 3 ，k 4 )＝(7，5，4，4)。
Finally, as shown in fig. 6, the invention trains to obtain a convolutional neural network model based on the processed data, and is used for evaluating the state of single equipment;
then, a third part is shown in fig. 8, and the invention provides a crowd environment state evaluation model based on multi-source data aiming at environment data; firstly, the invention collects the multi-source environmental data such as weather, temperature, humidity, air pressure, wind speed, altitude and the like in the environment where the crowd is located.
According to the data, the feature data is preprocessed by adopting a zero-mean normalization method, the mean value of the data is guaranteed to be 0, the standard deviation is guaranteed to be 1, the data features are scaled to the same scale, the problem of inconsistent dimensions among the features is avoided, and the training speed and the generalization capability of the model are improved. The invention firstly obtains the following data when zero-mean normalization is carried out on the data:
1) Mean value of overall data (mu)
2) Standard deviation (sigma) of the population data, this population being of the same order of magnitude as the population in 1).
3) Individual observations (x)
By substituting the above three values into the zero normalized formula, namely:
therefore, the invention can convert different data to the same magnitude, and realize standardization.
Then, the invention adopts a linear discriminant method to extract the characteristics of the characteristic data, namely, the high-dimensional data is projected into a low-order space, so as to realize the maximization of the distance between different categories and the minimization of the distance inside the same category, thereby realizing the classification of the data. Based on the preprocessed collected data, the method trains to obtain a cyclic neural network model so as to output crowd environment assessment results;
finally, as shown in fig. 9, according to the three models, the single comprehensive ability score is obtained by integrating the single physical quality score, the single equipment state score and the crowd environment state score through a variation coefficient method and a weighted rank sum ratio method.
The model algorithm mainly comprises two parts of calculating the evaluation result weight of the single body quality, the crowd environment state and the single equipment state by a variation coefficient method and the single comprehensive capacity by a weighted rank sum ratio method, and the specific implementation method of the algorithm comprises the following steps:
1) Determination of evaluation index weight by coefficient of variation method
Firstly, calculating the variation coefficients of three indexes of psychological stress, exercise capacity and equipment state, wherein the formula is
CV j ＝S j /X j ,(j＝1,2,3)
In CV j Is the variation coefficient of the j-th evaluation index, S j Is the standard deviation of the jth evaluation index, X j Is the average of the j-th evaluation index.
Then calculate the weight W of each index j The formula is
In which W is j Is the weight of the j-th evaluation index.
2) Weighted rank sum ratio
The basic idea of the Rank-sum ratio (RSR method for short) is that in an n-row (n evaluation objects) m-column (m evaluation indexes or grades) matrix, a dimensionless statistic RSR is obtained through Rank conversion, and the merits of the evaluation objects are ordered by RSR values.
The formula is expressed:
WRSR＝∑RW/n
wherein Σr represents the rank sum value of the evaluation target index, W represents the weight of each evaluation index, and n represents the number of evaluation indexes. Firstly, the high-priority index is ranked from small to large, the low-priority index is ranked from large to small, and the indexes with the same numerical value are ranked evenly.
3) Step ordering
The rank order and the frequency f of each WRSR are determined, the accumulated frequency is obtained, and then the corresponding probability unit is obtained. And according to the probability unit and the regression equation calculation result, sequencing the single comprehensive capacity by utilizing the result and the optimal grading principle.
The invention provides a method for detecting abnormal situations of crowd and single person movement states, and a method for realizing the technical scheme, wherein the method and the method are a plurality of methods, the method is only a preferred embodiment of the invention, and it should be pointed out that a plurality of improvements and modifications can be made by one of ordinary skill in the art without departing from the principle of the invention, and the improvements and modifications are also considered as the protection scope of the invention. The components not explicitly described in this embodiment can be implemented by using the prior art.
Claims (10)
1. The method for detecting the abnormal conditions of the crowd and the movement states of the single person is characterized by comprising the following steps:
step 1, collecting multi-source heterogeneous data in a crowd environment, wherein the multi-source heterogeneous data comprises: image data, overall environment data, and text information data;
step 2, crowd abnormal situation analysis and research based on multi-source heterogeneous data fusion;
step 3, collecting multi-perception data for single person targets in the crowd, wherein the multi-perception data comprises: vital sign data, equipment data, and individual environmental data;
and 4, detecting the single person movement state based on multi-perception data fusion.
