CN116030405A - Pedestrian target safety protection tool wearing monitoring method based on artificial intelligence - Google Patents
Pedestrian target safety protection tool wearing monitoring method based on artificial intelligence Download PDFInfo
- Publication number
- CN116030405A CN116030405A CN202111243169.6A CN202111243169A CN116030405A CN 116030405 A CN116030405 A CN 116030405A CN 202111243169 A CN202111243169 A CN 202111243169A CN 116030405 A CN116030405 A CN 116030405A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- safety
- safety protection
- algorithm
- 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
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a pedestrian target safety protection tool wearing monitoring method based on artificial intelligence, and belongs to the technical field of artificial intelligence. Including image recognition, which is a technique of identifying and locating objects in an image to identify targets and objects in various different modes. The identification process comprises the steps of image acquisition, image preprocessing, feature selection, extraction, classification decision and the like. And (3) completing image recognition target library construction by collecting and sorting defect pictures in service development as sample data, and completing image target detection algorithm library construction based on a neural network algorithm. Based on a strong visual library and an image algorithm, judging the wearing condition of the safety protection tool of the pedestrian target, establishing a safety protection tool sample database and an algorithm model, and intelligently detecting the detected pedestrian target. The invention solves the problems of diversification of safety protection tools in image recognition, interference of visual angle change and low recognition accuracy in multi-point start-up.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a pedestrian target safety protection wear monitoring method based on artificial intelligence.
Background
With the continuous construction of the power grid engineering, the problems of tight construction period of the power grid infrastructure engineering progress, single safety quality management means, shortage of field management staff and the like emerge, power grid engineering practitioners are continuously increased, engineering sub-packaging is carried out normally, the number of sub-packaging staff and mobility are continuously increased, and management difficulty is increased.
In order to strengthen the field management and control of the engineering, promote the field management capability of the engineering, promote the field safety management responsibility to be implemented, ensure the whole process, the omnibearing controllability, the energy control and the in-control of the construction of the basic construction engineering of the power supply company, coordinate and solve various problems in the construction process in time, explore the intelligent and informationized implementation of the comprehensive supervision mode of the engineering construction of the power supply company, and promote the field management level of the basic construction.
Chinese patent document CN201910300432.7 discloses a method and apparatus for identifying the wearing of a safety brace based on deep learning, by inputting an image to be identified into a pre-constructed safety brace identification model; the safety protection tool recognition model is obtained based on a pre-marked image sample and multi-layer convolutional neural network training; and obtaining a sub-image containing the safety protector in the image to be identified based on the output result of the safety protector identification model. The technical problems that in the prior art, the wearing recognition scheme of the safety protection tool is greatly influenced by illumination, and different feature extraction methods also have influence on the robustness of an algorithm are solved, and the technical effect of rapidly and accurately carrying out the wearing recognition of the safety protection tool is achieved. But it cannot solve the problem of how to overcome the background diversity and the disturbance of the viewing angle variation in the image recognition.
Chinese patent document CN202010566737.5 discloses a safety protection wear detection system based on 5G and deep learning target detection, and belongs to the technical field of wireless data transmission and image processing. The system comprises an image and data acquisition node, a monitoring center and a mobile phone terminal; the monitoring center comprises a computer and a gateway, and the gateway is in communication connection with the image and data acquisition node through a 5G network; compared with the existing manual inspection of the safety officer, the detection system provided by the invention greatly improves the detection instantaneity, reduces the safety detection labor intensity, reduces the safety risk of the staff, and has a certain benefit for the safety production guarantee of the factory. But the problem of safety control under the condition of all-round guarantee of multi-point startup and large-scale construction cannot be solved.
