Method and system for positioning personnel under mine

Abstract

The invention relates to a method and system for positioning personnel in a mine, and the method comprises the steps: determining any one person in the mine as a target person, and obtaining the multi-source positioning information of the target person: baseline label positioning information, geomagnetic positioning information and pose positioning information; obtaining multi-source positioning information corresponding to each time node of the target person in a preset historical time window, and generating an adjacent multi-source feature vector; the adjacent multi-source feature vectors are input into a pre-trained decision contribution prediction model, a decision contribution factor sequence corresponding to the multi-source positioning information is output, and decision contribution factors corresponding to all kinds of positioning information in the multi-source positioning information form the decision contribution factor sequence based on the multi-source positioning information and the decision contribution factor sequence of the multi-source positioning information. And inputting into a preset multi-source offset fusion algorithm model, and outputting target positioning information of the target person. Therefore, more accurate and more reliable underground personnel positioning is realized through differentiated decisions of different source positioning information.

Description

Technical Field

[0001] The present invention relates to the technical field of mine safety monitoring, and particularly relates to a method and system for positioning personnel underground in a mine. Background Art

[0002] Underground positioning is the basic guarantee for realizing underground safety production and supervision and management. At present, the underground positioning systems commonly installed underground mainly use electromagnetic wave emission and signal recognition methods to achieve underground positioning, such as RFID technology, Zigbee technology, PHS technology, WI-FI technology, etc. The positioning technology services of these methods are mainly for the management level to realize the supervision and management of underground personnel or dynamic targets. Their positioning process depends on the underground power supply and communication systems. Once there is an unstable situation in the underground power supply or signal, the positioning device cannot work normally and cannot meet the requirements of emergency positioning and rescue positioning in case of emergencies; in addition, for the research on the underground geomagnetic positioning method based on magnetic characteristics, due to the fact that the geomagnetic positioning technology is easily affected by metals such as iron, it is difficult to adapt to complex mine environments, and the positioning accuracy deviation is relatively large.

[0003] The Chinese invention patent with the patent application number 202210024380.7 discloses a method and device for underground positioning based on Baseline-RFMDR. The first positioning of underground personnel is obtained based on the angle between underground personnel and the baseline tags of the mine roadway, the second positioning of underground personnel is obtained based on geomagnetic data for geomagnetic matching, and the third positioning of underground personnel is obtained based on the average walking step length, the number of walking steps, and the walking direction; the fusion calculation with the minimum error is performed on the first positioning, the second positioning, and the third positioning to obtain the accurate positioning of underground personnel; thus, it has less dependence on the underground power supply and communication systems and is more suitable for complex mine environments.

[0004] However, in the actual underground positioning scenario in a mine, if there are outliers (due to reasons such as equipment failures and environmental interferences) in multiple positioning results, then the fusion calculation process may still be affected by these outliers and fluctuate, and it is easy to fall into the situation of the global optimal solution of multiple positionings, resulting in the accurate positioning result having no local tendency and thus deviating from the actual positioning. Summary of the Invention

[0005] In view of the technical problems existing in the prior art, the present invention provides a method and system for positioning personnel underground in a mine, and realizes more accurate and reliable underground personnel positioning through differential decision-making of positioning information from different sources.

[0006] The technical solution for the present invention to solve the above technical problems is as follows: A method for positioning personnel underground in a mine, comprising: S101. Designate any person in the current mine as the target person and obtain the multi-source positioning information of the target person. The multi-source positioning information includes positioning information from three sources: baseline tag positioning information, geomagnetic positioning information, and pose positioning information. The types of positioning information sources are denoted as a, b, and c respectively. S102. Obtain the multi-source positioning information corresponding to each time node within a preset historical time window for the target person, and generate an adjacent multi-source feature vector. S103. Input the adjacent multi-source feature vector into a pre-trained decision contribution prediction model, and output a decision contribution factor sequence corresponding to the multi-source positioning information. The decision contribution factors corresponding to each type of positioning information in the multi-source positioning information form the decision contribution factor sequence. S104. Based on the multi-source positioning information and its decision contribution factor sequence, input them into a preset multi-source offset fusion algorithm model, and output the target positioning information of the target person.

[0007] Preferably, the method for obtaining the baseline tag positioning information is as follows: Obtain the tag signals of multiple groups of baseline tags in the area, including tag coordinates and tag signal emission angles. Select the group of baseline tags with the largest tag signal emission angle as the main baseline tag. Based on the main baseline tag, select an adjacent group of baseline tags as the secondary baseline tag. Use the coordinate information of the main baseline tag and the secondary baseline tag, as well as the tag signal emission angle, to obtain the baseline tag positioning information of the target person.

