CN111259855A - Mobile safety helmet wearing detection method based on deep learning - Google Patents
Mobile safety helmet wearing detection method based on deep learning Download PDFInfo
- Publication number
- CN111259855A CN111259855A CN202010083323.7A CN202010083323A CN111259855A CN 111259855 A CN111259855 A CN 111259855A CN 202010083323 A CN202010083323 A CN 202010083323A CN 111259855 A CN111259855 A CN 111259855A
- Authority
- CN
- China
- Prior art keywords
- safety
- safety helmet
- training
- wearing
- people
- 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
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V20/40—Scenes; Scene-specific elements in video content
-
- 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
-
- 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
Abstract
The invention discloses a mobile safety helmet wearing detection method based on deep learning. The belt conveyor inspection robot runs on the track, the video images transmitted back by the cloud deck camera are analyzed, the SDD model is used for carrying out target identification, whether personnel change exists in the video is judged according to the number of detected human bodies in the image, the number of people without safety helmets and the number of people with safety helmets, the number of the human bodies of the current image and the number of the detected images of the previous three frames are used, when the personnel change, the detected object relation is used for judging whether people exist in the inspection robot along the way and whether the personnel wear safety helmets, and meanwhile, one frame of video image without safety helmets is stored, so that only one frame of illegal image is stored when the phenomenon that the safety helmets are not worn occurs on the site. According to the invention, a neural network model with safety helmet detection is obtained on a data set by adopting a deep learning algorithm, and the model achieves a good detection effect on a constructed test set.
Description
Technical Field
The invention relates to the field of artificial intelligence target detection, in particular to a mobile safety helmet wearing detection method based on deep learning.
Background
In actual scenes such as coal mines, power plants, transformer substations and construction sites, the safety helmet can effectively protect the head safety of operators, and is a safety measure which must be executed. However, as some people lack safety consciousness, the safety helmet is not worn according to regulations, and huge safety risks are brought. At present, the main management method is to carry out construction site video monitoring and adopt manual supervision to determine whether a safety helmet is worn. However, in the manual supervision method, on one hand, because the field is numerous, the monitoring screen is large, monitoring personnel are easy to fatigue, and monitoring is careless; on the other hand, a large amount of manpower is needed, which causes resource waste.
In recent years, artificial intelligence is rapidly developed, and computer vision is an important research direction of artificial intelligence, and the third hot tide is also met. As a research hotspot in the field of computer vision, a great number of excellent target detection algorithms based on convolutional neural networks have been successful, and more learners are encouraged to begin to focus on the research of deep learning target detection algorithms. As the safety helmet is taken as a safety protection article, the safety helmet mainly plays a role in protecting the head of a worker on a construction site, preventing falling of a high-altitude object and preventing hitting and collision of the object, plays an important role in safety production, and is valued for safety helmet detection of video images.
At present, the main management method is to carry out construction site video monitoring and adopt manual supervision to determine whether a safety helmet is worn. Even there is video detection, also be fixed camera, be used to building site, power plant entry more, it is not enough to on-the-spot supervision.
On one hand, due to the fact that the field is numerous, the monitoring screen is large, monitoring personnel are prone to fatigue, and monitoring omission is caused; on the other hand, a large amount of manpower is needed, which causes resource waste; fixed point detection can only monitor a local area due to limited visual angles, and can not exclude the operation standard condition of other personnel in uncovered areas.
Disclosure of Invention
The invention aims to solve the technical problem of overcoming the defects in the prior art and provides a method for detecting wearing of a movable safety helmet based on deep learning.
Coal, power plant's belt feeder inspection robot moves on the track, video image that passes through analysis cloud platform camera and transmit back, utilize the SDD model to carry out target identification, through the human number that detects in the picture, do not wear safety helmet people's head number, utilize the human number of current and three frames of the image that detect before, judge whether there is personnel's change in the video, when there is personnel's change, then utilize the object relation that detects, judge whether have people and whether personnel wear safety helmet in the inspection robot is along the way, simultaneously, save a frame video image of not wearing safety helmet, reduce data redundancy, prevent the phenomenon of same violation of rules and regulations, many times of storage. Therefore, when the phenomenon that the safety helmet is not worn on the spot occurs, only one frame of violation image is saved.
The method adopts a Tensorflow deep learning framework, which is a second-generation open-source deep learning platform of Google and is also the most popular machine learning framework for realizing the neural network at present.
Tensorflow widely supports a large number of functions including computer vision, speech recognition, human-computer gaming, and natural language processing. Therefore, the method provides a mobile safety helmet wearing detection method based on deep learning on the basis of the previous research. The process comprises sample collection, model training and detection. And the inspection robot holder camera realizes the detection and identification of the surrounding environment, and verifies the feasibility and effectiveness of the algorithm on the wearing detection of the safety helmet of the field worker. And alarm and illegal image storage are carried out on the phenomenon of wearing safety which is not in accordance with the standard.