Disclosure of Invention
The invention aims to solve the problems of diversification of the safety protection tool, interference of visual angle change and low accuracy of identification during multi-point operation in image identification.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a pedestrian target safety protection wear monitoring method based on artificial intelligence comprises image recognition, wherein the image recognition is a technology for recognizing and positioning objects in images to recognize targets and objects in various different modes. The identification process comprises the steps of image acquisition, image preprocessing, feature selection, extraction, classification decision and the like. And (3) completing image recognition target library construction by collecting and sorting defect pictures in service development as sample data, and completing image target detection algorithm library construction based on a neural network algorithm. Based on a strong visual library and an image algorithm, judging the wearing condition of the safety protection tool of the pedestrian target, establishing a safety protection tool sample database and an algorithm model, and intelligently detecting the detected pedestrian target
Preferably, the detected pedestrians are subjected to feature extraction and detection according to texture feature information of the safety appliance, so that whether illegal operation exists or not is judged.
Preferentially, according to the detection of the pedestrian frame by the YOLO, the hand area is selected in a self-adaptive mode to serve as a target area, and whether the constructor wears gloves can be judged according to relevant information according to wearing characteristics of the constructor.
Preferentially, as the types of the safety helmets worn by constructors are fixed, according to the geometric characteristic information of the safety helmets, the existing image library opencv can be used for threshold segmentation, the appointed types are extracted, the detected characteristic areas are filtered, and small noise is filtered, so that whether the safety helmets are worn or not is judged.
Preferentially, the target defects identified by the depth analysis image are selected, an optimal training algorithm is selected, and model training is carried out by combining with traditional image processing technologies such as image denoising and image enhancement. Image defect detection model frames trained based on neural network algorithms are used for constructing an image recognition model in a special power scene, intelligent applications such as power transmission line inspection, substation inspection, safety supervision field operation monitoring and the like in the power service field are formed, and power user experience and equipment on-line monitoring level are improved.
Further, according to the data transmitted back to the control center by the field monitoring equipment, the supervisory personnel remotely assist the supervisory field operation personnel to find out the abnormal operation and perform real-time early warning according to the field operation picture. The operation in the whole monitoring visual field can be monitored by the mode, but the operation condition in one monitoring visual field can only be monitored by a supervisory person at a time, so that the actual requirement of power operation is difficult to meet, and meanwhile, the condition is fluctuated due to different experiences of the supervisory person, so that the condition of irregular operation is difficult to be effectively avoided. The monitoring staff ' stares at ' the video picture that the on-site monitoring equipment transmits back to the control center, but in many cases, human beings are not a completely trusted observer, and no matter watch the real-time video stream or watch the video playback, because of own physiological weakness, we can't perceive the security threat often, thus lead to the occurrence of the phenomenon of missing report. Chinese patent document CN202010566737.5, which is disclosed in year 2020, month 19, discloses a safety harness wearing detection system based on 5G and deep learning target detection. Only solved the convenience that can be very big supervision personnel, reduced staff's security risk.
Even if artificial intelligence is adopted for monitoring, operators on duty can be relieved from the work of 'dead staring' monitors, but the problems that false alarm and missing alarm are the most common two problems in a video monitoring system still cannot be solved. Missing a report refers to when a security threat occurs at a monitored point, the threat not being found by the monitoring system or security personnel. False alarms refer to security activities at a monitoring point that are mistaken for a security threat, thereby producing false alarms. Chinese patent document CN112163572a, which is disclosed in 10/30/2020, discloses a method and apparatus for identifying an object. The technical problem of low detection efficiency caused by the illegal action of the manual inspection power construction site.
Conventional video surveillance systems are typically responded to and handled by security personnel for security threats that are sufficient to handle general, real-time response less demanding security threats. In many cases, however, multiple functional parts of the security system are required, and even multiple security-related departments are coordinated in a minimum amount of time to handle the crisis together when the threat occurs. At this time, the response speed of the monitoring system is directly related to the personal or property loss condition of the user.
Analysis of video data after an alarm is usually one of the tasks that security personnel have to do, and false alarm and missing alarm phenomena further exacerbate the need for data analysis. Security personnel are often required to find video data related to an alarm event, find a culprit, determine accident liability, or evaluate the security threat of the event. Because the traditional video monitoring system lacks intelligent factors, video data cannot be effectively classified and stored, and at most, only time labels can be marked, so that data analysis work becomes time-consuming and all relevant information is difficult to obtain, and frequently-occurring false alarm phenomenon further increases useless data, so that greater difficulty is brought to the data analysis work.
