Intelligent visual navigation system for transfer robot based on high-precision positioning

Abstract

The invention relates to the technical field of robot navigation, in particular to a transfer robot vision intelligent navigation system based on high-precision positioning, which is characterized in that a visual sensor, a laser radar and an infrared sensor are arranged on a transfer robot; analyzing the data of the visual sensor, outputting a sensor abnormal signal, and switching the visual sensor into a laser radar and infrared sensor combined navigation mode; generating three-dimensional point cloud data through the distance information, performing preprocessing in combination with the temperature data and the heat data to obtain a standard first data set, and performing space matching to obtain a mine-environment model; identifying a fixed obstacle, a dynamic obstacle and an abnormal area in the model, and generating a plurality of candidate passing paths; obtaining a security coefficient of each candidate path, and obtaining an optimal path in combination with historical path information; and monitoring data of the laser radar and the infrared sensor in real time, and when the visual sensor returns to normal and / or the transfer robot arrives at the target area, exiting the blind walking state.

Description

Technical Field

[0001] The present invention relates to the field of robot navigation technology, and in particular to a transport robot visual intelligent navigation system based on high-precision positioning. Background Art

[0002] In the existing technology, the visual navigation system of the handling robot has been widely used in the fields of industrial automation and intelligent manufacturing, but its application in special environments such as underground mines still faces many challenges.

[0003] Due to factors such as poor lighting conditions, complex environment, limited ventilation, and changing terrain inside underground mines, traditional navigation systems that rely on single vision or single positioning technology often find it difficult to achieve high-precision positioning and path planning, making robots susceptible to environmental interference when performing handling tasks, resulting in positioning errors or navigation failures. At the same time, existing high-precision positioning technologies, such as positioning methods based on satellite signals, cannot work properly in closed underground environments due to signal shielding, forcing existing systems to use other auxiliary means, but these methods generally have problems such as insufficient robustness and poor real-time performance.

[0004] Therefore, a visual intelligent navigation system for a handling robot based on high-precision positioning is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a visual intelligent navigation system for a handling robot based on high-precision positioning. The present invention arranges a visual sensor, a laser radar and an infrared sensor on the handling robot; analyzes the visual sensor data to output a sensor abnormality signal, and switches the visual sensor to a joint navigation mode of the laser radar and infrared sensor; generates three-dimensional point cloud data through distance information, and pre-processes the temperature data and heat data to obtain a standard first data set, performs spatial matching, and obtains a mine-environment model; identifies fixed obstacles, dynamic obstacles and abnormal areas in the model, and generates multiple candidate passage paths; obtains the safety factor of each candidate path, and obtains the optimal path in combination with historical path information; monitors the laser radar and infrared sensor data in real time, and exits the blind walking state when the visual sensor returns to normal and / or the handling robot reaches the target area.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A visual intelligent navigation system for a handling robot based on high-precision positioning, comprising:

[0008] The data acquisition module acquires underground mine maps and historical path information; and sets visual sensors, laser radars, and infrared sensors on the handling robot;

[0009] The status monitoring module analyzes the data collected by the visual sensor in real time and outputs sensor abnormality signals;

[0010] The switching control module switches the visual sensor to the joint navigation mode of the laser radar and infrared sensor according to the abnormal signal of the sensor;

[0011] The blind walking path planning module first obtains distance information, temperature data and heat data;

[0012] S10. Generate three-dimensional point cloud data through distance information, and pre-process the three-dimensional point cloud data, temperature data and heat data to obtain a standard first data set;

[0013] S20. Obtaining a mine-environment model by spatially matching the data in the standard first data set; identifying fixed obstacles, dynamic obstacles and abnormal areas in the mine-environment model in combination with the underground mine map;

[0014] S30. Based on fixed obstacles, locate the transport robot and generate multiple candidate paths;

[0015] S40. Perform a safety assessment on each candidate path to obtain a safety factor of the candidate path; and obtain the optimal path in combination with historical path information;

[0016] S50. Enter the blind walking state according to the optimal path; monitor the laser radar and infrared sensor data in real time, update the candidate path when deviating from the predetermined path and / or new risks arise, and re-evaluate the safety of the candidate path;

[0017] S60. When the visual sensor returns to normal and / or the transport robot reaches the target area, the blind walking state is exited.