[0008] Preferably, the method for obtaining the adjacent multi-source feature vector specifically includes: A1. Within the preset historical time window, collect the multi-source positioning information of each time node, including baseline tag positioning information, geomagnetic positioning information, and pose positioning information. A2. Based on the baseline tag positioning information, geomagnetic positioning information, and pose positioning information and their corresponding time nodes respectively, generate a number of data points. The longitude in each time node and its positioning information constitutes a data point, and perform fitting to obtain a trajectory curve within the preset historical time window. The horizontal axis of the trajectory curve represents the time node, and the vertical axis represents the longitude. A3. Based on the trajectory curves corresponding to each type of positioning information in the multi-source positioning information, calculate the offset values between the trajectory curves of the other two types of positioning information respectively, and generate offset pairs. A4. Combine the offset pairs corresponding to the baseline tag positioning information, geomagnetic positioning information, and pose positioning information respectively to form an adjacent multi-source feature vector.

[0009] Preferably, the offset value is calculated according to the following formula:

[0010] Wherein, D is the offset value between the trajectory curves of two positioning information, t1 is the start time node of the historical time window, and t2 is the end time node of the historical time window. is the longitude value of the trajectory curve of positioning information of type a at time node t. is the longitude value of the trajectory curve of positioning information of type b at time node t.

[0011] Preferably, the neighboring multi-source feature vector is denoted as [( , ), ( , ), ( , )], where ( , ) is the offset pair of baseline label positioning information, ( , ) is the offset pair of geomagnetic positioning information, ( , ) is the offset pair of pose positioning information. is the offset value between the trajectory curve of type a positioning information and the trajectory curve of type b positioning information. is the offset value between the trajectory curve of type a positioning information and the trajectory curve of type c positioning information.

[0012] Preferably, the acquisition method of the pre-trained decision contribution prediction model includes: B1. Collect the neighboring multi-source feature vectors of a large number of underground mine personnel at historical moments, and label each neighboring multi-source feature vector. The labeling content is set as: decision contribution factor sequence; specifically, the labeling method includes: the sequence composed of the ratio of the distance values between the actual positioning information corresponding to the historical moment and the corresponding baseline label positioning information, geomagnetic positioning information, and pose positioning information, as the decision contribution factor sequence of the neighboring multi-source feature vector. B2. Use all the labeled historical neighboring multi-source feature vectors as the training data set, and use the training data set to train the pre-selected neural network structure, optimize the model parameters, and obtain the final decision contribution prediction model.

[0013] Preferably, the pre-set multi-source offset fusion algorithm is:

[0014] Wherein, is the target positioning information of the target person. are the decision contribution factors of the baseline label positioning information, geomagnetic positioning information, and pose positioning information respectively. is the distance value between the baseline tag positioning information and the target positioning information, is the distance value between the geomagnetic positioning information and the target positioning information, is the distance value between the pose positioning information and the target positioning information.

[0015] Preferably, after the S103, the method further includes: S201, obtaining the underground mine image information collected by the roadway monitoring device closest to the main baseline tag at the current moment; S202, inputting the underground mine image information, the shooting parameters of the monitoring device, and the main baseline tag coordinate information into a pre-trained underground mine target recognition model to obtain the visual positioning information of the target person; wherein, the underground mine target recognition model includes a pixel classification model, a visual positioning prediction module, and a target person determination module. Inputting the underground mine image information into the pixel classification model to output an underground mine classification map; inputting the underground mine classification map and the shooting parameters into the visual positioning prediction module to output all target objects and their visual positioning information; inputting all target objects and their visual positioning information, and the main baseline tag coordinate information into the target person determination module to output the visual positioning information of the target person; S203, based on the relationship between the visual positioning information of the target person and the multi-source positioning information, correcting the decision contribution factor sequence to obtain an updated decision contribution factor sequence to replace the original decision contribution factor sequence.

[0016] Preferably, in the S203, correcting the decision contribution factor sequence specifically includes: D1. Calculate the correction factor of each decision contribution factor according to the following formula:

[0017]

[0018] wherein, i and j are used to represent the type numbers of the multi-source positioning information, including a, b, c, is the distance value between the visual positioning information and the i-th type of multi-source positioning information, is the reliability value of the i-th type of multi-source positioning information, is the original decision contribution factor of the i-th type of multi-source positioning information, is the distance value between the visual positioning information and the j-th type of multi-source positioning information, is the reliability value of the j-th type of multi-source positioning information, is a preset positive parameter for controlling the attenuation speed; D2. Use the correction factor of each decision contribution factor as the new decision contribution factor to generate a new decision contribution factor sequence.