Compared with the prior art, the invention has the beneficial effects that:
due to the exquisite network structure of the YOLO, 150 frames per second can be achieved by matching with the GPU, which is enough to be used in any real-time system. In consideration of the requirement of the project on the aspect of real-time processing of data, the position of the binding box and the category of the binding box are directly regressed by using YOLO at an output layer, namely the whole graph is used as the input of a network, and the target detection problem is converted into a regression problem.
The special deep framework and the strong learning ability of deep learning enable the deep learning to learn the characteristics for distinguishing objects from a large amount of training data, and the characteristics have more outstanding performance in classification and recognition than the characteristics designed by people.
In view of the superior performance of the neural network in terms of extracting features, the convolutional neural network is adopted to extract the picture features, and the differences among the categories are mapped to a higher dimension, so that different categories are effectively distinguished.
Because the sizes of the feature areas detected according to the features of the safety helmet, the glove, the safety belt and the like are different, an algorithm for self-adaptive change of the area threshold value is provided, and the self-adaptive change is carried out by combining the area of the pedestrian frame, so that the generation of false detection and omission is effectively avoided.
Detailed Description
For a better understanding of the present invention, the following examples are set forth to further illustrate the invention, but are not to be construed as limiting the invention. In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that the invention may be practiced without one or more of these details.
Example 1
A pedestrian target safety protection wear monitoring method based on artificial intelligence comprises image recognition, wherein the image recognition is a technology for recognizing and positioning objects in images to recognize targets and objects in various different modes. The identification process comprises the steps of image acquisition, image preprocessing, feature selection, extraction, classification decision and the like. And (3) completing image recognition target library construction by collecting and sorting defect pictures in service development as sample data, and completing image target detection algorithm library construction based on a neural network algorithm. Based on a strong visual library and an image algorithm, judging the wearing condition of the safety protection tool of the pedestrian target, establishing a safety protection tool sample database and an algorithm model, and intelligently detecting the detected pedestrian target.
And extracting and detecting the characteristics of the detected pedestrians according to the texture characteristic information of the safety tool and the geometric characteristics, so as to judge whether illegal operation exists or not.
The YOLO elaborate network architecture, in combination with the GPU, can reach 150 frames per second, which is sufficient for any real-time system. In consideration of the requirement of the project on the aspect of real-time processing of data, the position of the binding box and the category of the binding box are directly regressed by using YOLO at an output layer, namely the whole graph is used as the input of a network, and the target detection problem is converted into a regression problem.
YOLO divides the input picture into individual grids, leaving the grid where the object center point is located responsible for detecting the object. Each grid predicts N bounding boxes, each bounding box being accompanied by a prediction of an confidence value in addition to its own position. This confidence value represents the confidence that the predicted bounding box contains the object and how accurate the bounding box predicts.
According to the pedestrian frame detected by the YOLO, the hand area is selected in a self-adaptive mode to serve as a target area, and whether the constructor wears gloves or not can be judged according to relevant information according to wearing characteristics of the constructor. The principle is as follows: the target region has a fixed characteristic space in an hsv space domain, the current video frame is converted into the hsv space, the characteristic information of the target region is detected according to a parameter threshold value of the characteristic space, and the geometric information of the target region is calculated according to the detected characteristic continuous region, so that the user is judged to not wear the glove correctly.
Since the size of the feature area detected based on the glove feature is different, a certain fixed threshold is given. The threshold value is too large, so that missed detection is easy to cause; too small a threshold value is prone to false detection. Through analysis, the size of the glove characteristic area changes along with the distance between the pedestrian and the monitoring equipment, and when the constructor is closer to the monitoring equipment, the pedestrian frame area is larger, and the glove characteristic area is larger; when constructors are far away from the monitoring equipment, the area of the pedestrian frame is smaller, and the characteristic area of the glove is reduced. Therefore, an algorithm for self-adaptive change of the area threshold is provided, and the area of the pedestrian frame is combined to be self-adaptively changed, so that false detection and missing detection are effectively avoided.