[0018] Preferably, the underground mine map includes fixed obstacles, mine structure and safe passage information;

[0019] The visual sensor is used to collect image information of the internal environment of the mine;

[0020] The distance information is collected by the laser radar, and the temperature data and heat data are collected by the infrared sensor.

[0021] Preferably, the specific steps of acquiring the abnormal signal of the sensor are:

[0022] The data collected by the visual sensor in real time is image data collected continuously at a predetermined frame rate in the mine; each frame of the image data is accompanied by a timestamp and collection location information;

[0023] Preprocessing the image data to obtain standard image data;

[0024] Performing image quality and state analysis on the standard image data to obtain an image data set; the image data set includes image brightness data, image contrast data, image clarity data and image occlusion detection data;

[0025] An image quality index is obtained by comprehensively measuring the image data set and compared with a preset threshold; when the image quality index is inconsistent with the preset threshold range, a sensor abnormality signal is obtained.

[0026] Preferably, the calculation formula of the image quality index is:

[0027] IQI=ω 1 Q b +ω 2 Q c +ω 3 Q f +ω 4 (1-Q o );

[0028] Where IQI is the image quality index, ω 1 is the image brightness quality score weight, Q b is the image brightness quality score, ω 2 is the image contrast quality score weight, Q c is the image contrast quality score, ω 3 is the image clarity quality score weight, Q f is the image clarity quality score, ω 4 is the image occlusion ratio weight, Q o is the image occlusion ratio.

[0029] Preferably, the specific steps of obtaining the mine-environment model are:

[0030] S201. Preliminary registration of the three-dimensional point cloud data in the standard first data set with the pre-stored underground mine map, using the RANSAC algorithm based on feature extraction to achieve coarse matching and determine the initial relative position;

[0031] S202. Use an iterative closest point algorithm to perform fine matching on the data after preliminary registration, and adjust the position and posture between the point cloud data and the pre-stored map;

[0032] S203. By comparing the difference between the matching result and the pre-stored map, the difference area is divided using the region growing and cluster analysis method, where:

[0033] S2031. A fixed obstacle is an obstacle that always exists during multiple consecutive matching processes and is consistent with the corresponding position in the pre-stored map;

[0034] S2032. A dynamic obstacle is an obstacle that is significantly displaced from the fixed obstacle position in the pre-stored map and / or only appears locally in the current three-dimensional point cloud data;

[0035] S2033. The abnormal area is an area where the local light, temperature or heat is abnormal;

[0036] S204. Storing the information of the fixed obstacles, the dynamic obstacles and the abnormal areas in a structured form to form a complete mine-environment model.

[0037] Preferably, generating multiple candidate travel paths includes the following steps:

[0038] S301. Discretizing the fixed obstacles and channel information in the mine-environment model into a navigation map, wherein the nodes of the navigation map correspond to channel intersections, endpoints and key turning points, and the edges correspond to walkable channels between the nodes;

[0039] S302. According to the distance between nodes, channel width and safety distance from fixed obstacles, weights are assigned to each edge in the navigation map, and the weights of the edges are calculated by comprehensively calculating the path length, obstacle risk and turning angle;

[0040] S303. Using a graph search algorithm in the navigation map, searching from the transport robot positioning point to the predetermined target point, and generating multiple candidate travel paths.