[0019] A personnel positioning system underground in a mine, comprising: an acquisition module, a decision-making module, and a fusion positioning module; The acquisition module is used to determine any one person underground in the current mine as the target person, and obtain multi-source positioning information of the target person. The multi-source positioning information includes positioning information from three sources: baseline tag positioning information, geomagnetic positioning information, and pose positioning information. The types of positioning information sources are respectively denoted as a, b, and c; obtain the multi-source positioning information corresponding to each time node of the target person within a preset historical time window, and generate an adjacent multi-source feature vector; The decision-making module is used to input the adjacent multi-source feature vector into a pre-trained decision contribution prediction model, and output a decision contribution factor sequence corresponding to the multi-source positioning information. The decision contribution factors corresponding to each type of positioning information in the multi-source positioning information form a decision contribution factor sequence; The fusion positioning module is used to input the multi-source positioning information and its decision contribution factor sequence into a preset multi-source offset fusion algorithm model, and output the target positioning information of the target person.

[0020] The beneficial effects of the present invention are: The adjacent multi-source feature vector and the decision contribution prediction model are introduced to evaluate the contribution degree of the positioning information from each source. The multi-source offset fusion algorithm is adopted, and the decision contribution factor is combined to calculate the target positioning information, improving the accuracy and robustness of the positioning. By considering the relative consistency and difference of the multi-source positioning information, the influence that outliers may have on the positioning result is reduced. By down-weighting the current possible outliers through the decision contribution factor, the problem that the fusion calculation process may be affected by fluctuations when there are outliers in the multi-positioning fusion result is solved, realizing more accurate and reliable underground personnel positioning; By introducing visual positioning information and combining it with the multi-source positioning information for correction, the positioning result is made more accurate; the decision contribution factors of the positioning information of different source types are dynamically adjusted, reducing the influence of outliers on the positioning result, making up for the possible error defects of the model, and improving the robustness of the positioning; by obtaining image information and calculating correction factors in real time, the update and correction of real-time positioning information can be supported. Description of the Drawings

[0021] Figure 1 It is a schematic flowchart of a method for positioning personnel underground in a mine according to an embodiment of the present invention; Figure 2 It is a schematic structural diagram of a personnel positioning system underground in a mine according to an embodiment of the present invention. Detailed Embodiments

[0022] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0023] In the description of the present application, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present application, "a plurality of" means two or more, unless otherwise specifically defined.

[0024] In the description of the present application, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present application is not necessarily construed as being more preferred or having more advantages than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without the use of these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present application.

[0025] Embodiment 1 Figure 1 It is a schematic flowchart of a method for positioning personnel underground in a mine according to an embodiment of the present invention.

[0026] As Figure 1 shown, a method for positioning personnel underground in a mine includes the following steps: S101, determine any one person underground in the current mine as the target person, and obtain multi-source positioning information of the target person. The multi-source positioning information includes positioning information from three sources: baseline tag positioning information, geomagnetic positioning information, and pose positioning information. The types of positioning information sources are respectively denoted as a, b, and c.

[0027] Specifically, all personnel underground in a mine carry a positioning device. Several groups of baseline tags with equal density are equidistantly arranged on the side walls of the mine tunnels. Each group of baseline tags is symmetrically arranged on both sides of the tunnels. These tags are similar to barcodes or RFID tags in supermarkets and are used to identify specific locations in the tunnels. The baseline tags include tag coordinates, which are used to transmit signals to be received by the positioning devices carried by the personnel. The personnel's baseline tag positioning information is obtained as follows: the positioning device obtains the tag signals of multiple groups of baseline tags in the area, including tag coordinates and tag signal emission angles (the tag signal emission angle represents the angle between the tag signal emission direction and the tunnel direction, which is used to determine the relative position of the underground personnel and the tags). A group of baseline tags with the largest tag signal emission angle is selected as the main baseline tag, which usually means that the relative position of the underground personnel and this group of tags is closest or the signal is the strongest. Based on the main baseline tag, a secondary baseline tag is further selected to assist in calculating the position of the underground personnel. The coordinate information of the main baseline tag and the secondary baseline tag, as well as the tag signal emission angle, is used to calculate the position coordinates of the underground personnel through geometric relationships.

