CN114463502A - Non-contact type safety distance active early warning method and device and storage medium - Google Patents

Non-contact type safety distance active early warning method and device and storage medium Download PDF

Info

Publication number
CN114463502A
CN114463502A CN202210043944.1A CN202210043944A CN114463502A CN 114463502 A CN114463502 A CN 114463502A CN 202210043944 A CN202210043944 A CN 202210043944A CN 114463502 A CN114463502 A CN 114463502A
Authority
CN
China
Prior art keywords
data
personnel
error
engineering vehicle
positioning
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
Application number
CN202210043944.1A
Other languages
Chinese (zh)
Inventor
李连鹏
刘宁
袁超杰
刘福朝
赵辉
苏中
李擎
范军芳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Information Science and Technology University
Original Assignee
Beijing Information Science and Technology University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Information Science and Technology University filed Critical Beijing Information Science and Technology University
Priority to CN202210043944.1A priority Critical patent/CN114463502A/en
Publication of CN114463502A publication Critical patent/CN114463502A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/05—Geographic models
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/10—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration
    • G01C21/12—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning
    • G01C21/16—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning by integrating acceleration or speed, i.e. inertial navigation
    • G01C21/183—Compensation of inertial measurements, e.g. for temperature effects
    • G01C21/188—Compensation of inertial measurements, e.g. for temperature effects for accumulated errors, e.g. by coupling inertial systems with absolute positioning systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
    • G08B21/02—Alarms for ensuring the safety of persons
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Geometry (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Business, Economics & Management (AREA)
  • Emergency Management (AREA)
  • Evolutionary Biology (AREA)
  • Automation & Control Theory (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Graphics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Navigation (AREA)

Abstract

The invention discloses a non-contact type safety distance active early warning method, a non-contact type safety distance active early warning device and a storage medium. Wherein, the method comprises the following steps: acquiring positioning data of the engineering vehicle and the operating personnel based on the inertial sensor; performing error compensation on the inertial sensor by using Kalman filtering and inertial pose error estimation, and obtaining positioning data after error compensation; acquiring the working site environment data of the engineering vehicle and the personnel; and determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data after error compensation and the environment data of the operation field, and actively early warning when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation. The invention solves the technical problem that the early warning cannot be effectively carried out due to inaccurate positioning.

