CN112528885A - Identification method for platform staff in intelligent zoning - Google Patents
Identification method for platform staff in intelligent zoning Download PDFInfo
- Publication number
- CN112528885A CN112528885A CN202011485426.2A CN202011485426A CN112528885A CN 112528885 A CN112528885 A CN 112528885A CN 202011485426 A CN202011485426 A CN 202011485426A CN 112528885 A CN112528885 A CN 112528885A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- mask
- detection
- algorithm
- picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000013316 zoning Methods 0.000 title claims abstract description 10
- 238000001514 detection method Methods 0.000 claims abstract description 30
- 238000013135 deep learning Methods 0.000 claims abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 6
- 238000013145 classification model Methods 0.000 claims abstract description 4
- 230000004927 fusion Effects 0.000 claims description 6
- 239000003086 colorant Substances 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 5
- 230000007613 environmental effect Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a platform worker identification method for intelligent zoning, which comprises the following steps: collecting pedestrian data; training a pedestrian classification model by using marked data and a convolutional neural network; a color filter; performing significance detection on the picture after passing through the color filter, and automatically and accurately extracting a warning line region mask I by using SOD 100K; drawing a closed polygonal warning area according to the warning line area mask I to obtain a mask II; taking the image and operation graph of the original image and the mask II to obtain a target area picture; carrying out pedestrian detection on the obtained target area picture by using a deep learning-based method yolov4, and if a pedestrian exists, obtaining a corresponding pedestrian detection frame bbox; extracting a corresponding pedestrian picture for the obtained pedestrian frame bbox; and classifying and identifying the pedestrian pictures by using a pre-trained pedestrian classification network model to obtain the identification result which is the non-worker or the type of the worker.
Description
Technical Field
The invention relates to an identification method for platform workers in an intelligent zoning.
Background
Non-working personnel in certain areas on a platform of a train or a high-speed rail station are forbidden to enter, and if ordinary passengers enter the areas, certain danger may exist, so that the intelligent zoning and the identification of the working personnel are very important for early warning and warning of the non-working personnel. Because the station environment is complicated, the installation angle of a monitoring camera is usually not fixed, and how to intelligently mark out a designated area without being influenced by the angle of the camera according to an actual scene and identify the types of workers in the area is challenging.
With the development and progress of computer vision technology and machine learning, the deep learning method is used for extracting and identifying the depth features of the object, and a robust result with a good effect can be realized on the detection and identification of the object.
The defects of the prior art are mainly as follows:
1. the detection of the region based on the traditional machine learning method is easily influenced by other environmental factors, such as illumination and objects with similar colors, and generates more noise;
2. the traditional feature extraction method is based on manually designed features, and the features are easily influenced by factors such as illumination, brightness and the like, so that the detection and classification identification results of personnel are inaccurate.
Disclosure of Invention
The method aims to overcome the defects of the prior art, the regional lines are roughly extracted from the platform warning lines based on colors, and then the warning lines are finely extracted based on a deep learning method so as to delineate the warning regions. By using the pedestrian detection and classification identification method of deep learning to detect and identify pedestrians in the platform area, the used object depth characteristics have robustness on environmental influence.
The specific technical scheme of the invention is as follows:
disclosed is a method for identifying platform staff in an intelligent zoning. Comprises the following steps:
step one, acquiring pedestrian data, and manually marking and classifying;
secondly, training a pedestrian classification model by using the manually marked data in the first step and a convolutional neural network;
step three, color filter: firstly, converting a picture into HSV space by BGR, and then respectively determining H, S, V value ranges according to the colors of warning lines, wherein the value ranges are as follows: 11< H <26, S >43 or S <255, V >46 or V <255, zeroing values that are not simultaneously within the value range of H, S, V;
and fourthly, performing significance Detection on the picture after passing through the color filter, wherein an algorithm used for automatically and accurately extracting the mask I of the warning line region is SOD100K (high effective Efficient Object Detection with 100K Parameters), and a lightweight network provided by the algorithm mainly comprises a feature extractor and a cross-stage fusion part and can simultaneously process features of multiple scales. The feature extractor is stacked with the intra-layer multi-scale blocks proposed by SOD100K and is divided into 4 stages, each stage having 3, 4, 6, and 4 intra-layer multi-scale blocks, respectively, according to the resolution of the feature map. The cross-stage fusion part, which is a flexible convolution module (goctcnvs) proposed by SOD100K, processes features from stages of the feature extractor to obtain high-resolution output.