The invention relates to a mobile safety helmet wearing detection method based on deep learning, wherein a coal and power plant belt conveyor inspection robot runs on a track, and a video image transmitted back by a cloud deck camera is analyzed, and the specific detection method comprises the following steps:
1. collecting samples: the method is to obtain pictures of people in a monitored site, wherein the pictures need to comprise a human body with a safety helmet (construction site) and a human body without safety (other places can be used);
2. model training: extracting a training target coordinate through labelImg software by using the obtained sample, obtaining data required by training through a series of conversions, and sending the data into a target recognition model for training;
3. a detection section:
a. collecting a frame of video image;
b. starting image target detection;
c. counting and labeling the detection results, wherein the counting comprises the number of people wearing safety helmets, the number of people not wearing safety helmets and the number of human bodies;
d. judging whether the judgment is cancelled or not, if not, starting to judge whether a person does not wear the safety helmet or not, if so, not performing any treatment, and closing the detection mark;
e. judging whether to start a detection mark according to the number of people in the current frame and the number of people in the previous frame;
f. and updating and storing the number of the human bodies detected in the three adjacent types, and finally displaying the image with the identification result.
The method comprises the steps of collecting samples, and identifying workers wearing safety helmets and workers not wearing safety helmets in construction site images based on a Tensorflow frame; first, a certain number of worker images are collected as a training and testing set of models, and currently, since there is no complete data set about the detection of personal safety devices of workers, a data set needs to be developed and designed by oneself to train and test the ssd _ mobilenet _ v2 model. The research needs to detect the head image of a person without a safety helmet, the head image of a person with a safety helmet and the human body image. To save labor, a pre-trained model of ssd _ mobilene _ v2_ coco under the Tensorflow framework was used for human acquisition, 2000 images of the person wearing the helmet were acquired, and 2000 images of the person without the helmet were collected (possibly elsewhere).
The model training, downloading labelImg software, labeling the collected samples, which includes: the data can be automatically stored in an xml format after the label is marked on the head of the person wearing the safety helmet, the head of the person not wearing the safety helmet and the human body; then, the xml data set is divided into a training set accounting for 90% and a test set accounting for 10%. Then, converting the xml into a csv file, and converting the csv file into a tfrecrd format required by training; creating a label classified configuration file (label _ map.pbtxt) in an engineering file directory, wherein several targets need to be detected, and several ids are created; configuring a pipeline configuration file, and finally starting training according to the configured file.
Compared with the prior art, the invention has the beneficial effects that:
the invention constructs a training set and a testing set by collecting the scene picture information and a manual labeling method. By adopting a deep learning algorithm, a neural network model with safety helmet detection is obtained on a data set. Through verification, the model achieves a good detection effect on the constructed test set.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and specific examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Coal, power plant's belt feeder patrol and examine robot and move on the track, and the video image that transmits back through analysis cloud platform camera, detection method are as shown in figure 1, after the start-up system, at first carry out the relevant parameter setting of procedure, include: setting a network camera; detecting whether a stored folder without a safety helmet image exists or not, and if not, reestablishing the folder; the loading of the training model is detected, and the model loading data volume is large, so that long waiting time is needed.
After the initialization is completed, the system performs a frame of video image acquisition, and since the resolution of the original image is generally over 200 ten thousand, in order to reduce the operation pressure caused by detection, the image is processed into 960 × 540.
After a frame of video is read, the operation of an image recognition algorithm is carried out, and the detection results are divided into three types. Then, counting and labeling the detection results, and counting the number of the people wearing the safety helmet, the number of the people without the safety helmet and the number of the human bodies.
And judging whether the judgment is cancelled or not according to the statistical result, wherein the judgment standard is to detect whether the number of human bodies or the number of the human bodies is equal to the number of the worn safety helmets or not. If the number of the human bodies is zero or is equal to the number of the people wearing the safety helmet, the phenomenon that the safety helmet is not worn is avoided, and the detection mark is closed. Whether the person starts detection and judges whether a person wears no safety helmet or not, and the judgment standard is as follows: the human body without the safety helmet is larger than zero, the detection mark is opened, if the condition is met, the phenomenon that the safety helmet is not worn is judged, an alarm is output, the image of the safety helmet is stored, and finally the detection mark is closed.