[0041] Preferably, the specific calculation formula of the safety factor is:

[0042] S i =(d i α *r i β *s i γ *m i δ ) 1 / (α+β+γ+δ) ;

[0043] Among them, S i is the safety factor of the i-th candidate path, α is the obstacle spacing weight, d i is the obstacle spacing value of the i-th candidate path, β is the thermal risk value weight, r i is the thermal risk value of the i-th candidate path, γ is the path smoothness value weight, s i is the smoothness value of the i-th candidate path, δ is the weight of the historical path matching value, m i is the historical path matching value of the i-th candidate path.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention adopts multi-sensor data fusion, dynamic switching control, and pre-stored underground mine maps and historical path information, which significantly improves the positioning accuracy and navigation robustness of the robot in complex mine environments. When the visual sensor fails, it quickly switches to the joint navigation mode of the laser radar and infrared sensor, and builds a fine mine-environment model through real-time preprocessing and spatial matching, thereby achieving high-precision positioning and low collision risk, ensuring that the robot can operate continuously and safely in complex environments, and further improving the handling efficiency and operation reliability.

[0046] 2. The present invention generates sensor abnormality signals by preprocessing and quality analyzing visual data in real time, and then dynamically switches the control module. When an abnormal image quality index is detected, the navigation mode can be quickly switched from the visual sensor to the laser radar and infrared sensor combined mode to achieve short-term blind walking. This technology not only improves safety, but also can adjust the navigation strategy in time when the environment changes suddenly, ensuring navigation continuity and path safety.

[0047] 3. The present invention constructs a mine-environment model, uses fixed obstacles, channels and historical path data to generate multiple candidate paths, and combines risk assessment algorithms to evaluate the safety of the paths. In the blind walking state, the environment changes are monitored in real time, and the candidate paths are automatically updated according to deviations from the predetermined path or new risks to ensure that the robot always travels along the optimal safe channel. Not only does it improve the navigation efficiency of the robot when vision fails, but also through real-time path updates, it greatly reduces the risk of accidents caused by sudden changes in the environment. It is suitable for safe operation requirements in complex scenarios such as underground mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the structure of a visual intelligent navigation system for a handling robot based on high-precision positioning provided by the present invention;

[0049] Figure 2 A schematic diagram of the optimal path acquisition process provided by an embodiment of the present invention;

[0050] Figure 3 This is a logic diagram of the short-term blind walking of the transport robot provided in an embodiment of the present invention.

[0051] In the figure: S, starting point; G, target point; X1, first obstacle; X2, second obstacle; X3, third obstacle. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Embodiment 1

[0054] See also Figure 1 The present invention provides a visual intelligent navigation system for a handling robot based on high-precision positioning, and the technical solution is as follows:

[0055] The data acquisition module acquires underground mine maps and historical path information; and sets visual sensors, laser radars, and infrared sensors on the handling robot;

[0056] Further, the underground mine map includes fixed obstacles, mine structure and safe passage information;

[0057] The visual sensor is used to collect image information of the internal environment of the mine, the laser radar is used to collect distance information of the surrounding environment, and the infrared sensor is used to collect ambient temperature and heat information.

[0058] In this embodiment, by obtaining an underground mine map containing fixed obstacles, mine structure and safe passage information, a stable and accurate environmental prior is provided to the robot, thereby enhancing the accuracy and safety of positioning and path planning.

[0059] The status monitoring module analyzes the data collected by the visual sensor in real time and outputs sensor abnormality signals;

[0060] Furthermore, the specific steps of obtaining the abnormal signal of the sensor are:

[0061] The data collected by the visual sensor in real time is image data collected continuously in the mine at a predetermined frame rate; each frame of the image data is accompanied by a timestamp and collection location information;

[0062] Preprocessing the image data to obtain standard image data;

[0063] Performing image quality and state analysis on the standard image data to obtain an image data set; the image data set includes image brightness data, image contrast data, image clarity data and image occlusion detection data;

[0064] An image quality index is obtained by comprehensively measuring the image data set and compared with a preset threshold; when the image quality index is inconsistent with the preset threshold range, a sensor abnormality signal is obtained.

[0065] In this embodiment, by preprocessing and quality analysis of visual data, a sensor abnormality signal is generated in real time to ensure rapid switching of the navigation mode when the image is abnormal, thereby effectively avoiding positioning errors caused by visual failure.