The algorithm uses a novel dynamic weight attenuation scheme to reduce redundancy of feature representations, and the weight attenuation can be adjusted according to specific features of certain channels. In particular, during back propagation, the attenuation term may dynamically change according to the characteristics of certain channels. The weight update for dynamic weight decay can be represented as:
wherein λdIs a weight, x, of a dynamic weight decayiIs represented by wiCalculated features, and S (x)i) Is a measure of the feature, which may have multiple definitions, w, depending on the taskiIs the weight of the i-th layer,is the gradient to be updated. In this algorithm, the goal is to assign weights according to features between stable channels, using global average pooling as a special passThe index of the trace, the formula, can be expressed as:
xithe characteristic diagram is shown, and H and W respectively represent the height and width of the characteristic diagram.
Step five, a closed polygonal warning area is drawn according to the warning line area mask I, the pixel value in the closed area is set to be 255, the pixel values outside the area are all set to be 0, and a mask II is obtained;
sixthly, taking the image and the operation image of the original image and the mask II in the fifth step to obtain a target area image;
seventhly, performing pedestrian Detection on the target area picture obtained in the sixth step by using a deep learning-based method yolov4(Optimal Speed and Accuracy of Object Detection), if a pedestrian exists, obtaining a corresponding pedestrian Detection frame bbox, wherein the yolov4 algorithm is a universal target Detection algorithm, and is mainly characterized in that the target Detection precision is high, the Speed is high, and a loss function mainly used by the algorithm is CIoU and can be expressed as:
wherein the content of the first and second substances,
b,bgtrespectively representing the center points of the anchor frame and the target frame, p (-) represents the Euclidean distance, and c represents the distance between the anchor frame and the target frameThe diagonal distance of the smallest rectangle covering both the anchor and the target box, ω, h represents the width and height of the predicted box, ω, respectivelygt,hgtRespectively the width and height of the real box.
Step eight, extracting corresponding pedestrian pictures of the pedestrian frames bbox obtained in the step seven;
and step nine, classifying and identifying the pedestrian pictures by using a pre-trained pedestrian classification network model to obtain the identification result which is the non-worker or the type of the worker.
Technical effects
The traditional machine learning method is used for zoning, and because the angle for installing the monitoring camera is not fixed, the monitoring camera is easily influenced by environmental factors and noise, so that the region extraction is inaccurate. According to the method, firstly, a color filtering method is adopted, most of other objects irrelevant to a target area are removed preliminarily, and then the target area is extracted finely by using the saliency target detection SOD 100K.
The pedestrian detection method based on the yolov4 is used for detecting pedestrians by using the pictures in the target area, the detected pedestrian pictures are classified by using the convolutional neural network, the convolutional neural network uses the depth characteristics of objects, is less influenced by environmental factors, can accurately identify the types of the pedestrians, and can give out early warning if the stay time of non-workers in the specified area reaches the set threshold.
Drawings
Fig. 1 is a schematic diagram illustrating an identification method of platform staff for intelligent zoning according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
A platform staff identification method of intelligent zoning comprises the following overall steps.
Step one, acquiring pedestrian data, and manually marking and classifying;
secondly, training a pedestrian classification model by using the manually marked data in the first step and a convolutional neural network;
step three, color filter: firstly, converting a picture into HSV space by BGR, and then respectively determining H, S, V value ranges according to the colors of warning lines, wherein the value ranges are as follows: 11< H <26, S >43 or S <255, V >46 or V <255, zeroing values that are not simultaneously within the value range of H, S, V;
and fourthly, performing significance Detection on the picture after passing through the color filter, wherein an algorithm used for automatically and accurately extracting the mask I of the warning line region is SOD100K (high effective Efficient Object Detection with 100K Parameters), and a lightweight network provided by the algorithm mainly comprises a feature extractor and a cross-stage fusion part and can simultaneously process features of multiple scales. The feature extractor is stacked with the intra-layer multi-scale blocks proposed by SOD100K and is divided into 4 stages, each stage having 3, 4, 6, and 4 intra-layer multi-scale blocks, respectively, according to the resolution of the feature map. The cross-stage fusion part, which is a flexible convolution module (goctcnvs) proposed by SOD100K, processes features from stages of the feature extractor to obtain high-resolution output.