Judging whether the number of detected people changes or not, and starting a detection mark when the number of the detected people meets the following conditions according to the number of three times of human bodies currently detected and stored by the judged mark: the number of the human bodies in the current image is the same as that in the third frame; 2. the number of human bodies in the current frame is different from the number of human bodies in the first frame and the second frame; 3. the number of human bodies is not zero at present; 4. the number of the human bodies in the current frame is larger than the number of the detected people wearing the safety helmet.
And finally, updating and storing the number of the human bodies detected in the three adjacent types, and displaying the identification result on the image.
Claims (3)
1. A mobile safety helmet wearing detection method based on deep learning is characterized in that a belt conveyor inspection robot runs on a track, and a video image transmitted back by a cloud deck camera is analyzed, and the detection method is carried out according to the following steps:
1) collecting samples: the method comprises the steps of obtaining pictures of people in a monitored site, wherein the pictures need to comprise human body pictures with safety caps for a construction site, and the human body pictures without the safety caps are used in other places; the sample collection is to identify workers wearing safety helmets and not wearing safety helmets in construction site images based on a Tensorflow frame, and collect a certain number of worker images as a training and testing set of models;
2) model training: extracting a training target coordinate through labelImg software by using the obtained sample, obtaining data required by training through a series of conversions, and sending the data into a target recognition model for training;
3) a detection section:
a. collecting a frame of video image;
b. starting image target detection;
c. counting and labeling the detection results, wherein the counting comprises the number of people wearing safety helmets, the number of people not wearing safety helmets and the number of human bodies;
d. judging whether the judgment is cancelled or not, if not, starting to judge whether a person does not wear the safety helmet or not, if so, not performing any treatment, and closing the detection mark;
e. judging whether to start a detection mark according to the number of people in the current frame and the number of people in the previous frame;
f. and updating and storing the number of the human bodies detected in the three adjacent types, and finally displaying the image with the identification result.
2. The method of claim, wherein the sample collection, designing a data set to train and test the ssd _ mobilenet _ v2 model, collecting 2000 images of the person wearing the safety helmet from the field, and collecting 2000 images of the person without the safety helmet.
3. The method of claim, wherein the model training, downloading labelImg software, and labeling the collected samples comprises: the data can be automatically stored in an xml format after the label is marked on the head of the person wearing the safety helmet, the head of the person not wearing the safety helmet and the human body; then, the xml data set is divided into a training set accounting for 90 percent, and test is 10 percent of the test set; then, converting the xml into a csv file, and converting the csv file into a tfrecrd format required by training; creating a label-classified configuration file label _ map.pbtxt under an engineering file directory, detecting several targets, and creating several ids; configuring a pipeline configuration file, and finally starting training according to the configured file.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010083323.7A CN111259855A (en) | 2020-02-09 | 2020-02-09 | Mobile safety helmet wearing detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010083323.7A CN111259855A (en) | 2020-02-09 | 2020-02-09 | Mobile safety helmet wearing detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111259855A true CN111259855A (en) | 2020-06-09 |
Family
ID=70948235
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010083323.7A Pending CN111259855A (en) | 2020-02-09 | 2020-02-09 | Mobile safety helmet wearing detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111259855A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860242A (en) * | 2020-07-07 | 2020-10-30 | 北京海益同展信息科技有限公司 | Robot inspection method and device and computer readable medium |
| CN111914773A (en) * | 2020-08-07 | 2020-11-10 | 杭州微胜智能科技有限公司 | Equipment and method for capturing illegal boarding and alighting of passengers |
| CN112911255A (en) * | 2020-11-10 | 2021-06-04 | 泰州程顺制冷设备有限公司 | Building site safety state detecting system |
| CN113553963A (en) * | 2021-07-27 | 2021-10-26 | 广联达科技股份有限公司 | Detection method and device of safety helmet, electronic equipment and readable storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049787A (en) * | 2011-10-11 | 2013-04-17 | 汉王科技股份有限公司 | People counting method and system based on head and shoulder features |
| CN104902227A (en) * | 2015-05-06 | 2015-09-09 | 南京第五十五所技术开发有限公司 | Substation helmet wearing condition video monitoring system |