[0028] The method of obtaining geomagnetic positioning information is to obtain geomagnetic data of the path that underground personnel walk through in the mine. Geomagnetic data is specific data of the earth's magnetic field in the mine, similar to GPS signals, but not affected by sky obstructions. The measured geomagnetic data is compared with the pre-stored geomagnetic map to determine the location of the underground personnel. The geomagnetic map is pre-surveyed in the mine and contains geomagnetic features of various locations in the mine.

[0029] The method for acquiring the posture positioning information is as follows: based on the posture sensor installed in the positioning device, the average walking step length, number of steps, and walking direction of the underground personnel in the mine are obtained. The average walking step length is the average distance of each step when the underground personnel walk; the number of steps is the number of steps taken from the starting point to the current position; the walking direction is the direction in which the underground personnel walk (the angle based on a certain direction); based on the walking data, the posture positioning information is obtained.

[0030] It should be noted that for the detailed calculation and acquisition process of the baseline tag positioning information, geomagnetic positioning information, and posture positioning information, please refer to the relevant existing technology, and the present invention will not elaborate on this.

[0031] S102, obtaining multi-source positioning information corresponding to each time node of the target person within a preset historical time window, and generating a neighboring multi-source feature vector.

[0032] In some embodiments, in step S102, the method for obtaining the neighboring multi-source feature vectors specifically includes: A1. Within a preset historical time window (such as the past 2 minutes), collect multi-source positioning information at each time node (such as every second), including baseline tag positioning information, geomagnetic positioning information, and posture positioning information.

[0033] A2. Based on the baseline tag positioning information, geomagnetic positioning information, and pose positioning information and their corresponding time nodes respectively, generate a number of data points. Each time node and the longitude (or latitude) in its positioning information form a data point, and perform fitting (such as polynomial fitting, etc.) to obtain a trajectory curve within a preset historical time window. The horizontal axis of the trajectory curve represents the time node, and the vertical axis represents the longitude (or latitude).

[0034] A3. Based on the trajectory curves corresponding to each type of positioning information in the multi-source positioning information, calculate the offset values between the trajectory curves of each type of positioning information and the trajectory curves of the other two types of positioning information respectively, and generate offset pairs.

[0035] Specifically, the offset value is calculated according to the following formula:

[0036] where D is the offset value between the trajectory curves of the two types of positioning information, t1 is the start time node of the historical time window, t2 is the end time node of the historical time window, is the trajectory curve function of the positioning information of type a, that is, its longitude value at the time node t, is the trajectory curve function of the positioning information of type b, that is, its longitude value at the time node t.

[0037] Specifically, the offset pair corresponding to a certain type of positioning information includes: the offset values between the trajectory curve of this type of positioning information and the trajectory curves of the other two types of positioning information respectively. For example, the offset pair of the positioning information of type a is set as ( , ), where, is the offset value between the trajectory curve of the positioning information of type a and the trajectory curve of the positioning information of type b, is the offset value between the trajectory curve of the positioning information of type a and the trajectory curve of the positioning information of type c.

[0038] By comparing the offset values between different types of positioning information, the correlation between them can be understood. If the offset values of two types of positioning information are small, it indicates that the correlation between them is high; if the offset values are large, it indicates that the correlation between them is low. The adjacent multi-source feature vector can also indirectly reflect the accuracy of the positioning information. If the offset values of a certain type of positioning information and the other two types of positioning information are both large, then the accuracy of this type of positioning information may be low.

[0039] It should be noted that the calculation of the offset pair considers the overall offset situation of a certain type of positioning information relative to other types of positioning information within the time window, and the adjacent multi-source feature vector reflects the relative consistency and difference between different types of positioning information that are closest to the current moment.

[0040] A4. Combine the offset pairs corresponding to the baseline label positioning information, geomagnetic positioning information, and pose positioning information respectively to form an adjacent multi-source feature vector, denoted as [( , ), ( , ), ( , )].

[0041] Among them, ( , ) is the offset pair of the baseline label positioning information, ( , ) is the offset pair of the geomagnetic positioning information, and ( , ) is the offset pair of the pose positioning information.

[0042] S103. Input the adjacent multi-source feature vector into the pre-trained decision contribution prediction model, and output the decision contribution factor sequence corresponding to the multi-source positioning information. The decision contribution factors corresponding to each type of positioning information in the multi-source positioning information form the decision contribution factor sequence.

[0043] In some embodiments, the acquisition method of the pre-trained decision contribution prediction model includes: B1. Collect a large number of adjacent multi-source feature vectors of personnel in the mine at historical moments (select historical moments when the actual positioning information is determined to be standard (i.e., the positioning error is small and will not have an impact). Specifically, the training samples can be selected based on expert experience), and label each adjacent multi-source feature vector. The labeling content is set as: the decision contribution factor sequence.