The algorithm uses a novel dynamic weight attenuation scheme to reduce redundancy of feature representations, and the weight attenuation can be adjusted according to specific features of certain channels. In particular, during back propagation, the attenuation term may dynamically change according to the characteristics of certain channels. The weight update for dynamic weight decay can be represented as:
wherein λdIs a weight, x, of a dynamic weight decayiIs represented by wiCalculated features, and S (x)i) Is a measure of the feature, which may have multiple definitions, w, depending on the taskiIs the weight of the i-th layer,is the gradient to be updated. At the position ofIn the algorithm, the objective is to perform weight distribution according to the features between stable channels, and the global average pooling is used as an index of a specific channel, and the formula can be expressed as:
xithe characteristic diagram is shown, and H and W respectively represent the height and width of the characteristic diagram.
Step five, a closed polygonal warning area is drawn according to the warning line area mask I, the pixel value in the closed area is set to be 255, the pixel values outside the area are all set to be 0, and a mask II is obtained;
sixthly, taking the image and the operation image of the original image and the mask II in the fifth step to obtain a target area image;
seventhly, performing pedestrian Detection on the target area picture obtained in the sixth step by using a deep learning-based method yolov4(Optimal Speed and Accuracy of Object Detection), if a pedestrian exists, obtaining a corresponding pedestrian Detection frame bbox, wherein the yolov4 algorithm is a universal target Detection algorithm, and is mainly characterized in that the target Detection precision is high, the Speed is high, and a loss function mainly used by the algorithm is CIoU and can be expressed as:
wherein the content of the first and second substances,
b,bgtrepresents the center points of the anchor frame and the target frame, respectively, (. rho.) represents the Euclidean distance, c represents the diagonal distance of the smallest rectangle that can cover both the anchor and the target frame, ω, h represents the width and height of the prediction frame, ω, respectivelygt,hgtRespectively the width and height of the real box.
Step eight, extracting corresponding pedestrian pictures of the pedestrian frames bbox obtained in the step seven;
and step nine, classifying and identifying the pedestrian pictures by using a pre-trained pedestrian classification network model to obtain the identification result which is the non-worker or the type of the worker.
Claims (1)
1. A platform staff identification method of intelligent zoning is characterized by comprising the following steps:
step one, acquiring pedestrian data, and manually marking and classifying;
secondly, training a pedestrian classification model by using the manually marked data in the first step and a convolutional neural network;
step three, color filter: firstly, converting a picture into HSV space by BGR, and then respectively determining H, S, V value ranges according to the colors of warning lines, wherein the value ranges are as follows: 11< H <26, S >43 or S <255, V >46 or V <255, zeroing values that are not simultaneously within the value range of H, S, V;
fourthly, performing significance Detection on the picture passing through the color filter, automatically and accurately extracting a warning line area mask I, wherein the used algorithm is SOD100K (high effective Efficient Object Detection with 100K Parameters), and the lightweight network provided by the algorithm mainly comprises a feature extractor and a cross-stage fusion part and can simultaneously process features of multiple scales; the feature extractor is stacked with the intra-layer multi-scale blocks proposed by SOD100K and divided into 4 stages according to the resolution of the feature map, each stage having 3, 4, 6 and 4 intra-layer multi-scale blocks, respectively; the cross-stage fusion part of a flexible convolution module (goctcnvs) proposed by SOD100K processes features from stages of the feature extractor to obtain high resolution output;
the algorithm uses a novel dynamic weight attenuation scheme to reduce the redundancy of feature representation, and the weight attenuation can be adjusted according to the specific features of certain channels; specifically, during back propagation, the attenuation term will dynamically change according to the characteristics of certain channels; the weight update for dynamic weight decay can be represented as:
wherein λdIs a weight, x, of a dynamic weight decayiIs represented by wiCalculated features, and S (x)i) Is a measure of the feature, which may have multiple definitions, w, depending on the taskiIs the weight of the i-th layer,is the gradient to be updated; in this algorithm, the goal is to assign weights according to features between stable channels, using global average pooling as an indicator for a particular channel, the formula can be expressed as:
