A power operation site 5G control ball operation method
By using portable 5G mobile surveillance cameras and deep learning technology for intelligent monitoring of power operation sites, the problem of low efficiency in power operation site monitoring has been solved, and efficient, real-time safety monitoring and data interaction have been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 国网黑龙江省电力有限公司齐齐哈尔供电公司
- Filing Date
- 2022-01-25
- Publication Date
- 2026-06-12
AI Technical Summary
The current lack of scientific methods for supervising power operations at the site leads to high labor and time costs, as well as low supervision efficiency, which affects the normal operation of the power system and the safety of workers.
Portable 5G mobile surveillance cameras are used for image acquisition, and deep learning technology is used for human face and safety helmet detection. The images are then transmitted wirelessly to the power monitoring center via 4G/5G to achieve intelligent edge analysis. In addition, multi-network adaptive modules are used to improve network transmission quality and enable real-time security management.
It improves regulatory efficiency, reduces false positives and false negatives, reduces server load, enables real-time detection and rapid analysis, supports multi-platform data interaction, and reduces labor and time costs.
Smart Images
Figure CN114550031B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of operation control technology, and specifically relates to an operation method for a 5G deployment ball at a power operation site. Background Technology
[0002] On-site operation management has always been a significant constraint on power safety. Current on-site operation management schemes lack scientific regulatory methods, seriously impacting the normal operation of the power system and the personal safety of workers. Ensuring the safety of workers while adopting a series of scientific methods to guarantee the normal operation of the power system has become the goal of on-site operation management. Consequently, regulatory departments have invested considerable resources in research and have achieved some initial results. Currently, the main method for on-site safety supervision is on-site inspection, directly dispatching regulatory personnel to the work site for on-site checks. This method is considered relatively reliable.
[0003] However, in practical applications, each work site is scattered and its location is not fixed. It is also affected by distance, traffic, weather conditions, etc., which not only makes the labor and time costs extremely high, but also increases the difficulty of on-site management and restricts the efficiency of supervision. Summary of the Invention
[0004] The purpose of this invention is to provide a 5G deployment ball and operation method for power operation sites in order to solve the problems mentioned above.
[0005] The technical solution adopted in this invention is as follows: a 5G deployment ball and operation method for power operation sites, comprising the following steps:
[0006] S1: Image acquisition is performed. On-site operation image acquisition is achieved by a mobile intelligent monitoring terminal, including a portable 5G mobile deployment ball and recorder, which collects video of the entire operation process and supports zooming in, zooming out, and 360° rotation.
[0007] S2: Extract the key points of the two images respectively, determine the mapping relationship by calculating the similarity between the key points, and then calculate the geometric transformation relationship. Based on this relationship, transform all the pixels of the second image to the new position to eliminate jitter.
[0008] S3: Use 4G / 5G wireless transmission to transmit the video images from step S2;
[0009] S4: Store images and data;
[0010] S5: Deep Learning for Face Edge Detection: Using deep learning technology, faces are located based on seven feature points: Sliding window detection is performed on the entire image; a coarse candidate region for the face is selected; the size of the coarse candidate region is normalized; the normalized image is fed into the CNN face detection network; and the seven feature points are regressed and output.
[0011] S6: Conduct safety helmet inspection; Before the start of the operation, inspect the safety helmets of the workers and monitor them in real time during the operation; During the inspection, the method of coarse candidate region division and size normalization is used to detect the safety helmet area, and a deep learning network with stacked cumulative suppression network layer is added to calculate the feature vector value of the safety helmet to quickly detect the safety helmet. RGB pixel detection is added to detect the color of the safety helmet.
[0012] S7: Shifting computing from the platform to the front end to achieve edge-based intelligent analysis. Edge computing involves transmitting encrypted and authorized database data to intelligent terminals at the work site, where the terminals perform target analysis and comparison identification.
[0013] S8: Combine image recognition processing with operation control. The key points of operation control are closely related to the control objectives. Before the operation begins, the operation conditions are approved. During the operation, the safety of front-end operators is controlled. On-site operation videos are uploaded in real time.