Description

Non-contact type safety distance active early warning method and device and storage medium
Technical Field
The invention relates to the field of AI intelligence, in particular to a non-contact type active early warning method and device for a safety distance and a storage medium.
Background
The transformer substation is a hub of each level of voltage transmission lines, voltage lifting and resource distribution are carried out on power resources, the electromagnetic environment in the transformer substation is complex, the equipment spacing is small, and the line laying amount is greatly increased.
To SIS system, need place a large amount of safety monitoring instruments in operational environment, the cost is high, and can further compress originally narrow and small equipment interval, still have monitoring blind area and instrument scheduling problem that ages simultaneously, need carry out the input of manpower and materials for a long time.
For the DCS system, after a large amount of field information enters the computer control system, the sharing and interaction of the internal program and the field information of the computer lack a uniform connection standard, the industrial control software needs to develop a dedicated hardware device driver, once the hardware device is upgraded, the corresponding driver needs to be changed, and the system maintenance is not easy.
In addition, the DCS belongs to a process control system and a dynamic system, frequent manual intervention is needed, possibility of manual misoperation exists, a large amount of data needs to be processed by the DCS, and response speed is low, so that the adaptability of equipment with requirements on the response speed of an emergency stop system is low; usually, the SIS system and the DCS system cannot independently complete tasks, and need to be matched with each other to jointly maintain the safety of a working environment, so that the construction cost of the system is high.
In conclusion, with the development of the 5G technology and the mobile internet of things technology, aiming at the problems that the existing safety system is high in cost, dead zones exist in monitoring, site information benchmarks are not uniform and the like, the development of the new technology improves the safety of the working environment of the power station to a certain extent, guarantees the safety of constructors and equipment properties, fills up the technical blank in China, and ensures safe, reliable and efficient production of electric power, so that a non-contact type safe distance active early warning system is built, the digitization, informatization and science and technology of the transformer station are improved, and the development of the safety system of the transformer station is a further trend.
In view of the above problems, no effective solution has been proposed.
Disclosure of Invention
The embodiment of the invention provides a non-contact type active early warning method and device for a safety distance and a storage medium, and at least solves the technical problem that early warning cannot be effectively carried out due to inaccurate positioning.
According to an aspect of an embodiment of the present invention, a non-contact active safety distance early warning method is provided, including: acquiring positioning data of the engineering vehicle and the operating personnel based on the inertial sensor; performing error compensation on the inertial sensor by using Kalman filtering and inertial pose error estimation, and obtaining positioning data after error compensation; acquiring the working site environment data of the engineering vehicle and the personnel; and determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data after error compensation and the environment data of the operation field, and actively early warning when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation.
According to another aspect of the embodiments of the present invention, there is also provided a non-contact active warning device for safe distance, including: the precise positioning subsystem is configured to acquire positioning data of the engineering vehicle and the operating personnel based on the inertial sensor; the intelligent error compensation subsystem is configured to perform error compensation on the inertial sensor by utilizing Kalman filtering and inertial pose error estimation, and obtain positioning data after error compensation; the environment monitoring subsystem is configured to acquire work site environment data of the engineering vehicle and personnel; and the control and decision subsystem is configured to determine whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data subjected to error compensation and the environment data of the operation field, and actively early warn when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation.
According to still another aspect of embodiments of the present invention, there is also provided a computer-readable storage medium having stored thereon a program which, when executed, causes a computer to execute the method as described above.
In the embodiment of the invention, positioning data of an engineering vehicle and an operator are acquired based on an inertial sensor; performing error compensation on the inertial sensor by using Kalman filtering and inertial pose error estimation, and obtaining positioning data after error compensation; acquiring the working site environment data of the engineering vehicle and the personnel; and determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data after error compensation and the environment data of the operation field, and actively early warning when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation. By the method, the technical problem that early warning cannot be effectively performed due to inaccurate positioning is solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
fig. 1 is a flowchart of a non-contact active warning method for safe distance according to a first embodiment of the present invention;
FIG. 2 is a flow chart of a non-contact active warning method for safe distance according to a second embodiment of the present invention;
FIG. 3 is a flow chart of a workflow of a fine positioning subsystem according to an embodiment of the present invention;
FIG. 4 is a flow chart of the workflow of the error intelligent compensation subsystem according to an embodiment of the present invention;
FIG. 5 is a flow diagram of the workflow of the environmental monitoring subsystem according to an embodiment of the present invention;
FIG. 6 is a flow diagram of another workflow of the environmental monitoring subsystem according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of a non-contact active safety distance warning system according to an embodiment of the present invention;
FIG. 8 is a flow chart of a dynamic target fast perception method according to a first embodiment of the present invention;
FIG. 9 is a flow chart of a dynamic target fast perception method according to a second embodiment of the present invention;
FIG. 10 is a schematic diagram of feature extraction according to an embodiment of the present invention;
FIG. 11 is a flow chart of a dynamic target fast perception method according to a third embodiment of the present invention;
FIG. 12 is a flow chart of a dynamic target fast perception method according to a fourth embodiment of the present invention;
FIG. 13 is a schematic structural diagram of a dynamic target fast perception system according to an embodiment of the invention;
fig. 14 is a schematic structural diagram of a non-contact safe distance active warning system according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
SUMMARY
For dynamic target perception, dynamic target detection is the first step of recognition, namely, the changing region is accurately extracted from the background image from the sequence image. The main purpose of dynamic recognition perception is to determine an appropriate control quantity according to the angle deviation and the distance deviation obtained by path recognition so that the two deviations tend to be extremely small.
According to the method, a multi-source field information scale, time, relation and other multi-feature normalization expression method based on inertia, 5G, GNSS and a three-dimensional map database is adopted to construct a multi-space-time data unified reference frame, feature segmentation of map data is achieved, typical feature extraction and intelligent online recognition perception of a construction field live-action map are achieved by means of a sparse convolution neural network architecture, meanwhile, parameters are optimized and solved by means of a learning strategy, and rapid perception and recognition of dynamic targets under quantitative control of safety distances are achieved.
In addition, the safety control of a construction site is realized through the non-contact type active dynamic identification of the safety distance, the full management of the construction of an engineering project is effectively guaranteed, the safety, the timeliness and the science and technology of the construction of the power transmission and transformation project are greatly improved, and the safety of the construction of the power transmission and transformation project is effectively guaranteed; the aim of safe construction of all power transmission and transformation engineering machinery and personnel is achieved.
Example 1
According to an embodiment of the present invention, a non-contact type active early warning method for safety distance is provided, as shown in fig. 1, the method includes:
and S102, acquiring positioning data of the engineering vehicle and the operating personnel based on the inertial device.
Acquiring pose information of the engineering vehicle and the operating personnel on a construction site based on an inertial device such as an inertial sensor, and establishing a pose motion constraint model by analyzing the motion characteristics of the engineering vehicle and the operating personnel; estimating pose errors of the operation personnel and the engineering vehicle with motion constraint by using an optimal estimation method; and acquiring pose information of the operating personnel and the engineering vehicle by using a weak strapdown navigation failure error model.
Then, obtaining different frequency point signal intensities and multipath time delay clusters by using multi-frequency point channel estimation; the method comprises the steps of utilizing a multi-path space-time compensation model to realize accurate positioning in strong electric field interference so as to obtain auxiliary positioning information; and correcting the pose information of the operating personnel and the engineering vehicle by using the auxiliary positioning information to obtain corrected positioning data.
And S104, performing error compensation on the inertial device by using Kalman filtering and inertial pose error estimation, and obtaining error-compensated positioning data.
Using kalman filtering and inertial pose error estimation, error compensating the inertial device may include at least one of: compensating for a gyro output error of the inertial device based on at least one of: the method comprises the following steps of (1) inputting a gyro original output value vector, a rough calibration zero position output, a gyro angular velocity zero position output value vector, a temperature related zero position output, an acceleration related zero position output, a random noise signal, a rough calibration scale coefficient of a scale coefficient matrix, a scale coefficient of an angular velocity and a cross coupling coefficient matrix; and filtering and denoising the gyro noise under the dynamic state to compensate the random shaking error of the vehicle carrier.
The filtering and denoising processing on the gyro noise in the dynamic state may include: carrying out fast Fourier transform on the gyro noise in the motion state; carrying out first genetic operation on the gyro noise subjected to the fast Fourier transform, randomly generating a binary sequence C, and extracting a noise signal from the gyro noise subjected to the fast Fourier transform based on the binary sequence; performing inverse fast Fourier transform, converting the extracted noise signal into a continuous signal, and comparing the continuous signal with a gyro noise signal in a static state to obtain an inverted sequence; and multiplying the inverted sequence and the gyro noise after the fast Fourier transform to carry out filtering and noise reduction processing.