[0044] Among them, the specific labeling method includes: A sequence composed of the ratio of the distance values between the actual positioning information corresponding to the historical moment and the corresponding baseline label positioning information, geomagnetic positioning information, and pose positioning information respectively (for example, the ratio of the distance between the actual positioning information and the baseline label positioning information to the total distance) is used as the decision contribution factor sequence of the adjacent multi-source feature vector.

[0045] B2. Use all the labeled historical adjacent multi-source feature vectors as the training data set, and use the training data set to train the pre-selected neural network structure, optimize the model parameters, and obtain the final decision contribution prediction model.

[0046] With the arrival of new observation data, the adjacent multi-source feature vectors within the historical time window can be continuously updated, which enables the model to dynamically adjust its prediction and decision-making strategies to adapt to the changing environment and requirements.

[0047] S104, based on multi-source positioning information and its decision contribution factor sequence , , , input it into a pre-set multi-source offset fusion algorithm model, and output the target positioning information of the target person.

[0048] In some embodiments, the pre-set multi-source offset fusion algorithm is:

[0049] Among them, is the target positioning information of the target person, are the decision contribution factors of the baseline tag positioning information, geomagnetic positioning information, and pose positioning information respectively, is the distance value between the baseline tag positioning information and the target positioning information, is the distance value between the geomagnetic positioning information and the target positioning information, is the distance value between the pose positioning information and the target positioning information. Thus, the multi-source offset fusion algorithm obtains the target positioning information by minimizing the weighted distance sum, and the decision contribution factor is applied as the weight value, reflecting the reliability and accuracy of various types of positioning information at the current moment.

[0050] In summary, the adjacent multi-source feature vectors within the historical time window contain rich corresponding offsets in time series. These features play a positive guiding role in the final prediction result. The adjacent multi-source feature vectors quantify the consistency and difference of these positioning information within the historical time window by calculating the offset values between different types of positioning information (baseline tag positioning, geomagnetic positioning, pose positioning). This consistency and difference are important indicators for evaluating the reliability of various types of positioning information, which helps the model make more accurate judgments when fusing multiple positioning information; in the multi-source offset fusion algorithm model, the adjacent multi-source feature vectors serve as the input features of the decision contribution model, providing a basis for the multi-source offset fusion algorithm model to fuse different positioning information, adjusting the weights of different positioning information, and thus obtaining a more accurate fused positioning result. By calculating the adjacent multi-source feature vectors in real time and inputting them into the model, it can support the update and correction of real-time positioning information, which helps to maintain the accuracy and reliability of positioning information in a dynamic environment, helps to reduce the impact of outliers or sudden interferences on the positioning result, and improves the robustness of positioning.

[0051] A neighboring multi-source feature vector and a decision contribution prediction model are introduced to evaluate the contribution degree of the positioning information from each source. A multi-source offset fusion algorithm is adopted, and the decision contribution factor is combined to calculate the target positioning information, improving the accuracy and robustness of the positioning. By considering the relative consistency and difference of the multi-source positioning information, the influence of outliers on the positioning result is reduced. The current possible outliers are down-weighted by the decision contribution factor to solve the problem that the fusion calculation process may be affected by fluctuations when there are outliers in the multi-positioning fusion result, realizing a more accurate and reliable underground personnel positioning.

[0052] Embodiment 2 In some embodiments, after step S103, the method further includes: S201, obtaining the image information of the underground mine collected by the roadway monitoring device closest to the main baseline tag at the current moment.

[0053] In the mine roadway, monitoring devices (such as cameras) are arranged at key positions to monitor the activities in the underground mine. The monitoring device closest to the main baseline tag is selected to obtain the image information of the underground mine at the current moment. The main baseline tag is determined in step S101 before and is used to identify a specific position in the roadway. By obtaining the image of the closest monitoring device, it can be ensured that the image information is closely related to the position of the target person.

[0054] S202, inputting the image information of the underground mine, the shooting parameters of the monitoring device, and the coordinate information of the main baseline tag into the pre-trained mine target recognition model to obtain the visual positioning information of the target person.

[0055] Specifically, the mine target recognition model includes a pixel classification model, a visual positioning prediction module, and a target person determination module. The image information of the underground mine is input into the pixel classification model to output a mine classification map; the mine classification map and the shooting parameters are input into the visual positioning prediction module to output all target objects and their visual positioning information; all target objects and their visual positioning information, and the coordinate information of the main baseline tag are input into the target person determination module to output the visual positioning information of the target person.