Example 2
The method for monitoring the wearing of the safety protection device based on the pedestrian target of artificial intelligence is different from the embodiment 1 in that: because the types of the safety helmets worn by constructors are fixed, according to the geometric characteristic information of the safety helmets, the existing image library opencv can be used for threshold segmentation, the appointed type is extracted, the detected characteristic area is filtered, and small noise is filtered, so that whether the safety helmets are worn or not is judged.
Since the feature areas detected according to the helmet features are different in size, a certain fixed threshold value is given. The threshold value is too large, so that missed detection is easy to cause; too small a threshold value is prone to false detection. Through analysis, the size of the characteristic area of the safety helmet changes along with the distance between the pedestrian and the monitoring equipment, and when constructors are closer to the monitoring equipment, the area of the pedestrian frame is larger, and the area of the safety helmet is larger along with the pedestrian frame; when constructors are far away from the monitoring equipment, the area of the pedestrian frame is smaller, and the area of the safety helmet is smaller. Therefore, an algorithm for self-adaptive change of the area threshold is provided, and the area of the pedestrian frame is combined to be self-adaptively changed, so that false detection and missing detection are effectively avoided.
Example 3
The method for monitoring the wearing of the safety protection device based on the pedestrian target of artificial intelligence is different from the embodiment 1 in that: and carrying out face recognition by adopting a neural network method. While neural network methods can obtain implicit expressions of these rules and rules through a learning process, neural network methods generally require a face to be input as a one-dimensional vector, so that input nodes are huge, and one important object of recognition is dimension reduction.
The face in the figure is detected, and a frame is marked for the face. After the face is detected, the face can be analyzed, and 72 key points such as the outline of eyes, mouths and noses can be obtained to accurately identify various face attributes such as gender, age, expression and the like. The method can be suitable for various actual environments such as large-angle side faces, shielding, blurring, expression change and the like.
When the pole tower works aloft, the double-safety belt with the backup rope should be used, the safety belt and the protection rope should be hung on the firm components at different parts of the pole tower, and the safety belt should be prevented from falling off from the pole top or being damaged by sharp objects. When the person indexes, the hand-held member should be firm and the protection of the backup protection rope should not be lost.
The camera device is arranged on the pole tower to ensure the safe operation of operators.
Based on the convolutional neural network model, the local perception and weight sharing structure of the convolutional neural network model enables the convolutional neural network model to be closer to a biological neural network in the real world, the weight sharing structure reduces the complexity of the neural network, the complexity of the feature extraction and classification process during data reconstruction is avoided, and the convolutional neural network model has the unique advantages in speech recognition and image processing.
Finally, it is noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present invention, and that other modifications and equivalents thereof by those skilled in the art should be included in the scope of the claims of the present invention without departing from the spirit and scope of the technical solution of the present invention.
Claims (5)
1. The method is characterized in that the image recognition is a technology for recognizing and positioning objects in images to recognize targets and objects in various modes, the recognition process comprises the steps of image acquisition, image preprocessing, feature selection, extraction, classification decision and the like, the construction of an image recognition target library is completed by collecting defect pictures in the development of finishing business as sample data, the construction of an image target detection algorithm library is completed based on a neural network algorithm, the wearing condition of the safety guard of the pedestrian target is judged based on a powerful visual library and an image algorithm, a safety guard sample database and an algorithm model are established, and the detected pedestrian target is intelligently detected.
2. The method for monitoring the wearing of the safety protection device based on the pedestrian target of the artificial intelligence according to claim 1, wherein the detected pedestrian is subjected to feature extraction and detection according to the texture feature information of the safety device, and the geometric feature is used for judging whether illegal operation exists or not.
3. The method for monitoring the wearing of the safety protection device based on the pedestrian target of claim 1, wherein the pedestrian frame is detected according to YOLO, the hand area is selected as the target area in a self-adaptive mode, and whether the constructor wears gloves or not can be judged according to the wearing characteristics of the constructor and the related information.