Step S106, acquiring the work site environment data of the engineering vehicle and the personnel;
for example, based on sensor self-calibration and system combined calibration, the environment of the operation field is sensed, and field data is collected; carrying out data preprocessing on the acquired data to filter interference factors and extract features; and performing multiple feature matching on the extracted features by using a three-dimensional map data effective feature point extraction method so as to reconstruct the three-dimensional scene of the engineering site in real time.
Then, segmenting superpixels of typical environmental feature data of the map data based on map data imaging expression technology and data automatic association analysis technology, and analyzing semantic relation; and carrying out target identification and active monitoring on the operators and the engineering vehicles in the three-dimensional scene of the engineering field by utilizing a sparse convolutional neural network architecture based on the semantic relation and the superpixels of the typical environment characteristic data of the segmented map data.
And S108, determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data after error compensation and the environment data of the operation field, and actively early warning when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation.
For example, based on the positioning data after error compensation and the working site environment data, classifying the risk states of the engineering vehicle and the working personnel by using a multidimensional scene linkage visualization engine; and determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on pre-configured early warning parameters and the classified risk state.
The method utilizes inertial positioning to detect the interference between the electrified construction environment and the vehicle, compensates the shaking error of the engineering vehicle, and solves the problem of space positioning in the weak/no-satellite environment; in addition, the satellite enhanced positioning technology is used for assisting inertial positioning, high-precision space positioning is provided, and safety control of a construction site is realized.
The non-contact active early warning method in the embodiment can perform quick, effective and safe early warning before an accident occurs, effectively avoids the loss of personnel and equipment, can improve the scientific and technological property and safety of the existing power plant, and promotes the digitization, informatization and scientific and technological upgrading of the power plant.
Example 2
According to the embodiment of the invention, a non-contact type active early warning method for the safety distance is provided. The method can be applied in the following scenarios: the project application environment comprises 5 conditions such as a 500-dry-V alternating-current field, a 220-dry-V alternating-current field, a 35-dry-branch alternating-current field, a 500-dry-V flexible direct-current field, a large-scale deep foundation pit operation scene and the like. As shown in fig. 2, the method comprises the steps of:
step S201, a precise positioning subsystem is called.
The precise positioning subsystem takes inertial autonomous positioning orientation as a core, a satellite navigation positioning enhancement technology as an auxiliary, and establishes a multi-source fusion autonomous precise positioning system according to factors such as signal intensity, environmental conditions, motion forms and the like.
And step S202, calling an error intelligent compensation subsystem.
The intelligent error compensation subsystem performs periodic division and characteristic extraction on the vehicle working inertia data, establishes a model by using a deep neural network, and solves compensation by using Kalman filtering and inertial pose error estimation.
Step S203, calling an environment monitoring subsystem.
The environment monitoring subsystem establishes a data unified reference frame under multiple time and space based on inertia, 5G, GNSS and a three-dimensional map database, extracts effective characteristic points, and completes fusion display by combining a deep learning algorithm and using a visual display technology.
The environment monitoring subsystem mainly executes the following three steps: risk superposition evaluation, early warning evaluation, multi-dimensional risk evaluation, grading early warning and multi-dimensional statistical analysis.
And (3) risk superposition evaluation: and a multidimensional scene linkage visualization engine is utilized to realize visualization simulation display, realize accurate positioning of equipment and operation on a live-action map, classification of risk points and risk state display, and realize hierarchical display of risk superposition evaluation results in an area on the map.
Early warning evaluation: and establishing a multi-target resource scheduling model for emergency rescue under the multilevel grid, realizing visualization and nodularization of identifying early warning parameter configuration, visualized display and operation of flow, and realizing self-defined flexible configuration of early warning evaluation indexes.
Multidimensional risk assessment, hierarchical early warning and multidimensional statistical analysis: the method comprises the steps of establishing a visual intelligent recognition early warning platform supporting an edge calculation micro-service framework and elastic dynamic convergence, supporting multi-dimensional risk assessment, grading early warning and multi-dimensional statistical analysis, and comprehensively supporting monitoring field command decision.
And step S204, judging whether the operator operates by mistake.
Step S205, a control and decision subsystem is invoked.
The control decision subsystem is based on the subsystems and carries out safety protection and active early warning when the field operation personnel has misoperation or exceeds a safety distance.
Step S206, an alarm is sent out and response is made.
Step S207, no alarm is given, and the monitoring is continued.
The embodiment has effectively solved construction safety distance quantization control, the monitoring means is single, environmental monitoring data volume is big, the blank scheduling problem of construction environment monitoring, through non-contact safety distance initiative dynamic identification, eliminate constructor, the error of vehicle, make safety precaution before the occurence of failure, realize job site safety management and control, avoid taking place serious production accident, the safety control of engineering project construction has effectively been ensured, the security of power transmission and transformation engineering construction has been improved by a wide margin, ageing and science and technology nature, effectively ensure the construction of power transmission and transformation engineering, and the scheme has the universality, can be applied to most work scenes after the demand suitable improvement according to different work scenes.
In addition, by unifying multi-source data under the unified space-time reference, the data management, statistics and arrangement efficiency is obviously improved, the problem that data processing in the system is numerous and complicated is solved, the technical blank of lack of wearable portable high-precision positioning equipment in China is filled, the standardization, the productization and the industrialization of key technologies, information services and instruments and equipment are promoted, and the scientific and technological guarantee is provided for the continuous stable safe operation and development of a power plant.
Example 3
According to the embodiment of the invention, a non-contact type active early warning method for the safety distance is provided. The steps in this embodiment are similar to those in embodiments 1 and 2, except for the method performed by the fine positioning sub-system.
The precise positioning subsystem comprises an inertia autonomous positioning module and an elastic fusion precise positioning technology module. The work flow of the precise positioning subsystem is shown in fig. 3, and includes the following steps S302 to S304.
And step S302, acquiring pose information based on the inertial device.
An inertial autonomous positioning module of the precise positioning subsystem acquires pose information of the engineering vehicle and the operating personnel on the construction site, and a vehicle pose motion constraint model is established through analyzing the motion characteristics of the personnel and the engineering vehicle. And (3) obtaining a motion constraint personnel and vehicle pose error estimation method by using an optimal estimation method, and calculating a pose error which best meets the actual situation. And by combining a weak strapdown navigation failure error model, the accurate acquisition of the pose information of the operating personnel and the engineering vehicle is realized.
And step S304, performing accurate positioning based on the 5G region fusion.
An elastic fusion accurate positioning technology module of the accurate positioning subsystem uses a 5G-based regional elastic fusion accurate co-positioning method to perform channel estimation on multiple frequency points to obtain the strength of different frequency point signals, multipath time delay cluster distribution and other information, and establishes a multipath space-time compensation model to realize accurate positioning in a strong electric field interference environment.
Example 4
According to the embodiment of the invention, a non-contact type active early warning method for the safety distance is provided. The steps in this embodiment are similar to those in embodiments 2 and 3, except for the method performed by the error intelligent compensation subsystem. The work flow of the error intelligent compensation subsystem is shown in fig. 4.
First, a gyro output error characteristic is analyzed.
Error sources include deterministic errors of inertial devices, random errors, random vehicle wobble errors, and the like. The errors of the inertial device are divided into: zero error, scale coefficient error, cross coupling error, random error.
Firstly, analyzing the output error characteristics of the gyroscope, wherein an error model is as follows:
Figure RE-GDA0003589568380000091
Figure RE-GDA0003589568380000092
outputting a vector for the angular velocity after error compensation; a. theqThe original output value vector of the triaxial gyroscope is obtained; n is a radical of1For coarse calibration zero position output, N represents a gyro angular velocity zero position output value vector, N (T) is temperature-related zero position output, and N (a) is acceleration-related zero position output; v (t) is a random noise signal; the scale factor matrix is represented by SF, SF1For coarse scaling of the scale factor, SF2(Ω) is a proportionality coefficient corresponding to the input angular velocity; cross couplingThe coefficient matrix is denoted by CR.
And secondly, correcting the random shaking error of the vehicle carrier.
1) And establishing an inertial sensor model by using a deep neural network method.
2) According to the self-adaptive moment estimation principle, the sensing precision is improved by utilizing the parameter optimization method of the engineering vehicle sensor.
And finally, filtering and denoising the gyro noise under the dynamic condition.
The method selects a gyro noise signal in a static state as a reference, and in order to improve the quality of the gyro signal under normal walking, the gyro noise under the dynamic state is a main processing object of a filtering noise reduction algorithm, and comprises the following steps:
1) and carrying out fast Fourier transform on the gyro data with noise in the motion state.
2) The first genetic manipulation is performed to randomly generate a binary sequence C.
3) And performing fast Fourier inverse transformation, converting the extracted noise signals into continuous signals, and comparing the continuous signals with gyro noise signals in a static state.