[0056] Among them, the target objects may include iconic objects in the underground mine such as mine cars and personnel, and the shooting parameters include the focal length, optical center, tilt angle, horizontal field of view, vertical field of view, and camera height of the monitoring device camera; Among them, the pixel classification module is used to output a mine classification map with the same resolution as the input mine image information, and the classification value of each pixel represents the target object identifier to which the pixel belongs in the input mine image information. For the training of the pixel classification model, this will not be elaborated here and reference can be made to the existing technology. A training data set is obtained by annotating a large number of historical mine image information, and the pre-set neural network structure is trained using the training data set to obtain the final pixel classification model.

[0057] Specifically, the visual positioning prediction module is specifically used for: C1. Based on all pixel points in the mine classification map and their annotated target object identifiers, pixel points are merged according to the target object identifiers (pixel points with the same target object identifier and a dispersion degree less than the dispersion threshold are merged) to obtain at least one target object class, each target object class corresponds to a target object identifier, and the original mine classification map is updated. The latest mine classification map includes at least one target object class.

[0058] C2. Convert the pixel points within the target object class from the image coordinate system to the visual positioning information in the same coordinate system (i.e., the scene coordinate system) as the position coordinates corresponding to the main baseline label, specifically including: using camera calibration technology to obtain the internal parameters (such as focal length, optical center, etc.) and external parameters (such as rotation matrix, translation vector) of the camera, and applying the geometric transformation formula to convert the image coordinates to the actual scene coordinates: Through the image scale and shooting parameters, the positioning information of each sub-pixel point is converted from the image coordinate system to the scene coordinate system. Among them, the coordinate information of the pixel points of the target object class in the image coordinate system is determined according to the camera parameters (focal length, optical center position, and distortion coefficient), and the image scale is preset according to the camera parameters and the scene coordinate system, which is used to convert the pixel unit in the image to the physical unit in the real world. The determination of the image scale usually depends on the camera calibration process, which usually involves the use of camera internal parameters, and this invention will not elaborate on this; for each pixel point within the target object class, its image coordinates and the internal and external parameters of the camera are used for conversion. The conversion formula usually involves converting the image coordinates to the three-dimensional coordinates in the camera coordinate system, and then converting to the scene coordinate system through the external parameters. The specific conversion formula may vary depending on the camera model and calibration method, and this invention will not elaborate on this and reference can be made to the existing related technology.

[0059] It should be noted that the main baseline label coordinate information provides a known reference point in the scene. The main baseline label coordinate information and the monitoring device coordinate information (i.e., the position information of the camera) can be used to further correct and verify the conversion result. By comparing the relative position relationship between the converted coordinates and the main baseline label coordinates, the parameters in the conversion formula can be adjusted to improve the accuracy of the conversion.

[0060] Specifically, the target person determination module is specifically used for: Filter out the visual positioning information with the target object identified as a person, compare it with the multi-source positioning information and the main baseline label coordinate information, and use the visual positioning information that meets the preset conditions as the visual positioning information of the target person.

[0061] Among them, the preset conditions are set as follows: on the premise that the distance value between the visual positioning information and all the baseline label coordinate information received by the target person is the smallest for the main baseline label coordinate information, the distance value between the visual positioning information and the multi-source positioning information is the smallest (the distance values with each type of positioning information can be calculated respectively, and then weighted and averaged based on the corresponding decision contribution factors). It should be noted that all the baseline label coordinate information appears in pairs, and the distance value between the visual positioning information and all the baseline label coordinate information received by the target person is set as the sum of the distance values between each coordinate information in each group of baseline label coordinate information and the visual positioning information.

[0062] S203. Based on the relationship between the visual positioning information of the target person and the multi-source positioning information, correct the decision contribution factor sequence to obtain an updated decision contribution factor sequence to replace the original decision contribution factor sequence.

[0063] Among them, correcting the decision contribution factor sequence specifically includes: D1. Calculate the correction factor of each decision contribution factor according to the following formula:

[0064]

[0065] Among them, i and j are used to represent the type numbers of the multi-source positioning information, including a, b, c, is the distance value between the visual positioning information and the i-th type of multi-source positioning information, is the reliability value of the i-th type of multi-source positioning information, is the original decision contribution factor of the i-th type of multi-source positioning information, is the distance value between the visual positioning information and the j-th type of multi-source positioning information, is the reliability value of the j-th type of multi-source positioning information, is the Gaussian function, which is used to attenuate the consistency of the positioning information according to the distance, is a preset positive parameter, which is used to control the attenuation speed.

[0066] D2. Use the correction factor of each decision contribution factor as the new decision contribution factor to generate a new decision contribution factor sequence.