2. The method for detecting abnormal crowd conditions and single person exercise state according to claim 1, wherein the analyzing and researching abnormal crowd conditions based on multi-source data fusion in the step 2 comprises the following steps:
step 2-1, establishing a crowd environment-oriented multi-target recognition model aiming at image data, and carrying out crowd environment-oriented multi-target recognition;
step 2-2, establishing target recognition based on multi-source heterogeneous data aiming at individual environment data;
step 2-3, establishing a crowd detail analysis model based on multiple visual angles according to the target recognition results in the step 2-1 and the step 2-2, and carrying out crowd detail analysis based on multiple visual angles;
and 2-4, analyzing the target situation based on the multi-mode data aiming at the text information data.
3. The method for detecting abnormal crowd conditions and single person movement states according to claim 2, wherein the crowd environment-oriented multi-target recognition model in step 2-1 is composed of a crowd target number estimation module and a crowd target detection output module;
the crowd target number estimating module comprises the following steps:
firstly, initializing a multi-view module aiming at collected image data, wherein the multi-view module consists of a feature extraction layer, a view coding layer, a view fusion layer and a final classification layer; the multi-view module initialization includes three steps: firstly, establishing a crowd information classification model based on a multi-view image training data set, wherein the model comprises basic quintuple, and the basic quintuple comprises: the method comprises the steps of describing a group of people environment, collecting individual targets in a group of people, collecting individual target positions in a group of people, collecting position information of personnel carrying equipment in a group of people and collecting position information of equipment carried by individuals in a group of people, and performing fine-grained classification on image data collected on site; secondly, independently initializing the image data of multiple visual angles; thirdly, an image standardization module is constructed to correct the crowd environment image data; after the multi-view module initialization task is completed, fusing the data in the manually collected crowd information database with the multi-view module to form multi-view environment information; carrying out character feature extraction and fine granularity classification de-duplication on the multi-view environment information by adopting a Resnet50 deep neural network model, and completing the construction of the crowd target number estimation module;
The crowd target detection output module comprises the following steps:
firstly, extracting crowd information based on a crowd information database acquired manually, preprocessing crowd image information and loading a crowd abnormal state monitoring model according to the crowd information database; performing target feature extraction by using a Darknet53 deep convolutional neural network, and obtaining a final target detection result by using an image feature pyramid method; finally, adopting feature layer decoding regression classification, extracting image features on multiple scales based on an image feature pyramid by using a deep neural network, extracting high-level features of an original image by using a single neural network, and performing layer-by-layer downsampling to obtain feature layers with different scales, thereby completing the output of crowd target detection.
4. The method for detecting abnormal crowd conditions and single person movement states according to claim 3, wherein the establishing of the target recognition based on the multi-source heterogeneous data in the step 2-2 adopts an infrared fusion detection method, and the specific method is as follows:
the infrared image data is enhanced by utilizing an infrared information base acquired manually, then infrared human feature extraction is carried out, and feature stitching is carried out by adopting a PANet neural network model; and screening the confidence coefficient in the target identification result by adopting a confidence coefficient threshold filtering method, and eliminating the detection result with lower confidence coefficient.
5. The method for detecting abnormal crowd conditions and single person exercise status according to claim 4, wherein the target situation analysis based on multi-modal data in step 2-4 specifically comprises:
aiming at text information data, multi-mode data is adopted for analysis, wherein the multi-mode data comprises image information and text information; establishing a text decoding model, and extracting key information from a text information base; fusing the target characteristic information and text information extracted from the multi-view module; and detecting and outputting the fused information by using confidence threshold filtering to finish text fusion detection, namely analyzing the target situation.
6. The method for detecting abnormal conditions of crowd and single person movement state according to claim 5, wherein the single person movement state detection based on multi-perception data fusion in step 4 comprises the following steps:
step 4-1, building a physical quality assessment model according to vital sign data;
step 4-2, establishing a single equipment state evaluation model according to the equipment data;
step 4-3, establishing a crowd environment state evaluation model according to the environment data;
and 4-4, establishing a single comprehensive capacity evaluation model according to the three models, and performing single comprehensive capacity evaluation.