xirepresenting a feature map, H, W representing the height and width of the feature map, respectively;
step five, a closed polygonal warning area is drawn according to the warning line area mask I, the pixel value in the closed area is set to be 255, the pixel values outside the area are all set to be 0, and a mask II is obtained;
sixthly, taking the image and the operation image of the original image and the mask II in the fifth step to obtain a target area image;
seventhly, performing pedestrian Detection on the target area picture obtained in the sixth step by using a deep learning-based method yolov4(Optimal Speed and Accuracy of Object Detection), if a pedestrian exists, obtaining a corresponding pedestrian Detection frame bbox, wherein the yolov4 algorithm is a universal target Detection algorithm, and is mainly characterized in that the target Detection precision is high, the Speed is high, and a loss function mainly used by the algorithm is CIoU and can be expressed as:
wherein the content of the first and second substances,
b,bgtrepresents the center points of the anchor frame and the target frame, respectively, (. rho.) represents the Euclidean distance, c represents the diagonal distance of the smallest rectangle that can cover both the anchor and the target frame, ω, h represents the width and height of the prediction frame, ω, respectivelygt,hgtWidth and height of the real frame respectively;
step eight, extracting corresponding pedestrian pictures of the pedestrian frames bbox obtained in the step seven;
and step nine, classifying and identifying the pedestrian pictures by using a pre-trained pedestrian classification network model to obtain the identification result which is the non-worker or the type of the worker.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011485426.2A CN112528885A (en) | 2020-12-16 | 2020-12-16 | Identification method for platform staff in intelligent zoning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011485426.2A CN112528885A (en) | 2020-12-16 | 2020-12-16 | Identification method for platform staff in intelligent zoning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112528885A true CN112528885A (en) | 2021-03-19 |
Family
ID=75000599
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011485426.2A Pending CN112528885A (en) | 2020-12-16 | 2020-12-16 | Identification method for platform staff in intelligent zoning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112528885A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114067360A (en) * | 2021-11-16 | 2022-02-18 | 国网上海市电力公司 | Pedestrian attribute detection method and device |
-
2020
- 2020-12-16 CN CN202011485426.2A patent/CN112528885A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114067360A (en) * | 2021-11-16 | 2022-02-18 | 国网上海市电力公司 | Pedestrian attribute detection method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108759973B (en) | Water level measuring method | |
CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
CN109785337B (en) | In-column mammal counting method based on example segmentation algorithm | |
CN109101924B (en) | Machine learning-based road traffic sign identification method | |
CN104112269B (en) | A kind of solar battery laser groove parameter detection method and system based on machine vision | |
CN110866430B (en) | License plate recognition method and device | |
CN102867417B (en) | Taxi anti-forgery system and taxi anti-forgery method | |
CN109902676B (en) | Dynamic background-based violation detection algorithm | |
CN105844621A (en) | Method for detecting quality of printed matter | |
CN112149543B (en) | Building dust recognition system and method based on computer vision | |
CN113011283B (en) | Non-contact type rail sleeper relative displacement real-time measurement method based on video | |
CN111753797A (en) | Vehicle speed measuring method based on video analysis | |
CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
CN113436157A (en) | Vehicle-mounted image identification method for pantograph fault | |
CN113435407B (en) | Small target identification method and device for power transmission system | |
CN106056078B (en) | Crowd density estimation method based on multi-feature regression type ensemble learning | |
CN113378659A (en) | Method for identifying busy operation | |
KR101874968B1 (en) | Visibility measuring system base on image information and method for using the same | |
CN113435452A (en) | Electrical equipment nameplate text detection method based on improved CTPN algorithm | |
CN110348307B (en) | Path edge identification method and system for crane metal structure climbing robot | |
CN110516524A (en) | Vehicle number of axle recognition methods based on Mask R-CNN in a kind of traffic scene | |
CN112528885A (en) | Identification method for platform staff in intelligent zoning | |
CN115239621A (en) | Transformer substation visual intelligent inspection method based on virtual-real matching | |
KR102178202B1 (en) | Method and apparatus for detecting traffic light | |
CN113963230A (en) | Parking space detection method based on deep learning |
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 |