[0067] In summary, by introducing visual positioning information and combining it with multi-source positioning information for correction, the visual positioning information provides an intuitive image basis, which helps to reduce the errors of other positioning information and makes the positioning result more accurate. Dynamically adjusting the decision contribution factors of different types of positioning information from various sources reduces the impact of outliers on the positioning result, compensates for the possible error defects of the model, and improves the robustness of positioning. By obtaining image information and calculating correction factors in real time, it is possible to support the update and correction of real-time positioning information and adapt to the complex and changeable mine environment.

[0068] Embodiment 3 Figure 2 It is a schematic structural diagram of a personnel positioning system in a mine according to an embodiment of the present invention.

[0069] As Figure 2 shown, a personnel positioning system in a mine includes: an acquisition module, a decision module, and a fusion positioning module; Specifically, the acquisition module is used to determine any person in the current mine as the target person and obtain the multi-source positioning information of the target person. The multi-source positioning information includes three types of positioning information: baseline tag positioning information, geomagnetic positioning information, and pose positioning information, and the types of positioning information sources are respectively denoted as a, b, and c; obtain the multi-source positioning information corresponding to each time node of the target person within a preset historical time window, and generate an adjacent multi-source feature vector; The decision module is used to input the adjacent multi-source feature vector into a pre-trained decision contribution prediction model, and output a decision contribution factor sequence corresponding to the multi-source positioning information. The decision contribution factors corresponding to each type of positioning information in the multi-source positioning information form a decision contribution factor sequence; The fusion positioning module is used to input the multi-source positioning information and its decision contribution factor sequence into a preset multi-source offset fusion algorithm model, and output the target positioning information of the target person.

[0070] It should be noted that in the above embodiments, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0071] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] The present invention will be described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0073] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0075] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention.

[0076] Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for locating personnel in a mine, characterized in that: include: S101, any person in the current mine is determined as a target person, and multi-source positioning information of the target person is obtained. The multi-source positioning information includes positioning information from three sources: baseline tag positioning information, geomagnetic positioning information, and posture positioning information. The types of positioning information sources are respectively recorded as a, b, and c; S102, obtaining multi-source positioning information corresponding to each time node of the target person within a preset historical time window, and generating a neighboring multi-source feature vector; S103, inputting the neighboring multi-source feature vector into a pre-trained decision contribution prediction model, outputting a decision contribution factor sequence corresponding to the multi-source positioning information, wherein the decision contribution factor corresponding to each positioning information in the multi-source positioning information constitutes the decision contribution factor sequence; S104, based on the multi-source positioning information and its decision contribution factor sequence, it is input into a pre-set multi-source offset fusion algorithm model to output the target positioning information of the target person.

2. The method for locating personnel in a mine according to claim 1, characterized in that: The method for acquiring the baseline tag positioning information is as follows: acquiring tag signals of multiple groups of baseline tags in the area, including tag coordinates and tag signal emission angles, selecting a group of baseline tags with the largest tag signal emission angle as the main baseline tags, and based on the main baseline tags, selecting a group of adjacent baseline tags as secondary baseline tags, and using the coordinate information of the main baseline tags and the secondary baseline tags, as well as the tag signal emission angle, to obtain the baseline tag positioning information of the target person.

3. The method for locating personnel in a mine according to claim 1, characterized in that: The method for obtaining the adjacent multi-source feature vector specifically includes: A1. Collect multi-source positioning information at each time node within the preset historical time window, including baseline label positioning information, geomagnetic positioning information and posture positioning information; A2. Generate several data points based on the baseline tag positioning information, geomagnetic positioning information, posture positioning information and their corresponding time nodes. Each time node and the longitude in its positioning information constitute a data point. Perform fitting to obtain a trajectory curve within a preset historical time window. The horizontal axis of the trajectory curve represents the time node and the vertical axis represents the longitude. A3. Based on the trajectory curve corresponding to each positioning information in the multi-source positioning information, respectively calculate the offset values ​​between the trajectory curves of the other two positioning information to generate an offset pair; A4. The offset pairs corresponding to the baseline label positioning information, the geomagnetic positioning information and the posture positioning information are respectively formed into a neighboring multi-source feature vector.

4. The method for locating personnel in a mine according to claim 3, characterized in that: The offset value is calculated according to the following formula: Where D is the offset value between the trajectory curves of the two positioning information, t1 is the starting time node of the historical time window, and t2 is the ending time node of the historical time window. is the longitude value of the trajectory curve with positioning information type a at time node t, It is the longitude value of the trajectory curve with positioning information type b at time node t.