4) And performing multiplication operation on the inverted sequence and the gyro noise after the motion state fast Fourier transform to complete the whole filtering and noise reduction process.
In the embodiment, the vehicle working inertia data is subjected to periodic division and characteristic extraction, a model is established by using a deep neural network, and compensation is solved by using Kalman filtering and inertia pose error estimation, so that the engineering vehicle and operators can be positioned more accurately.
Example 5
According to the embodiment of the invention, a non-contact type active early warning method for the safety distance is provided. The steps in this embodiment are similar to those in embodiments 2 through 4, except for the method performed by the environmental monitoring subsystem.
In the embodiment, the positioning data under a space rectangular coordinate system or a space geodetic coordinate system is adopted, and point cloud data and the positioning data under different benchmarks are required to be unified under the same space-time benchmark. The working flow of the environment monitoring subsystem of the present embodiment is shown in fig. 5, and includes the following steps S502 to S506.
And step S502, environment perception.
Based on calibration means such as sensor self-calibration and system combined calibration, the environment sensing of the operation field is carried out.
And step S504, feature extraction and real-time reconstruction of the three-dimensional scene.
And carrying out data preprocessing on the acquired data to finish interference factor filtering and feature extraction. The map data is segmented by utilizing a map data imaging expression technology and a data automatic association analysis technology through superpixel segmentation and semantic relation analysis of typical environment characteristic data. And performing multiple feature matching by using a three-dimensional map data effective feature point extraction method to realize real-time reconstruction of a three-dimensional scene of a construction site.
And step S506, actively early warning.
And combining deep learning, and realizing typical target identification and active monitoring on the construction site live-action map by utilizing a sparse convolutional neural network architecture. And (5) fusing the instantaneous attitude high-precision calculation sum of the IMU and 5G, GNSS, and completely acquiring the scene.
Example 6
According to the embodiment of the invention, a non-contact type active early warning method for the safety distance is provided. The steps in this embodiment are similar to those in embodiments 2 through 4, except for the method performed by the environmental monitoring subsystem.
In the embodiment, the positioning data under a space rectangular coordinate system or a space geodetic coordinate system is adopted, and point cloud data and the positioning data under different benchmarks are required to be unified under the same space-time benchmark. The working flow of the environment monitoring subsystem of the present embodiment is shown in fig. 6, and includes the following steps S602 to S606.
Step S602, environmental perception.
First, data acquisition is performed, for example, map data, inertial data, and 5G + GNSS data are acquired.
And then carrying out system calibration. Based on calibration means such as sensor self-calibration and system combined calibration, the environment sensing of the operation field is carried out.
Step S604, information processing.
The information processing mainly comprises two parts: data processing and data fusion.
The data processing comprises data preprocessing, multivariate segmentation, feature extraction and the like. Data fusion includes uniform benchmarking, feature levels, and the like. And carrying out data preprocessing on the acquired data through information processing to finish interference factor filtering and feature extraction.
Data fusion is mainly a unified benchmark and determines the feature level.
Step S606, environment reconstruction and target positioning.
The environment reconstruction and target positioning mainly comprise: environmental modeling, target identification, and active monitoring.
The environment modeling is mainly based on environment characteristics, and three-dimensional reconstruction is carried out through multiple characteristic matching. The target identification mainly comprises template matching and feature extraction. Active monitoring is mainly target localization and target monitoring.
In the embodiment, a map data imaging expression technology and a data automatic association analysis technology are combined, a superpixel segmentation and semantic relation analysis method of typical environment characteristic data is used for achieving the segmentation of map data, deep learning is combined, and a sparse convolution neural network architecture is used for achieving typical target recognition and active monitoring of a real scene map on a construction site.
Example 7
According to the embodiment of the invention, a non-contact type active early warning device for the safe distance is provided. As shown in fig. 7, the non-contact active safety distance early warning device in this embodiment includes:
the precise positioning subsystem 72 is configured to acquire positioning data of the engineering vehicle and the operator based on the inertial device; an error intelligent compensation subsystem 74 configured to perform error compensation on the inertial device by using kalman filtering and inertial pose error estimation, and obtain error-compensated positioning data; an environment monitoring subsystem 76 configured to acquire work site environment data of the engineering vehicle and the personnel; the control and decision subsystem 78 is configured to determine whether the work vehicle or the personnel is within the substation non-contact safety distance based on the error compensated positioning data and the job site environment data, and to proactively warn if outside the substation non-contact safety distance.
The non-contact active safety distance early warning device in this embodiment can implement the methods in embodiments 1 to 6, and details are not repeated here.
The system comprises an accurate positioning subsystem, an error intelligent compensation subsystem, an environment monitoring subsystem and a control and decision subsystem, is respectively responsible for personnel and equipment positioning, disturbance and noise elimination, working environment monitoring of a power plant and safety measure response functions, and has the beneficial effect of accurate early warning.
An important component in the non-contact construction safety distance active dynamic identification early warning is dynamic target perception. For dynamic target perception, dynamic target detection is the first step of recognition, namely, the changing region is accurately extracted from the background image from the sequence image. The main purpose of dynamic recognition perception is to determine an appropriate control quantity according to the angle deviation and the distance deviation obtained by path recognition so that the two deviations tend to be extremely small.
The following examples will focus on the dynamic object fast sensing method. On the basis of the dynamic target rapid perception method, when the dynamic target exceeds a safety distance threshold value, active early warning can be carried out. In other words, the dynamic target fast sensing method in the following embodiments is part of a non-contact construction safety distance active dynamic identification early warning method. The non-contact construction safety distance active dynamic identification early warning method comprises the following steps: the early warning method is to actively carry out early warning when the dynamic target identified by the dynamic target fast sensing method exceeds a safe distance threshold.
Example 8
According to an embodiment of the present invention, a method for fast sensing a dynamic target based on deep learning is provided, as shown in fig. 8, the method includes:
s802, based on the combined calibration of a sensing system, environment sensing is carried out, and data are collected from different data sources;
step S804, scene acquisition is carried out based on the acquired data, and map data and pose information are acquired;
step S806, based on the obtained map data and pose information, preprocessing the normalized data of the multivariate features, and performing feature matching on the preprocessed data of the normalized data through superpixel segmentation and semantic analysis.
In an exemplary embodiment, based on the acquired map data and pose information, the preprocessing of the normalized data of the multivariate features comprises at least one of: carrying out normalization data preprocessing on the information scale on the acquired map data and pose information; preprocessing the time series normalization data of the acquired map data and pose information; and carrying out global normalization data preprocessing on the acquired map data and pose information.
In an exemplary embodiment, the preprocessing of the normalized data of the information scale on the acquired map data and pose information includes: performing normalization processing on the acquired map data and pose information, wherein the normalization processing comprises at least one of the following steps: unifying data encoding modes of data from different data sources into the same mode; unifying data formats of data from different data sources into the same format; matching the map data and the pose information after the standardization processing, wherein the matching processing comprises at least one of the following steps: matching different attributes of different data sources identifying the same thing together; identifying identical data of different data sources as duplicates; and carrying out data screening on the map data and the pose information after matching processing to select consistent dimension data.
In an exemplary embodiment, the time-series normalized data preprocessing the acquired map data and pose information includes: based on the pattern correlation between adjacent time sequences and the characteristic that data are continuously generated in the time dimension, the time registration of the acquired map data and pose information from different data sources is performed by a subsequent algorithm.
In an exemplary embodiment, the global normalization data preprocessing of the acquired map data and pose information includes: after windowing and segmenting the time sequence, acquiring a plurality of segments of time subsequences with fixed window lengths; and based on the relation factor of the adjacent time sequences, carrying out normalized data preprocessing on the time subsequences in the current window by utilizing the maximum value and the minimum value of the time subsequences in the current window.
In an exemplary embodiment, before feature matching the normalized data preprocessed data by superpixel segmentation and semantic analysis, the method further comprises: defining an error state vector of the perception system as a column vector; determining angular velocity and acceleration of inertial devices of the sensing system in relation to measurement values based on the column vectors; determining an update equation for the inertial device based on the determined angular velocity and acceleration of the inertial device versus the measured values; determining an attitude error of the inertial device using the determined update equation.
For example, determining the attitude error of the inertial device using the determined update equation includes: obtaining a transfer matrix between each frame of the map data according to the map feature point extraction and matching relation; and resolving corresponding attitude angle information by using the transfer matrix, using the angle increment change value of the inertial device as a weight adjustment factor, and resolving an attitude error based on the attitude angle information.
And step S808, based on the data obtained by the feature matching, sensing the dynamic target by utilizing the fusion of a sparse convolution network and motion estimation.