4. The pedestrian target safety protection wear monitoring method based on artificial intelligence according to claim 1, wherein the type of a safety helmet worn by a constructor is fixed, the existing image library opencv can be used for threshold segmentation according to the fixed geometric characteristic information of the safety helmet, the appointed type is extracted, the detected characteristic area is filtered out, and small noise is filtered, so that whether the safety helmet is worn or not is judged.
5. The pedestrian target safety protection tool wearing monitoring method based on artificial intelligence according to claim 1, wherein target defects identified by depth analysis images are selected, an optimal training algorithm is selected, model training is developed by combining traditional image processing technologies such as image denoising and image enhancement, an image defect detection model frame trained based on a neural network algorithm is constructed, an image recognition model under a power special scene is constructed, intelligent applications such as power transmission line inspection, substation inspection and safety supervision on-site operation monitoring in the power service field are formed, and power user experience and equipment on-line monitoring level are improved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111243169.6A CN116030405A (en) | 2021-10-25 | 2021-10-25 | Pedestrian target safety protection tool wearing monitoring method based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111243169.6A CN116030405A (en) | 2021-10-25 | 2021-10-25 | Pedestrian target safety protection tool wearing monitoring method based on artificial intelligence |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116030405A true CN116030405A (en) | 2023-04-28 |
Family
ID=86078150
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111243169.6A Pending CN116030405A (en) | 2021-10-25 | 2021-10-25 | Pedestrian target safety protection tool wearing monitoring method based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116030405A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117830956A (en) * | 2024-01-26 | 2024-04-05 | 南京关宁电子信息科技有限公司 | High-altitude live working safety protection method, system, equipment and storage medium |
-
2021
- 2021-10-25 CN CN202111243169.6A patent/CN116030405A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117830956A (en) * | 2024-01-26 | 2024-04-05 | 南京关宁电子信息科技有限公司 | High-altitude live working safety protection method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112396658B (en) | Indoor personnel positioning method and system based on video | |
| CN107145851A (en) | Constructions work area dangerous matter sources intelligent identifying system | |
| CN106412501A (en) | Construction safety behavior intelligent monitoring system based on video and monitoring method thereof | |
| CN113903081A (en) | Visual identification artificial intelligence alarm method and device for images of hydraulic power plant | |
| CN110765848A (en) | Chemical plant personnel safety guarantee system and early warning method based on artificial intelligence image processing algorithm | |
| CN113516076A (en) | Improved lightweight YOLO v4 safety protection detection method based on attention mechanism | |
| CN113989711B (en) | Power distribution construction safety tool use identification method and system | |
| CN109241847A (en) | The Oilfield Operation District safety monitoring system of view-based access control model image | |
| CN117035419B (en) | Intelligent management system and method for enterprise project implementation | |
| CN113111771A (en) | Method for identifying unsafe behaviors of power plant workers | |
| Wang et al. | A safety helmet and protective clothing detection method based on improved-yolo v 3 | |
| CN112287823A (en) | Facial mask identification method based on video monitoring | |
| CN110441320A (en) | A kind of gangue detection method, apparatus and system | |
| CN114926778A (en) | Safety helmet and personnel identity recognition system under production environment | |
| CN116030405A (en) | Pedestrian target safety protection tool wearing monitoring method based on artificial intelligence | |
| Li et al. | Application research of artificial intelligent technology in substation inspection tour | |
| CN113095160A (en) | Power system personnel safety behavior identification method and system based on artificial intelligence and 5G | |
| CN117173791A (en) | Distribution network constructor violation detection method and system based on action recognition | |
| CN116682034A (en) | Dangerous behavior detection method under complex production operation scene | |
| CN116311499A (en) | Wearing detection method and device for safety equipment | |
| CN101286238A (en) | Bionic intelligent monitoring system | |
| CN115829324A (en) | Personnel safety risk silent monitoring method | |
| CN115223249A (en) | Quick analysis and identification method for unsafe behaviors of underground personnel based on machine vision | |
| CN115860144A (en) | Machine learning system for anti-electricity-stealing site | |
| CN116030404A (en) | Artificial intelligence-based construction and safety monitoring method for electronic warning fence of operation area |
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 |