5. The method for locating personnel in a mine according to claim 3, characterized in that: The neighboring multi-source feature vector is recorded as [( , ),( , ),( , )],in,( , ) is the offset pair of the baseline label positioning information, ( , ) is the offset pair of geomagnetic positioning information, ( , ) is the offset pair of pose positioning information, is the offset value between the trajectory curve of type a positioning information and the trajectory curve of type b positioning information, It is the offset value between the trajectory curve of type a positioning information and the trajectory curve of type c positioning information.

6. The method for locating personnel in a mine according to claim 1, characterized in that: The method for obtaining the pre-trained decision contribution prediction model includes: B1. Collect a large number of neighboring multi-source feature vectors of underground personnel at historical moments, and label each neighboring multi-source feature vector, where the labeling content is set as: decision contribution factor sequence; wherein the labeling method specifically includes: a sequence composed of the ratio of the distance values ​​between the actual positioning information at the corresponding historical moment and the corresponding baseline label positioning information, geomagnetic positioning information, and posture positioning information, as the decision contribution factor sequence of the neighboring multi-source feature vector; B2. All the annotated historical neighboring multi-source feature vectors are used as training data sets, and the pre-selected neural network structure is trained using the training data sets to optimize the model parameters and obtain the final decision contribution prediction model.

7. The method for locating personnel in a mine according to claim 1, characterized in that: The pre-set multi-source offset fusion algorithm is: in, Target positioning information for the target person, They are the decision contribution factors of baseline tag positioning information, geomagnetic positioning information, and posture positioning information, respectively. is the distance value between the baseline label positioning information and the target positioning information, is the distance between the geomagnetic positioning information and the target positioning information, It is the distance value between the pose positioning information and the target positioning information.

8. The method for locating personnel in a mine according to claim 2, characterized in that: After S103, the method further includes: S201, obtaining underground mine image information collected by the tunnel monitoring equipment closest to the main baseline tag at the current moment; S202, inputting the underground mine image information, the shooting parameters of the monitoring equipment, and the main baseline label coordinate information into the pre-trained mine target recognition model to obtain the visual positioning information of the target personnel; wherein the mine target recognition model includes a pixel classification model, a visual positioning prediction module and a target personnel determination module, inputting the underground mine image information into the pixel classification model, and outputting the mine classification map; inputting the mine classification map and the shooting parameters into the visual positioning prediction module, and outputting all target objects and their visual positioning information; inputting all target objects and their visual positioning information, and the main baseline label coordinate information into the target personnel determination module, and outputting the visual positioning information of the target personnel; S203, based on the relationship between the visual positioning information of the target person and the multi-source positioning information, the decision contribution factor sequence is modified to obtain an updated decision contribution factor sequence to replace the original decision contribution factor sequence.

9. The method for locating personnel in a mine according to claim 8, characterized in that: In S203, the decision contribution factor sequence is modified, specifically including: D1. Calculate the correction factor of each decision contribution factor according to the following formula: Among them, i and j are used to indicate the type number of information given by multi-source positioning, including a, b, c, is the distance value between the visual positioning information and the i-th multi-source positioning information, is the reliability value of the i-th multi-source positioning information, is the original decision contribution factor of the i-th multi-source positioning information, is the distance value between the visual positioning information and the j-th multi-source positioning information, is the reliability value of the j-th multi-source positioning information, It is a preset positive parameter used to control the decay speed; D2. Use the correction factor of each decision contribution factor as a new decision contribution factor to generate a new decision contribution factor sequence.

10. A personnel positioning system in a mine, applied to the personnel positioning method in a mine as claimed in any one of claims 1 to 9, characterized in that: include: Acquisition module, decision module and fusion positioning module; The acquisition module is used to identify any person in the current mine as a target person, and obtain the multi-source positioning information of the target person. The multi-source positioning information includes positioning information from three sources: baseline label positioning information, geomagnetic positioning information, and posture positioning information. The types of positioning information sources are recorded as a, b, and c respectively; the multi-source positioning information corresponding to each time node of the target person in the preset historical time window is obtained to generate a neighboring multi-source feature vector; The decision module is used to input the neighboring multi-source feature vectors into the pre-trained decision contribution prediction model, and output the decision contribution factor sequence corresponding to the multi-source positioning information. The decision contribution factors corresponding to each positioning information in the multi-source positioning information constitute the decision contribution factor sequence; The fusion positioning module is used to input the multi-source positioning information and its decision contribution factor sequence into a pre-set multi-source offset fusion algorithm model to output the target positioning information of the target person.

Images