| CN105469069A (en) * | 2015-12-09 | 2016-04-06 | 上海工业自动化仪表研究院 | Safety helmet video detection method for production line data acquisition terminal |
| CN106372662A (en) * | 2016-08-30 | 2017-02-01 | 腾讯科技（深圳）有限公司 | Helmet wearing detection method and device, camera, and server |
| CN108596111A (en) * | 2018-04-26 | 2018-09-28 | 四川隧唐科技股份有限公司 | Safety cap recognition methods and device |
| CN110008804A (en) * | 2018-12-12 | 2019-07-12 | 浙江新再灵科技股份有限公司 | Elevator monitoring key frame based on deep learning obtains and detection method |
| CN110602449A (en) * | 2019-09-01 | 2019-12-20 | 天津大学 | Intelligent construction safety monitoring system method in large scene based on vision |
-
2020
- 2020-02-09 CN CN202010083323.7A patent/CN111259855A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049787A (en) * | 2011-10-11 | 2013-04-17 | 汉王科技股份有限公司 | People counting method and system based on head and shoulder features |
| CN104902227A (en) * | 2015-05-06 | 2015-09-09 | 南京第五十五所技术开发有限公司 | Substation helmet wearing condition video monitoring system |
| CN105469069A (en) * | 2015-12-09 | 2016-04-06 | 上海工业自动化仪表研究院 | Safety helmet video detection method for production line data acquisition terminal |
| CN106372662A (en) * | 2016-08-30 | 2017-02-01 | 腾讯科技（深圳）有限公司 | Helmet wearing detection method and device, camera, and server |
| CN108596111A (en) * | 2018-04-26 | 2018-09-28 | 四川隧唐科技股份有限公司 | Safety cap recognition methods and device |
| CN110008804A (en) * | 2018-12-12 | 2019-07-12 | 浙江新再灵科技股份有限公司 | Elevator monitoring key frame based on deep learning obtains and detection method |
| CN110602449A (en) * | 2019-09-01 | 2019-12-20 | 天津大学 | Intelligent construction safety monitoring system method in large scene based on vision |
Non-Patent Citations (1)
| Title |
|---|
| 樊钰: "基于深度学习的安全帽检测系统设计与实现", 《中国优秀硕士学位论文全文数据库工程科技I辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860242A (en) * | 2020-07-07 | 2020-10-30 | 北京海益同展信息科技有限公司 | Robot inspection method and device and computer readable medium |
| CN111914773A (en) * | 2020-08-07 | 2020-11-10 | 杭州微胜智能科技有限公司 | Equipment and method for capturing illegal boarding and alighting of passengers |
| CN112911255A (en) * | 2020-11-10 | 2021-06-04 | 泰州程顺制冷设备有限公司 | Building site safety state detecting system |
| CN112911255B (en) * | 2020-11-10 | 2021-12-07 | 江苏国兴建设项目管理有限公司 | Building site safety state detecting system |
| CN113553963A (en) * | 2021-07-27 | 2021-10-26 | 广联达科技股份有限公司 | Detection method and device of safety helmet, electronic equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111259855A (en) | Mobile safety helmet wearing detection method based on deep learning | |
| CN110110613B (en) | Track traffic abnormal personnel detection method based on motion recognition | |
| CN108319926A (en) | A kind of the safety cap wearing detecting system and detection method of building-site | |
| CN111967400B (en) | Campus safety monitoring and early warning management system based on artificial intelligence | |
| CN111191586B (en) | Method and system for inspecting wearing condition of safety helmet of personnel in construction site | |
| CN111598040A (en) | Construction worker identity identification and safety helmet wearing detection method and system | |
| CN112183317B (en) | Live working field violation behavior detection method based on space-time diagram convolutional neural network | |
| CN111920129A (en) | Intelligent safety helmet system | |
| CN108566538A (en) | Based on the circular coal yard personnel safety guard of Infrared-Visible fusion tracking and the monitoring system and method for spontaneous combustion | |
| CN114665608B (en) | Intelligent sensing inspection system and method for transformer substation | |
| CN113206978A (en) | Security intelligent monitoring early warning system and method for oil and gas pipeline station | |
| CN111257507A (en) | Gas concentration detection and accident early warning system based on unmanned aerial vehicle | |
| CN114565845A (en) | Intelligent inspection system for underground tunnel | |
| CN115393566A (en) | Fault identification and early warning method and device for power equipment, storage medium and equipment | |
| CN115223249A (en) | Quick analysis and identification method for unsafe behaviors of underground personnel based on machine vision | |
| CN116071708A (en) | Image recognition analysis method for fire risk security protection | |
| CN104574729A (en) | Alarming method, device and system | |
| CN114187542A (en) | Insulating glove detection method and system in electric power scene | |
| CN113807240A (en) | Intelligent transformer substation personnel dressing monitoring method based on uncooperative face recognition | |
| CN111652128B (en) | High-altitude power operation safety monitoring method, system and storage device | |
| CN115829324A (en) | Personnel safety risk silent monitoring method | |
| CN110751125A (en) | Wearing detection method and device | |
| CN116433029A (en) | Power operation risk assessment method, system, equipment and storage medium | |
| CN113822242B (en) | Image recognition technology-based helmet wearing recognition method and device | |
| CN112926488B (en) | Electric power pole tower structure information-based operator violation identification method |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200609 |
|
| WD01 | Invention patent application deemed withdrawn after publication |