[0014] S9: The production site is equipped with multiple types of 4G / 5G video acquisition terminal equipment, which are connected to power monitoring centers at all levels through 4G / 5G wireless networks, and a real-time image transmission platform is deployed in the power supply bureau's monitoring center.
[0015] Preferably, in step S3, a multi-network adaptive module is used, and the terminal selects the operator according to the network conditions of the on-site operation location, and adopts a multi-mode binding simultaneous transmission method;
[0016] Preferably, in step S3, a multi-network adaptive module is used, specifically implemented as follows:
[0017] The device-side composite module of the multi-card router distributes the data sent by third-party network applications on the intranet to multiple wireless network cards and sends it to the first physical network card of the unbinding server. The first physical network card of the unbinding server transmits the received data to the server-side composite module of the unbinding server through the server-side socket interface.
[0018] The server-side composite module of the unbinding server first restores the data from the balanced transmission of the multi-card router to the data before balanced transmission, separates the data starting from the IP header, and sets a unique corresponding server source address by combining the source address and the network identifier of the multi-card router. It then establishes a network address translation table, performs network address translation, and then sends the IP packet to the virtual network card driver module. The virtual network card driver module simulates the IP packet as an IP packet received from the network and transmits it to the routing module. The routing module then passes it to the second physical network card that actually needs to send the data. Finally, the second physical network card sends it to the target service of the third-party network application.
[0019] The unbinding server transmits data sent by the target service of the third-party network application on the external network to the server-side composite module of the unbinding server. The server-side composite module then sends the data to the multiple wireless network cards of the multi-card router through the server-side socket interface and the first physical network card.
[0020] The device-side composite module of the multi-card router first restores the data from the unbinding server to the data before the balanced transmission, then unpacks the data to restore the data starting from the IP header, and then the internal network physical card and the device-side network card driver module send the data to the third-party network application on the internal network.
[0021] Preferably, in step S4, the video recording includes front-end storage and platform storage.
[0022] Preferably, in step S5, a deep learning network with stacked cumulative suppression network layers is added to calculate the feature vector value of the face, and the best matching vector is selected by calculating the Euclidean distance between the vectors to perform face recognition.
[0023] Preferably, in step S6, the identification process based on multi-feature fusion for identifying the safety helmet includes: normalizing the image of the safety helmet to be identified; improving HOG features to obtain the edge features of the safety helmet; obtaining the grayscale features of the safety helmet; obtaining the texture features of the safety helmet based on LBP feature theory; fusing grayscale features, edge features, and texture features to obtain a fused feature vector; and predicting the classification result based on an SVM classifier.
[0024] Preferably, in step S8, intelligent detection alarms and snapshots are taken for violations of operator rules, and photos are taken, reported, reviewed, summarized, saved, and exported for each work process.
[0025] Preferably, in step S7, after analysis and comparison, the development of a handheld mobile terminal for on-site operations is considered to match abnormal scene alarms and allow real-time viewing of on-site information. In summary, due to the adoption of the above technical solution, the beneficial effects of this invention are:
[0026] 1. In this invention, multiple video streams are combined into one file for storage during the image and data storage process, which reduces the number of seek times of the magnetic head and can significantly improve the performance of concurrent writing; the method also combines the recording source into the file, so that the recording source and the recording will not be separated when the file is transferred, which is beneficial to improving the analysis efficiency after the data is transferred.
[0027] 2. In this invention, a deep learning network with stacked cumulative suppression network layers is added to calculate the feature vector value of the face when detecting the edge of the human face. This method makes the detection results more accurate and reduces false positives and false negatives. At the same time, the model consumes less memory space and computing power, and the detection algorithm can be ported to the terminal for independent operation, reducing dependence on the network and server, resulting in lower latency and reduced pressure on the server. The cascaded structure combining fast moving target coarse detection and deep learning fine detection is adopted, which makes the target detection speed faster and enables real-time detection.