The embodiment unifies the standard of the data under multiple time and space conditions, and improves the multi-source information processing capability of the data; the sparse convolutional neural network architecture is utilized to realize the typical feature extraction and intelligent online identification of the construction site live-action map, is suitable for various targets, and has strong self-adaptive capacity and high identification accuracy.
The embodiment can be applied to a live construction site and used for quickly identifying operators and engineering vehicles. On one hand, the method can effectively reduce the occurrence probability of power failure accidents, personal accidents and equipment accidents caused by subjective experience and traditional sensor signal judgment at present, effectively solves the main potential safety hazards existing in the live-line construction operation process, on the other hand, the method is simultaneously beneficial to supervision of a manager in the construction process, further improves the image of the power grid industry, effectively improves the power supply reliability and the user power utilization satisfaction degree, and has obvious social benefits, so that the method provided by the embodiment has practical value.
In addition, the non-contact safe distance active dynamic target identification algorithm provided by the embodiment is a multi-feature normalization expression method based on inertia, 5G, GNSS and multi-source field information scale, time, relation and the like of a three-dimensional map database, a multi-space-time data unified reference frame is constructed, feature segmentation of map data is realized, a sparse convolutional neural network architecture is combined with a deep learning algorithm to realize typical feature extraction and intelligent online identification and sensing of a construction field live-action map, meanwhile, a learning strategy is used for optimizing parameters, a space unified reference is that targets to be sensed and environment pose information are established in a unified coordinate system or can be obtained through matrix transformation, and identification and monitoring of different space targets are improved.
In addition, the sensing system in the embodiment can rapidly and accurately calculate the state and the category parameters of the dynamic target in a complex environment, and the accuracy and the rapidity of the dynamic target identification are obviously improved.
Besides being applied to live construction sites, the dynamic target rapid sensing method provided by the embodiment can be widely applied to various fields, and provides a new technical route for realizing a dynamic identification rapid sensing technology with universality, initiative, high precision and high cost performance. In particular, the dynamic target fast perception strategy based on deep learning provided by the embodiment can be applied to target recognition monitoring based on deep learning in other fields, so that the method provided by the embodiment has stronger universality.
Example 9
According to the embodiment of the invention, a dynamic target rapid perception method based on deep learning is provided.
The dynamic target rapid perception method based on deep learning is realized on the premise that a perception system, scene acquisition, normalized expression and feature extraction are completed. The whole process includes the steps that firstly, a perception system collects multi-source data by using a three-dimensional map library, a 5G + GNSS and an inertia device, information scale, time and space relation scene information is obtained, then, normalization expression of multiple features is carried out, and a three-dimensional operation scene is constructed; based on methods such as a nonlinear least square method, a relaxation method, a Bayes estimation theory and the like, super-pixel segmentation and semantic analysis are carried out to carry out feature matching, and a dynamic target perception autonomous learning model is established by utilizing information obtained by feature matching through a sparse convolution neural network and a motion estimation fusion model to complete rapid identification and live-action map display of a dynamic target.
The multi-element characteristic normalization in the application refers to multi-element characteristic normalization expression of multi-source field information scale, time, relation and the like. The target identification means that a part belonging to a target is screened from the spatial or temporal distribution data information and characteristic judgment is carried out.
Fig. 9 is a flowchart of a method for fast sensing a dynamic target based on deep learning according to this embodiment, and as shown in fig. 9, the method includes:
step S902, the sensing system senses environmental data.
The sensing system of the live working site mainly comprises 5G, GNSS, an inertial device and a high-precision three-dimensional map library.
Step S904, scene acquisition.
And obtaining map data, and pose information of the operating personnel and the engineering vehicle according to the information obtained by the sensing system.
And step S906, normalizing expression and feature matching.
Aiming at a field operation environment, multi-feature normalized expression of information scale, time and space relation is completed based on inertia, 5G, GNSS and multi-source data of a three-dimensional map database, a three-dimensional operation scene is constructed, and superpixel segmentation and semantic analysis are carried out based on methods such as a nonlinear least square method, a relaxation method and a Bayes estimation theory, so as to carry out feature matching. In addition, the map data is filtered and segmented, feature data is extracted, and real-time reconstruction of the three-dimensional scene of the construction site is achieved through multiple feature matching. The process of feature extraction is shown in fig. 10.
In this embodiment, the normalized expression mainly includes normalized expression of information scale, normalized expression of time series, and global normalized expression.
1. Information scale normalization method
1) Standardization aims to make data encoding modes, data formats and the like of different data sources identical, and lays a foundation for next data matching (code standardization process in data standardization).
2) Matching, wherein the data matching work has two aspects, namely, different attributes of different data sources for identifying the same object are matched together, so that the data is more complete; and the other is that the same data of different data sources are identified into repetition, so that the basis is laid for the next screening. For example: from duplicate customer names in different data sources.
3) And screening, wherein the main purpose of data screening is to select a consistency dimension as main data, namely consistency dimension data delivered finally.
2. Time series normalization method
The time registration of multi-source data is mainly completed in time series normalization, and due to the particularity of the time series, such as mode correlation between adjacent series, the data is a characteristic which is generated continuously in a time dimension, and the selection of a normalization method is also beneficial to the subsequent algorithm/model work as much as possible.
3. Global normalization method
On the basis of global normalization, the concept of windowing in wavelet/Fourier transform is combined to generate windowed normalization.
The time sequence in the window is normalized by utilizing the maximum value and the minimum value of the sequence in the window, and the method solves the problem of sequence amplitude difference under different measurement conditions when the time span is large to a certain extent. Given a time sequence S, the window length of S is n segments after S is segmented by windowingThe sequence of L constitutes: s ═ S1，s2，…，sn. Performing self-adaptive normalization on the S, wherein the sequence after normalization is as follows: r ═ R1，r2，…，rnTherein of
Figure RE-GDA0003589568380000171
Figure RE-GDA0003589568380000172
Figure RE-GDA0003589568380000173
Where α and β are relationship factors of adjacent time series, i ═ 1, 2iIs an iterative process variable, di-1Is an iterative process variable, riIs the normalized sequence, fiIs an iterative process variable, fi-1Is an iterative process variable.
In step S908, a dynamic object is identified.
The target identification process is to screen out the part belonging to the target from the spatial or temporal distribution data information and to perform characteristic judgment. Machine learning is an advantageous tool to address such problems. In the machine learning process, a proper algorithm and a classifier are adopted and proper prior characteristics are selected to identify the dynamic target.
Example 10
According to the embodiment of the invention, a dynamic target rapid perception method based on deep learning is provided.
In this embodiment, the environment sensing of the operation site is performed based on calibration means such as sensor self-calibration and system joint calibration. And carrying out data preprocessing on the acquired data to finish interference factor filtering and feature extraction. And the live-action map fusion comprises feature level and space unified reference fusion. By researching an effective characteristic point extraction method of three-dimensional map data, real-time reconstruction of a three-dimensional scene of a construction site is realized through multiple times of characteristic matching; the map data image expression technology and the data automatic correlation analysis technology are combined, the super-pixel segmentation and semantic relation analysis method of typical environment characteristic data is researched, the map data is segmented, deep learning is combined, and the sparse convolution neural network architecture is utilized to realize rapid identification of typical dynamic targets of the real scene map on the construction site.
First, context awareness is performed. After the data acquisition is completed, for example, map data, inertial data, and 5G + GNSS data are acquired. And then carrying out system calibration. Based on calibration means such as sensor self-calibration and system combined calibration, the environment sensing of the operation field is carried out.
Next, information processing is performed. The information processing mainly comprises two parts: data processing and data fusion. The data processing comprises data preprocessing, multivariate segmentation, feature extraction and the like. Data fusion includes uniform benchmarking, feature levels, and the like. And carrying out data preprocessing on the acquired data through information processing to finish interference factor filtering and feature extraction. Data fusion is mainly a unified benchmark and determines the feature level.
And finally, carrying out environment reconstruction and target positioning. The environment reconstruction and target positioning mainly comprise: environmental modeling, target identification, and active monitoring. The environment modeling is mainly based on environment characteristics, and three-dimensional reconstruction is carried out through multiple characteristic matching. The target identification mainly comprises template matching and feature extraction. Active monitoring is mainly target localization and target monitoring.
In the embodiment, a map data imaging expression technology and a data automatic association analysis technology are combined, a superpixel segmentation and semantic relation analysis method of typical environment characteristic data is used for achieving the segmentation of map data, deep learning is combined, and a sparse convolution neural network architecture is used for achieving typical target recognition and active monitoring of a real scene map on a construction site.
Example 11
According to the embodiment of the invention, a dynamic target rapid perception method based on deep learning is provided. This method is similar to that in embodiments 8 to 10, except that a dynamic object is identified.
FIG. 11 is a flow chart of a method of dynamic object recognition according to an embodiment of the present invention, as shown in FIG. 11, the method comprising the steps of:
in step S1102, information data is acquired.