7. The method for detecting abnormal conditions of a crowd and movement states of a single person according to claim 6, wherein the building of the physical fitness assessment model in step 4-1 comprises:
collecting single basic physiological data and preprocessing, wherein the preprocessing comprises the steps of carrying out abnormal data processing on physiological features with missing values by adopting a mean value substitution method and carrying out L2 norm feature standardization preprocessing; training to obtain a long-short-period memory neural network model as a physical quality assessment model according to the preprocessed acquired data, and outputting a single body state result, namely a single body quality assessment result;
the mean value replacement method specifically comprises the following steps:
the first step: firstly, sorting physiological characteristic data from small to large in ascending order;
and a second step of: calculating the position: the position of the calibration parameter Q1 is (n+1) ×0.25; the position of the calibration parameter Q2 is (n+1) ×0.5; the position of the calibration parameter Q3 is (n+1) multiplied by 0.75, and n is the total number of data;
and a third step of: calculating the quantile value:
if the calculation result in the second step is an integer, directly taking the numerical value corresponding to the position;
if the result of the second step is a decimal place, the value corresponding to the 2 nd place is (1-decimal part) +the value corresponding to the 3 rd place is the decimal part;
Fourth step: calculating a quartile range IQR:
IQR＝Q3-Q1
fifth step: judging whether the physiological characteristic data is an abnormal value or not:
if the physiological characteristic data exceeds the range of [ (Q1-1.5 x IQR) to (Q3 +1.5 x IQR) ] then judging the physiological characteristic data as abnormal values;
sixth step: calculating a physiological characteristic data mean value;
seventh step: replacing the abnormal value screened in the fifth step with the average value;
the L2 norm characteristic standardization pretreatment is carried out, and is concretely as follows:
physiological characteristic data vectors x (x 1, x2, …, x n ) The L2 norm of (2) is defined as:
wherein x is n Representing nth physiological characteristic data; to normalize x to the L2 norm, i.e., build a mapping from x to x ', such that the L2 norm of x' is 1, each dimension of the vector x is divided by II x II 2 Obtaining a new vector X 2 I.e.
8. The method for detecting abnormal conditions of people and single person exercise state according to claim 7, wherein the establishing of the single person equipment state evaluation model in step 4-2 comprises the following steps:
collecting single equipment data, adopting the maximum normalized characteristic processing, and scaling different characteristic values to the same range to obtain the equipment characteristic data; performing feature extraction on the equipment feature data by using a principal component analysis method, and removing noise and redundant features by mapping high-dimensional data into a low-dimensional space to finish dimension reduction feature extraction; and training based on the processed data to obtain a convolutional neural network model as a single equipment state evaluation model, wherein the convolutional neural network model is used for evaluating the single equipment state to obtain a single equipment state evaluation result.
9. The method for detecting abnormal crowd conditions and single person exercise status according to claim 8, wherein the establishing a crowd environment status assessment model in step 4-3 comprises:
collecting environmental data in the environment where people are located, preprocessing by adopting a zero-mean value standardization method, ensuring that the mean value of the data is 0 and the standard deviation is 1, and scaling the preprocessed data to the same scale; performing feature extraction on the scaled data by adopting a linear discriminant method, projecting high-dimensional data into a low-order space, and completing maximization of the distance between different categories and minimization of the distance inside the same category, thereby completing dimension reduction and classification of the data; based on the processed data, training to obtain a circulating neural network model as a crowd environment state evaluation model, and outputting crowd environment evaluation results.