[0028] 3. In this invention, the platform system interface is open, supports platform docking, can push data to the power platform or other application platforms, or obtain data from other platforms, realize data interaction and a wider range of functional integration, and provide support for the development of richer applications. Attached Figure Description
[0029] Figure 1 This is a flowchart of the operation method of the 5G deployment ball at the power operation site according to the present invention. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0031] Reference Figure 1
[0032] Example:
[0033] This invention provides a method for operating a 5G surveillance camera at a power work site, comprising the following steps:
[0034] S1: First, image acquisition is performed. On-site operation image acquisition is achieved by a mobile intelligent monitoring terminal, including a portable 5G mobile deployment ball and recorder, which collects video of the entire operation process and supports operations such as zooming in, zooming out, and 360° rotation.
[0035] S2: First, extract the key points (such as corner points) of the two images respectively. Then, determine the mapping relationship by calculating the similarity between the key points, thereby calculating the geometric transformation relationship, which is generally an affine transformation system. Then, based on this relationship, transform all the pixels of the second image to the new position to eliminate jitter.
[0036] S3: Use 4G / 5G wireless transmission to transmit the video images from step S2;
[0037] S4: Store images and data;
[0038] S5: Deep learning is used for face edge detection; deep learning technology is used to locate faces based on seven feature points: sliding window detection is performed on the entire image; a coarse candidate region for the face is selected; the size of the coarse candidate region is normalized; the normalized image is fed into the CNN face detection network; the seven feature points are regressed and output.
[0039] S6: Conduct safety helmet inspection; Before the start of the operation, inspect the safety helmets of the workers and monitor them in real time during the operation; During the inspection, the method of coarse candidate region division and size normalization is used to detect the safety helmet area, and a deep learning network with stacked cumulative suppression network layer is added to calculate the feature vector value of the safety helmet to quickly detect the safety helmet. RGB pixel detection is added to detect the color of the safety helmet.
[0040] S7: Shifting computing from the platform to the front end to achieve edge-based intelligent analysis. Edge computing involves transmitting encrypted and authorized database data to intelligent terminals at the work site, where the terminals perform target analysis and comparison identification.
[0041] S8: Combine image recognition processing with operation control. The key points of operation control are closely related to the control objectives. Through the operation safety control system software, the operation conditions are approved before the operation begins, and the safety control of front-end operators is carried out during the operation. On-site operation videos are uploaded in real time.
[0042] S9: The production site is equipped with multiple types of 4G / 5G video acquisition terminal equipment, which are connected to power monitoring centers at all levels through 4G / 5G wireless networks, and a real-time image transmission platform is deployed in the power supply bureau's monitoring center.
[0043] In step S3, a multi-network adaptive module was first adopted, and the terminal selected the operator according to the network conditions of the on-site operation location. Secondly, a multi-mode binding simultaneous transmission method was studied, which can improve the wireless image transmission quality by more than 1.6 times and reduce the drop rate by 60%.
[0044] In step S3, a multi-network adaptive module is used, specifically implemented as follows:
[0045] (1): The device-side composite module of the multi-card router distributes the data sent by the third-party network application in the intranet to multiple wireless network cards and sends it to the first physical network card of the unbinding server. The first physical network card of the unbinding server transmits the received data to the server-side composite module of the unbinding server through the server-side socket interface.
[0046] (2): The server-side composite module of the unbinding server first restores the data from the balanced transmission of the multi-card router to the data before balanced transmission, separates the data starting from the IP header, and sets a unique corresponding server source address by combining the source address and the network identifier of the multi-card router. It then establishes a network address translation table, performs network address translation, and then sends the IP packet to the virtual network card driver module. The virtual network card driver module simulates the IP packet as an IP packet received from the network and transmits it to the routing module. The routing module then passes it to the second physical network card that actually needs to send data. Finally, the second physical network card sends it to the third-party network application target service.
[0047] (3): The unbinding server transmits the data sent by the target service of the third-party network application on the external network to the server-side composite module of the unbinding server, and then the server-side composite module sends the data to the multiple wireless network cards of the multi-card router through the server-side socket interface and the first physical network card.