[0066] Furthermore, the calculation formula of the image quality index is:

[0067] IQI=ω 1 Q b +ω 2 Q c +ω 3 Q f +ω 4 (1-Q o );

[0068] Where IQI is the image quality index, ω 1 is the image brightness quality score weight, Q b is the image brightness quality score, ω 2 is the image contrast quality score weight, Q c is the image contrast quality score, ω 3 is the image clarity quality score weight, Q f is the image clarity quality score, ω 4 is the image occlusion ratio weight, Q o is the image occlusion ratio.

[0069] The calculated IQI is compared with a preset total quality threshold. If the IQI is less than the threshold, it is considered that the image quality does not meet the requirement, thereby generating a sensor abnormality signal. In this embodiment, the preset total quality threshold is set to 0.8.

[0070] Image brightness quality score Q b It is obtained by the following method: convert the collected image data into a grayscale image and calculate the average grayscale value L of all pixels avg ; According to the preset brightness lower limit B min and upper limit B max , get the image brightness quality score The preset brightness lower limit B min and upper limit B max Determined by statistical analysis of image data under historical working conditions;

[0071] Image contrast quality score Q c It is obtained by the following method: Calculate the standard deviation σ of the image grayscale as a direct indicator of image contrast, and then obtain Among them C max It is the preset ideal maximum contrast ratio, which is statistically determined by collecting image data under normal working conditions;

[0072] Image clarity quality score Q f It is obtained by applying the Laplacian operator to the image to extract the edge information of the image and calculating the variance V of the gradient image. lap ,get Among them, F max It is the preset ideal clarity index value, which is determined by measuring the Laplace variance of the clear image under experimental conditions;

[0073] Image occlusion ratio Q o It is obtained by the following method: using Otsu adaptive threshold segmentation algorithm to segment the image into foreground and background, and using morphological processing to extract continuous occlusion areas; counting the area A of the occlusion area o With the total image area A t The ratio of

[0074] In this embodiment, the present invention adopts brightness, contrast, clarity and occlusion ratio to weightedly calculate the image quality index, accurately evaluate the visual state, detect anomalies in real time, and quickly switch the sensor mode, effectively ensuring the safe, stable and efficient operation of the navigation system.

[0075] The switching control module switches the visual sensor to the joint navigation mode of the laser radar and infrared sensor according to the abnormal signal of the sensor; the multi-sensor data fusion, dynamic switching control and pre-stored underground mine map and historical path information are adopted to significantly improve the positioning accuracy and navigation robustness of the robot in the complex mine environment. The data is shown in Table 1.

[0076] Table 1 Dynamic sensor switching control data table

[0077] Image Quality Index Switching Response Time (ms) Number of Abnormal Triggers 0.65 80 5 0.60 85 6 0.68 78 4

[0078] The blind walking path planning module first obtains distance information, temperature data and heat data;

[0079] S10. Generate three-dimensional point cloud data through distance information, and pre-process the three-dimensional point cloud data, temperature data and heat data to obtain a standard first data set;

[0080] S20. Obtaining a mine-environment model by spatially matching the data in the standard first data set; identifying fixed obstacles, dynamic obstacles and abnormal areas in the mine-environment model in combination with the underground mine map;

[0081] Furthermore, the specific steps of obtaining the mine-environment model are as follows:

[0082] S201. Preliminary registration of the three-dimensional point cloud data in the standard first data set with the pre-stored underground mine map, using the RANSAC algorithm based on feature extraction to achieve coarse matching and determine the initial relative position;

[0083] S202. Use the iterative closest point algorithm to perform fine matching on the data after preliminary registration, adjust the relative position and posture between the point cloud data and the pre-stored map, and use parameters such as the maximum number of iterations of 50 times and the convergence tolerance of 0.001m;

[0084] S203. By comparing the difference between the matching result and the pre-stored map, the difference area is divided using the region growing and cluster analysis method, where:

[0085] S2031. A fixed obstacle is an obstacle that always exists during multiple consecutive matching processes and is consistent with the corresponding position in the pre-stored map;

[0086] S2032. A dynamic obstacle is an obstacle that is significantly displaced from the fixed obstacle position in the pre-stored map or that only appears locally in the current three-dimensional point cloud data;

[0087] S2033. Abnormal areas are areas that deviate significantly from the pre-stored map due to temporary environmental factors (such as local illumination, temperature or heat anomalies);

[0088] S204. Storing the information of the fixed obstacles, the dynamic obstacles and the abnormal areas in a structured form to form a complete mine-environment model.