10. The method for detecting abnormal conditions of a crowd and movement states of a person according to claim 9, wherein the step 4-4 of performing a comprehensive ability evaluation of the person specifically comprises:
a single body quality evaluation result, a single equipment state evaluation result and a crowd environment evaluation result are fused through a variation coefficient method and a weighted rank sum ratio method, and a single comprehensive capacity score is obtained through calculation;
The variation coefficient method is used for determining the weight of the evaluation index, and the specific method is as follows:
the method for calculating the variation coefficient of the three indexes of psychological stress, exercise capacity and equipment state comprises the following steps:
CV l ＝S l /X l ,(l＝1,2,3)
in CV l Is the variation coefficient of the first evaluation index, S l Is the standard deviation of the first evaluation index, X l Is the average of the first evaluation index;
then calculate the weight W of each index l The method comprises the following steps:
in which W is l Is the weight of the first evaluation index, m represents the number of the evaluation indexes;
the weighted rank sum ratio method is specifically as follows:
WRSR＝CRW/m
wherein WRSR represents a weighted rank sum ratio, RSR represents a rank sum ratio, Σr represents a rank sum value of an evaluation target index, W represents a weight of each evaluation index, m represents the number of evaluation indexes, k represents, and l represents; firstly, the high-priority index is ranked from small to large, the low-priority index is ranked from large to small, and the indexes with the same numerical value are ranked evenly.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310587781.8A CN116758469A (en) | 2023-05-24 | 2023-05-24 | Crowd abnormal condition and single person movement state detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310587781.8A CN116758469A (en) | 2023-05-24 | 2023-05-24 | Crowd abnormal condition and single person movement state detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116758469A true CN116758469A (en) | 2023-09-15 |
Family
ID=87948710
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310587781.8A Pending CN116758469A (en) | 2023-05-24 | 2023-05-24 | Crowd abnormal condition and single person movement state detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116758469A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117009876A (en) * | 2023-10-07 | 2023-11-07 | 长春光华学院 | Motion state quantity evaluation method based on artificial intelligence |
| CN117828280A (en) * | 2024-03-05 | 2024-04-05 | 山东新科建工消防工程有限公司 | Intelligent fire information acquisition and management method based on Internet of things |
| CN117828280B (en) * | 2024-03-05 | 2024-06-07 | 山东新科建工消防工程有限公司 | Intelligent fire information acquisition and management method based on Internet of things |
-
2023
- 2023-05-24 CN CN202310587781.8A patent/CN116758469A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117009876A (en) * | 2023-10-07 | 2023-11-07 | 长春光华学院 | Motion state quantity evaluation method based on artificial intelligence |
| CN117009876B (en) * | 2023-10-07 | 2024-01-09 | 长春光华学院 | Motion state quantity evaluation method based on artificial intelligence |
| CN117828280A (en) * | 2024-03-05 | 2024-04-05 | 山东新科建工消防工程有限公司 | Intelligent fire information acquisition and management method based on Internet of things |
| CN117828280B (en) * | 2024-03-05 | 2024-06-07 | 山东新科建工消防工程有限公司 | Intelligent fire information acquisition and management method based on Internet of things |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107609601B (en) | Ship target identification method based on multilayer convolutional neural network | |
| CN103268495B (en) | Human body behavior modeling recognition methods based on priori knowledge cluster in computer system | |
| CN111666843B (en) | Pedestrian re-recognition method based on global feature and local feature splicing | |
| CN106408030B (en) | SAR image classification method based on middle layer semantic attribute and convolutional neural networks | |
| CN105488456B (en) | Method for detecting human face based on adaptive threshold adjustment rejection sub-space learning | |
| CN110674748A (en) | Image data processing method, image data processing device, computer equipment and readable storage medium | |
| CN114743020B (en) | Food identification method combining label semantic embedding and attention fusion | |
| CN109299657B (en) | Group behavior identification method and device based on semantic attention retention mechanism | |
| CN114283469B (en) | Improved YOLOv4-tiny target detection method and system | |
| CN110796199B (en) | Image processing method and device and electronic medical equipment | |
| CN112200021B (en) | Target crowd tracking and monitoring method based on limited range scene | |
| CN110674875A (en) | Pedestrian motion mode identification method based on deep hybrid model | |
| CN112464701A (en) | Method for detecting whether people wear masks or not based on light weight characteristic fusion SSD | |
| CN116758469A (en) | Crowd abnormal condition and single person movement state detection method | |
| CN108762503A (en) | A kind of man-machine interactive system based on multi-modal data acquisition | |
| Li et al. | A review of deep learning methods for pixel-level crack detection | |
| CN111931616A (en) | Emotion recognition method and system based on mobile intelligent terminal sensor equipment | |
| CN114022726A (en) | Personnel and vehicle monitoring method and system based on capsule network | |
| CN114723764A (en) | Parameterized edge curve extraction method for point cloud object | |
| CN116824335A (en) | YOLOv5 improved algorithm-based fire disaster early warning method and system | |
| CN117217368A (en) | Training method, device, equipment, medium and program product of prediction model | |
| CN114067128A (en) | SLAM loop detection method based on semantic features | |
| CN114358279A (en) | Image recognition network model pruning method, device, equipment and storage medium | |
| CN113762166A (en) | Small target detection improvement method and system based on wearable equipment | |
| CN117219127A (en) | Cognitive state recognition method and related equipment |
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 |