And the perception system performs environment perception to acquire environment data.
In step S1104, the column vector is expanded.
The inertial, 5G, GNSS systematic error state vector is defined as a 15 x 1 column vector:
Figure RE-GDA0003589568380000191
wherein the content of the first and second substances,
Figure RE-GDA0003589568380000192
defining the difference value of attitude angle errors measured by IMU, 5G and GNSS at the moment k as a three-dimensional column vector comprising a roll angle, a pitch angle and a yaw angle;
Figure RE-GDA0003589568380000193
zero offset value output by the gyroscope; delta rkThe three-axis position error under the navigation system; delta vkThe three-axis speed error under the navigation system;
Figure RE-GDA0003589568380000194
is the zero offset value output by the accelerometer.
The relationship of the angular velocity ω and the acceleration a to the measured value can be expressed as:
Figure RE-GDA0003589568380000195
wherein, ω ismFor measuring angular velocity, ngFor the compensation of angular velocity under the navigation system, amFor measuring acceleration, naThe acceleration compensation quantity under the navigation system.
The update equation described by the dynamic system state space convolution model can be expressed as:
Figure RE-GDA0003589568380000196
wherein k is a discrete time,
Figure RE-GDA0003589568380000197
is the system noise; gkIs the noise matrix coefficient; vkTo observe the noise matrix; hkIs an observation matrix; zkAs an observed quantity, Xk+1Is the state of the system at time (k +1), Xk+1∈Rn，Zk+1∈RmAs an observation matrix for the corresponding signal, f (X)k) As a function of the state of the system at time k,
Figure RE-GDA0003589568380000198
for transfer matrices, omega, of IMU to navigation coordinate systemsbAngular velocity under IMU, abAcceleration under IMU. f (X)k) The definition is as follows:
Figure RE-GDA0003589568380000199
wherein the content of the first and second substances,
Figure RE-GDA00035895683800001910
is the difference in the attitude angle error of the gyroscope output,
Figure RE-GDA00035895683800001911
is the zero offset value of the accelerometer output,
Figure RE-GDA0003589568380000201
zero bias, Δ t, for gyroscope outputIFor IMU sampling time interval of Δ tCThe time interval is sampled for the camera as,
Figure RE-GDA0003589568380000202
for transferring map pose to IMU poseMoving matrix (known from IMU mounting structure)
Figure RE-GDA0003589568380000203
Is a constant value),
Figure RE-GDA0003589568380000204
is the attitude angle error at time k.
A transfer matrix R between frames can be obtained according to the map feature point extraction and matching relationship, and then the attitude matrix in the navigation coordinate system can be expressed as:
Figure RE-GDA0003589568380000205
wherein, C11To C33Representing the acquired attitude data information, and solving the corresponding attitude angle information:
Figure RE-GDA0003589568380000206
(7)
wherein the content of the first and second substances,
Figure RE-GDA0003589568380000207
is the attitude angle at time k, ΨkCYaw angle at time k, θkCAt time k, pitch angle, γkCRoll angle at time k.
Wherein, the attitude error as an observed quantity can be expressed as:
Figure RE-GDA0003589568380000208
wherein the content of the first and second substances,
Figure RE-GDA0003589568380000209
for the difference of the attitude angles of the combination positions of the adaptive dynamic matching,
Figure RE-GDA00035895683800002010
is the attitude angle measured under IMU and 5G,
Figure RE-GDA00035895683800002011
attitude angles measured for GNSS.
The combined pose resolving attitude error of the adaptive dynamic matching can be expressed as:
Figure RE-GDA00035895683800002012
wherein the content of the first and second substances,
Figure RE-GDA00035895683800002013
in order to adjust the factors for the weights,
Figure RE-GDA00035895683800002015
is the average angular increment per unit of time,
Figure RE-GDA00035895683800002014
is the attitude error. The angular increment change value of the IMU is used as the calculation amount of the weight adjustment factor gamma, so that the problem that the solution cannot be realized due to overlarge positioning error or failure is solved.
Step S1106, three-dimensional reconstruction.
After the information scale, time and space relation scene information is acquired, the normalization expression of the multivariate characteristics is carried out, and a three-dimensional operation scene is constructed.
In step S1108, the output relationship is input.
During deep learning training, the optimizer optimizes the model parameters to find a set of optimal solutions. The input-output parameter relationship of each layer in the convolutional neural network can be closely related by the weight W of the neuron link, the number of input-output channels and the like.
In step S1110, the data feature analysis result is deeply learned.
And establishing a dynamic target perception autonomous learning model by utilizing a sparse convolutional neural network and a motion estimation fusion model to complete the rapid identification of the dynamic target and the display of a live-action map.
In the embodiment, the IMU instantaneous attitude high-precision calculation and the 5G, GNSS attitude calculation are fused, the advantages of time drift and the like do not exist, the problem of inertial attitude calculation random divergence is solved, the IMU attitude measurement advantages are brought into play, the attitude calculation accuracy is further improved, and the integrity map fusion recognition under dynamic motion is realized.
Example 12
According to an embodiment of the present invention, there is provided a dynamic target fast perception method based on deep learning, as shown in fig. 12, the method includes:
step S1201, pose information is obtained.
And the three-axis accelerometer and the three-axis gyroscope acquire data, and the acquired data is settled through strapdown inertial navigation to obtain the position, the speed and the posture.
In step S1202, motion control is triggered.
The zero velocity and the zero angular velocity are adjusted and corrected through negative feedback of motion control, the advantage of IMU attitude measurement is played, the accuracy of attitude calculation is further improved, and integrity map fusion under dynamic motion is realized.
In step S1203, position information is acquired.
And extracting and matching the features based on the three-dimensional map database to obtain the position information of the engineering vehicle and the operator.
And step S1204, performing error correction on the position information and the pose information.
And inputting the obtained pose information such as position, speed and attitude data and the position information into an extended Kalman filter, and performing error correction on the calculation results of the strapdown inertial navigation of the inertial device and the calculation results of the position information of the map.
And step S1205, pose updating is carried out.
And performing depth reconstruction based on the position information to realize three-dimensional reconstruction. And performing loop monitoring and similarity calculation based on the position, speed and attitude information solved by the strapdown inertial navigation. And updating the semi-dense map and the pose based on the output result of the three-dimensional reconstruction and the output result of the similarity calculation to acquire a complete scene.
Example 13
According to an embodiment of the present invention, there is provided a dynamic target fast perception system based on deep learning, as shown in fig. 13, the system includes:
the perception module 132 is configured to perform environmental perception based on perception system joint calibration, and collect data from different data sources; a scene acquisition module 134 configured to perform scene acquisition based on the acquired data, and acquire map data and pose information; a data processing module 136 configured to perform normalization data preprocessing of multiple features based on the acquired map data and pose information, and perform feature matching on the data after normalization data preprocessing by superpixel segmentation and semantic analysis; and the identification module 138 is configured to sense the dynamic target by utilizing sparse convolution network and motion estimation fusion based on the data obtained by feature matching.
The system in this embodiment is capable of implementing the method of any one of embodiments 8 to 12, and therefore, the details are not repeated here.
Example 14
According to the embodiment of the invention, a non-contact active early warning system for safe distance is provided, and the system is shown in fig. 14 and comprises a visual recognition early warning platform 142, a positioning reference station 143 and a plurality of mobile positioning terminals 141.
The visual recognition early warning platform 142 comprises a 5G and deep learning based non-contact type safe distance active early warning system which runs on a Linux operating system. The master station is based on a good dimensional industrial personal computer TPCIPC-100 and an Intel Bay trailer J1900 processor, and the processor has stable performance; the memory adopts DDR3 memory technology, and the size is 8G; the hard disk is a 512G SSD solid state disk and a 1T HDD mechanical hard disk. In the aspect of software, a wisdom tooth Windows and Linux platform can change 422/485 protocol according to a default RS232 protocol. The positioning reference station 143 selects a dry-searching magic cube MC262M PRO GNSS positioning module, a dual-frequency RTK algorithm is built in, a BDS B2a frequency band is selected, and the positioning precision is as follows: 0.02+1ppm, and the speed measurement precision is less than 0.05 m/s. The mobile positioning terminal 141 selects PA-IMU-03D as the mobile positioning terminal, and the parameters thereof are as follows: the starting time is less than or equal to 5S, the measuring range is +/-200 degrees/S, the zero offset repeatability is less than or equal to 0.3 degrees/h, the zero offset stability is less than or equal to 0.3 degrees/h, the scale factor nonlinearity is less than or equal to 200ppm, the scale factor repeatability is less than or equal to 200ppm, the bandwidth is more than or equal to 300Hz, the random walk coefficient is less than or equal to 0.05 degrees/h, the type quartz accelerometer has the measuring range of +/-10 g (various measuring ranges can be selected), the offset value is less than or equal to 1mg, the offset value monthly comprehensive error is less than 5 multiplied by 10 < -5 > g (1 sigma), the offset value temperature coefficient is less than or equal to 50 ug/DEG C, the scale factor monthly comprehensive error is less than 80ppm (1 sigma), and the scale factor temperature coefficient is less than 80 ppm/DEG C.
During the operation of the system, for example, three mobile positioning terminals are installed at designated positions on the engineering vehicle (carrier), and are respectively installed above a cab a, a highest point B of a boom and an operation room C of the engineering vehicle as shown in fig. 14. The mobile positioning terminal of the point A is used for measuring the actual position and the vehicle posture of the vehicle, and the mobile positioning terminals of the point B and the point C are used for measuring the extension length of the suspension arm and the rotation angle of the suspension arm relative to the carrier. And the position information of A, B, C points can complete the real-time information acquisition of the position and the angle of the engineering vehicle.
Real-time information of the position and the angle of the engineering vehicle is transmitted back to the visual recognition early warning platform, the platform utilizes the existing dangerous area coordinates to compare and recognize with the position information of the engineering vehicle, and active early warning is carried out after the set early warning safety threshold value is exceeded. The visual identification early warning platform in this embodiment is equivalent to the non-contact active early warning device for the safety distance in embodiment 7, and can implement the methods in embodiments 1 to 6, and therefore, details are not repeated here.