[0089] In the above steps, the coarse matching adopts the RANSAC algorithm to improve the matching robustness, and the fine matching adopts the iterative closest point algorithm to achieve high-precision alignment. Each parameter (such as the maximum number of iterations, convergence tolerance, etc.) is determined by statistical experimental data in the underground mine environment, thereby ensuring that the mine-environment model can accurately reflect the distribution of fixed obstacles, dynamic obstacles and abnormal areas in the mine.

[0090] S30. Based on fixed obstacles, locate the transport robot and generate multiple candidate paths;

[0091] Furthermore, generating multiple candidate travel paths includes the following steps:

[0092] S301. Discretizing the fixed obstacles and channel information in the mine-environment model into a navigation map, wherein the nodes of the navigation map correspond to channel intersections, endpoints and key turning points, and the edges correspond to walkable channels between the nodes;

[0093] S302. Weights are assigned to each edge in the navigation graph based on the distance between nodes, channel width, and safety distance from fixed obstacles. The weights of the edges are calculated by comprehensively calculating the path length, obstacle proximity risk, and turning angle;

[0094] S303. Using a graph search algorithm in the navigation map, searching from the transport robot positioning point to the predetermined target point, and generating multiple candidate travel paths.

[0095] A navigation map is constructed based on fixed obstacles and a graph search algorithm is used to generate candidate paths, effectively integrating obstacle information and safe channel data, providing the robot with multiple safe passage path options and improving navigation efficiency.

[0096] S40. Perform a safety assessment on each candidate path to obtain a safety factor of the candidate path; and obtain the optimal path in combination with historical path information;

[0097] Furthermore, the specific calculation formula of the safety factor is:

[0098]

[0099] Among them, S i is the safety factor of the i-th candidate path, α is the obstacle spacing weight, d i is the obstacle spacing value of the i-th candidate path, β is the thermal risk value weight, r i is the thermal risk value of the i-th candidate path, γ is the path smoothness value weight, s i is the smoothness value of the i-th candidate path, δ is the weight of the historical path matching value, m i is the historical path matching value of the i-th candidate path, d meas is the average distance between the key node of the i-th candidate path and the nearest fixed obstacle, d ref Preset the upper limit of the safe distance for the i-th candidate path, r meas is the thermal risk value along the candidate path obtained by the infrared sensor for the i-th candidate path (the higher the thermal risk value, the greater the risk). The thermal risk value is mainly obtained through the temperature data collected by the infrared sensor. ref is the maximum thermal risk reference value preset for the i-th candidate path, obtained through a large amount of data statistics, s meas is the smoothness index of the i-th candidate path, s ref is the reference value of the ideal smooth path of the ith candidate path. The historical path matching value of the ith candidate path is obtained through correlation analysis. min() is the minimum function.

[0100] The safety factor of the candidate path comprehensively considers obstacle spacing, thermal risk, path smoothness and historical matching, evaluates path safety through a reasonable formula, provides a quantitative basis for path selection, and ensures driving safety. At the same time, the safety factor formula normalizes each key indicator and performs a nonlinear combination to objectively reflect the comprehensive risk and safety of the candidate path, which helps to automatically select the optimal and safest path.