[0048] (4): The device-side composite module of the multi-card router first restores the data from the unbinding server to the data before the balanced transmission, then unpacks the data to restore the data starting from the IP header, and then the internal network physical card and the device-side network card driver module send the data to the third-party network application in the internal network.
[0049] In step S4, video recording is mainly done in two ways: front-end storage and platform storage. When storing at the work site, the number of cameras stored at the same time is usually relatively small, and the performance requirements are not too high. However, since the deployment environment may be harsh, more attention is paid to the stability of storage and the speed of analysis.
[0050] In step S4, when storing data on the platform, the storage devices are all placed in the computer room, so the storage devices themselves have high stability. However, when storing data on the platform, there are usually a large number of cameras being managed, so the performance requirements for the storage are very high.
[0051] In step S5, a deep learning network with stacked cumulative suppression network layers is added to calculate the feature vector values of the face. The best matching vector is selected by calculating the Euclidean distance between the vectors to perform face recognition.
[0052] In step S6, the identification process based on multi-feature fusion for identifying safety helmets mainly includes: normalizing the image of the safety helmet to be identified; improving HOG features to obtain edge features of the safety helmet; obtaining grayscale features of the safety helmet; obtaining texture features of the safety helmet based on LBP feature theory; fusing grayscale features, edge features, and texture features to obtain a fused feature vector; and predicting the classification result based on an SVM classifier.
[0053] In step S8, intelligent detection alarms and snapshots are taken for violations of operator rules, and photos are taken, reported, reviewed, summarized, saved and exported for each work process; this reduces work approval time and ensures that no unqualified work order enters the work site and no required action to ensure basic safety is missed.
[0054] In step S9, the platform has the ability to access equipment video, intelligently analyze and process data, and store video recordings, providing video support for on-site supervision. The system platform can adopt a cascaded structure, with each county power department building its own platform to manage its subordinate resources, and then uniformly connecting to the municipal platform, thus balancing the localization of monitoring and the unification of management.
[0055] In step S7, after analysis and comparison, the development of a handheld mobile terminal for on-site operations can be considered to match abnormal scene alarms and view on-site information in real time.
[0056] The method combines multiple video streams into a single file during image and data storage, reducing the number of head seeks and significantly improving concurrent write performance. It also combines the recording source into the file, preventing separation of the recording source and the recording itself during file transfer, thus improving analysis efficiency. Furthermore, this method fully utilizes the file system's hard drive management capabilities, with no limit on the length of each data write, making the development of upper-level applications relatively simple.
[0057] When detecting edges in human faces, a deep learning network with stacked cumulative suppression network layers is added to calculate the feature vector values of the face. This method results in more accurate detection results with fewer false positives and false negatives. At the same time, the model consumes less memory and computing power, and the detection algorithm can be ported to the terminal for independent operation, reducing dependence on the network and server, resulting in lower latency and reduced server pressure. A cascaded structure combining fast moving target coarse detection and deep learning fine detection is adopted, which makes target detection faster and enables real-time detection.
[0058] The platform system has open interfaces, supports platform integration, and can push data to power platforms or other application platforms, or obtain data from other platforms, to achieve data interaction and a wider range of functional integration, and can provide support for the development of richer applications.