Claims (10)

1. A non-contact type active early warning method for safety distance is characterized by comprising the following steps:
acquiring positioning data of the engineering vehicle and the operating personnel based on the inertial sensor;
performing error compensation on the inertial sensor by using Kalman filtering and inertial pose error estimation, and obtaining positioning data after error compensation;
acquiring the working site environment data of the engineering vehicle and the personnel;
and determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data after error compensation and the environment data of the operation field, and actively early warning when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation.
2. The method of claim 1, wherein using kalman filtering and inertial pose error estimation to error compensate the inertial sensors comprises at least one of:
compensating for a gyro output error of the inertial sensor based on at least one of: the method comprises the following steps of (1) inputting a gyro original output value vector, a rough calibration zero position output, a gyro angular velocity zero position output value vector, a temperature related zero position output, an acceleration related zero position output, a random noise signal, a rough calibration scale coefficient of a scale coefficient matrix, a scale coefficient of an angular velocity and a cross coupling coefficient matrix;
and filtering and denoising the gyro noise under the dynamic state to compensate the random shaking error of the vehicle carrier.
3. The method of claim 1, wherein filtering and denoising the gyro noise in a dynamic state comprises:
carrying out fast Fourier transform on the gyro noise in the motion state;
carrying out first genetic operation on the gyro noise subjected to the fast Fourier transform, randomly generating a binary sequence C, and extracting a noise signal from the gyro noise subjected to the fast Fourier transform based on the binary sequence;
performing inverse fast Fourier transform, converting the extracted noise signal into a continuous signal, and comparing the continuous signal with a gyro noise signal in a static state to obtain an inverted sequence;
and multiplying the inverted sequence and the gyro noise after the fast Fourier transform to carry out filtering and noise reduction processing.
4. The method of claim 1, wherein acquiring positioning data of the work vehicle and the operator based on the inertial sensor comprises:
acquiring pose information of the engineering vehicle and the operating personnel on a construction site, and establishing a pose motion constraint model by analyzing the motion characteristics of the engineering vehicle and the operating personnel;
estimating pose errors of the operation personnel and the engineering vehicle with motion constraint by using an optimal estimation method;
and acquiring pose information of the operating personnel and the engineering vehicle by using a weak strapdown navigation failure error model.
5. The method of claim 4, wherein after acquiring positioning data of the work vehicle and the operator based on the inertial sensor, the method further comprises:
obtaining different frequency point signal intensity and multipath time delay clusters by using multi-frequency point channel estimation;
the method comprises the steps of utilizing a multi-path space-time compensation model to realize accurate positioning in strong electric field interference so as to obtain auxiliary positioning information;
and correcting the pose information of the operating personnel and the engineering vehicle by using the auxiliary positioning information to obtain corrected positioning data.
6. The method of claim 1, wherein obtaining work site environment data for the work vehicle and personnel comprises:
sensing the environment of the operation site based on sensor self-calibration and system combined calibration, and collecting site data;
carrying out data preprocessing on the acquired data to filter interference factors and extract features;
and performing feature matching on the extracted features for multiple times by using a three-dimensional map data effective feature point extraction method so as to reconstruct the three-dimensional scene of the engineering site in real time.
7. The method of claim 6, wherein after reconstructing the three-dimensional scene of the project site in real-time, the method further comprises:
segmenting superpixels of typical environmental feature data of the map data based on map data imaging expression technology and data automatic association analysis technology, and analyzing semantic relation;
and carrying out target identification and active monitoring on the operators and the engineering vehicles in the three-dimensional scene of the engineering field by utilizing a sparse convolutional neural network architecture based on the semantic relation and the superpixels of the typical environment characteristic data of the segmented map data.
8. The method of claim 1, wherein determining whether the work vehicle or the personnel is within the contactless safe distance of the substation based on the error-compensated positioning data and the job site environment data comprises:
classifying the risk states of the engineering vehicle and the operating personnel by utilizing a multidimensional scene linkage visualization engine based on the positioning data after error compensation and the operating site environment data;
and determining whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on pre-configured early warning parameters and the classified risk state.
9. A non-contact type active early warning device for safety distance is characterized by comprising:
the precise positioning subsystem is configured to acquire positioning data of the engineering vehicle and the operating personnel based on the inertial sensor;
the intelligent error compensation subsystem is configured to perform error compensation on the inertial sensor by utilizing Kalman filtering and inertial pose error estimation, and obtain positioning data after error compensation;
the environment monitoring subsystem is configured to acquire work site environment data of the engineering vehicle and personnel;
and the control and decision subsystem is configured to determine whether the engineering vehicle or the personnel is within the non-contact safety distance of the transformer substation based on the positioning data subjected to error compensation and the environment data of the operation field, and actively early warn when the engineering vehicle or the personnel is out of the non-contact safety distance of the transformer substation.
10. A computer-readable storage medium having stored thereon a program which, when executed, causes a computer to perform the method of any one of claims 1 to 8.
CN202210043944.1A 2022-01-14 2022-01-14 Non-contact type safety distance active early warning method and device and storage medium Pending CN114463502A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210043944.1A CN114463502A (en) 2022-01-14 2022-01-14 Non-contact type safety distance active early warning method and device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210043944.1A CN114463502A (en) 2022-01-14 2022-01-14 Non-contact type safety distance active early warning method and device and storage medium

Publications (1)

Publication Number Publication Date
CN114463502A true CN114463502A (en) 2022-05-10

Family

ID=81409161

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210043944.1A Pending CN114463502A (en) 2022-01-14 2022-01-14 Non-contact type safety distance active early warning method and device and storage medium

Country Status (1)

Country Link
CN (1) CN114463502A (en)

Similar Documents

Publication Publication Date Title
CN107481292B (en) Attitude error estimation method and device for vehicle-mounted camera
CN114463932B (en) Non-contact construction safety distance active dynamic identification early warning system and method
KR20210052409A (en) Lane line determination method and apparatus, lane line positioning accuracy evaluation method and apparatus, device, and program
JP2013186899A (en) Method and system for monitoring structure
DE112016006692T5 (en) Method for predicting a movement of an object
CN104809724A (en) Automatic precise registration method for multiband remote sensing images
CN114140761A (en) Point cloud registration method and device, computer equipment and storage medium
CN111145251A (en) Robot, synchronous positioning and mapping method thereof and computer storage device
CN115205718B (en) Geographic information measuring system and measuring method thereof
US11928813B2 (en) Method and system for detecting change to structure by using drone
CN115063017A (en) Monitoring and evaluating system and method for small and medium-span bridge structure
CN110287455A (en) A kind of PM2.5 deep learning inversion method of combination remotely-sensed data and social perception data
CN114445404A (en) Automatic structural vibration response identification method and system based on sub-pixel edge detection
CN109147322B (en) Multi-source data self-adaptive fusion method in urban traffic big data processing
CN115311617A (en) Method and system for acquiring passenger flow information of urban rail station area
CN114092660A (en) High-precision map generation method and device and vehicle for generating map
CN110532969B (en) Slope unit dividing method based on multi-scale image segmentation
CN117115063A (en) Multi-source data fusion application method
CN114463502A (en) Non-contact type safety distance active early warning method and device and storage medium
CN115542362A (en) High-precision space positioning method, system, equipment and medium for electric power operation site
CN114386522A (en) Dynamic target rapid sensing method and system based on deep learning
CN113762153B (en) Novel tailing pond detection method and system based on remote sensing data
CN115761265A (en) Method and device for extracting substation equipment in laser radar point cloud
JP2023152480A (en) Map data generating method, map data generating apparatus, and map data generating program
CN114741806A (en) Suspension optimization method, system, device, equipment and medium

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