[0059] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0060] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for operating a 5G surveillance camera at a power work site, characterized in that: Includes the following steps: S1: Image acquisition is performed. On-site operation image acquisition is achieved by a mobile intelligent monitoring terminal, including a portable 5G mobile deployment ball and recorder, which collects video of the entire operation process and supports zooming in, zooming out, and 360° rotation. S2: Extract the key points of the two images respectively, determine the mapping relationship by calculating the similarity between the key points, and then calculate the geometric transformation relationship. Based on this relationship, transform all the pixels of the second image to the new position to eliminate jitter. S3: Use 4G / 5G wireless transmission to transmit the video images from step S2; In step S3, a multi-network adaptive module is used, and the terminal selects the operator according to the network conditions of the on-site operation location, and adopts a multi-mode binding simultaneous transmission method. In step S3, a multi-network adaptive module is used, specifically implemented as follows: The device-side composite module of the multi-card router distributes the data sent by third-party network applications on the intranet to multiple wireless network cards and sends it to the first physical network card of the unbinding server. The first physical network card of the unbinding server transmits the received data to the server-side composite module of the unbinding server through the server-side socket interface. The server-side composite module of the unbinding server first restores the data from the balanced transmission of the multi-card router to the data before balanced transmission, separates the data starting from the IP header, and sets a unique corresponding server source address by combining the source address and the network identifier of the multi-card router. It then establishes a network address translation table, performs network address translation, and then sends the IP packet to the virtual network card driver module. The virtual network card driver module simulates the IP packet as an IP packet received from the network and transmits it to the routing module. The routing module then passes it to the second physical network card that actually needs to send the data. Finally, the second physical network card sends it to the target service of the third-party network application. The unbinding server transmits data sent by the target service of the third-party network application on the external network to the server-side composite module of the unbinding server. The server-side composite module then sends the data to the multiple wireless network cards of the multi-card router through the server-side socket interface and the first physical network card. The device-side composite module of the multi-card router first restores the data from the unbinding server to the data before the balanced transmission, then unpacks the data to restore the data starting from the IP header, and then the internal network physical card and the device-side network card driver module send the data to the third-party network application on the internal network. S4: Store images and data; S5: Deep learning-based face edge detection: Using deep learning technology, faces are located based on seven feature points: sliding window detection is performed on the entire image; a coarse candidate region for the face is selected; the size of the coarse candidate region is normalized; The normalized image is fed into the CNN face detection network; The regression outputs seven feature points; In step S5, a deep learning network with stacked cumulative suppression network layers is added to calculate the feature vector value of the face, and the best matching vector is selected by calculating the Euclidean distance between the vectors to perform face recognition. S6: Conduct safety helmet inspection; Before the start of the operation, inspect the safety helmets of the workers and monitor them in real time during the operation; During the inspection, the method of coarse candidate region division and size normalization is used to detect the safety helmet area, and a deep learning network with stacked cumulative suppression network layer is added to calculate the feature vector value of the safety helmet to quickly detect the safety helmet. RGB pixel detection is added to detect the color of the safety helmet. In step S6, the identification process, based on multi-feature fusion, identifies the safety helmet, including: normalizing the image of the safety helmet to be identified; improving HOG features to obtain the edge features of the safety helmet; obtaining the grayscale features of the safety helmet; obtaining the texture features of the safety helmet based on LBP feature theory; fusing grayscale features, edge features, and texture features to obtain a fused feature vector; and predicting the classification result based on an SVM classifier. S7: Shifting computing from the platform to the front end to achieve edge-based intelligent analysis. Edge computing involves transmitting encrypted and authorized database data to intelligent terminals at the work site, where the terminals perform target analysis and comparison identification. S8: Combine image recognition processing with operation control. The key points of operation control are closely related to the control objectives. Before the operation begins, the operation conditions are approved. During the operation, the safety of front-end operators is controlled. On-site operation videos are uploaded in real time. S9: The production site is equipped with multiple types of 4G / 5G video acquisition terminal equipment, which are connected to power monitoring centers at all levels through 4G / 5G wireless networks, and a real-time image transmission platform is deployed in the power supply bureau's monitoring center.
2. The operation method of the 5G deployment ball at the power operation site as described in claim 1, characterized in that: In step S4, the video recording includes front-end storage and platform storage.
3. The operation method of the 5G deployment ball at the power operation site as described in claim 1, characterized in that: In step S8, intelligent detection alarms and snapshots are taken for violations of operator rules, and photos are taken, reported, reviewed, summarized, saved and exported for each work process.
4. The operation method of the 5G deployment ball at the power operation site as described in claim 1, characterized in that: In step S7, after analysis and comparison, the development of handheld mobile terminals for on-site operations is considered to match abnormal scene alarms and view on-site information in real time.