[0101] S50. Enter the blind walking state according to the optimal path; monitor the laser radar and infrared sensor data in real time, update the candidate path when deviating from the predetermined path and / or new risks arise, and re-evaluate the safety of the candidate path;

[0102] In this embodiment, combined with Figure 2 To further illustrate the optimal path acquisition process, according to the mine-environment model, the fixed obstacles and channel information are discretized into a navigation graph. The nodes of the navigation graph correspond to channel intersections, endpoints, and key turning points, and the edges represent the traversable paths between nodes. Figure 2 For a simplified 5*5 grid diagram, Figure 2 Where S is the starting point when the visual sensor fails, located at (1,1), G is the target point, located at (5,5), X1 is the first obstacle, X2 is the second obstacle, and X3 is the third obstacle; the first obstacle is a fixed obstacle, the second obstacle is a dynamic obstacle, and the third obstacle is an abnormal area; the following three candidate paths can be obtained:

[0103] Path A: (1,1)→(1,2)→(2,3)→(3,3)→(3,4)→(3,5)→(4,5)→(5,5);

[0104] Path B: (1,1)→(2,1)→(3,2)→(3,3)→(4,3)→(5,3)→(5,4)→(5,5);

[0105] Path C: (1,1)→(1,2)→(1,3)→(1,4)→(1,5)→(2,5)→(3,5)→(4,5)→(5,5);

[0106] In the safety assessment stage, the present invention adopts a comprehensive assessment method based on obstacle spacing, thermal risk, path smoothness, and matching degree with historical safe paths to evaluate the above three paths. After normalization, path A has a large average distance from the nearest obstacle at key nodes, ensuring a high obstacle spacing score; at the same time, the infrared sensor data collected along path A shows that the thermal risk is low, and the thermal risk score obtained after normalization is high; in addition, path A has a smooth overall direction, a small turning angle, and a smoothness index that is better than other candidate paths; finally, the path has a high matching degree with the historical safe path, indicating that it has been verified as a safe and stable route in actual operation.

[0107] Taking all the above indicators into consideration, according to the nonlinear safety coefficient calculation formula adopted by the present invention, the safety coefficient of path A is significantly higher than that of other candidate paths, and thus it is selected as the optimal path.

[0108] When the system detects an abnormal image quality index, it can quickly switch the navigation mode from the visual sensor to the combined mode of the laser radar and infrared sensor to achieve "short-term blind walking". This technology not only improves the safety of the system, but also can adjust the navigation strategy in time when the environment changes suddenly, ensuring navigation continuity and path safety. For data, see Table 2.

[0109] Table 2 Blind walking path planning and real-time dynamic update data table

[0110] Number of Candidate Paths Optimal Path Safety Coefficient Path Update Time (ms) Duration of Blind Walking (seconds) 4 0.85 150 30 5 0.80 148 29 6 0.86 145 32

[0111] S60. When the visual sensor returns to normal and / or the transport robot reaches the target area, the blind walking state is exited.

[0112] The present invention realizes high-precision positioning and intelligent navigation in underground mine environments through multi-sensor collaboration. The system integrates visual, laser radar and infrared sensor data collection, constructs mine maps and historical path information in real time, and uses image quality detection to automatically switch to the laser radar and infrared sensor joint navigation mode when vision fails, effectively overcoming positioning errors caused by low light and dust interference. Using the three-dimensional point cloud data generated by the laser radar and the pre-stored map, a fine mine-environment model is constructed through the RANSAC and ICP algorithms to accurately identify fixed obstacles, dynamic obstacles and abnormal areas; on this basis, multiple candidate passage paths are generated, and the path safety is comprehensively evaluated in combination with indicators such as obstacle spacing, thermal risk, path smoothness and historical matching, and the optimal path is selected to enter the blind walking state. During the blind walking process, the system monitors the sensor data in real time, dynamically updates the candidate paths and safety assessments, ensures that the robot operates safely, continuously and efficiently in complex environments, and significantly improves the overall operating efficiency and system robustness.

[0113] Embodiment 2

[0114] The present invention provides a visual intelligent navigation system for a handling robot based on high-precision positioning, including a data acquisition module, a state monitoring module, a switching control module and a blind path planning module. The technical solution is as follows:

[0115] The data acquisition module acquires underground mine maps and historical path information; and sets visual sensors, laser radars, and infrared sensors on the handling robot;

[0116] The status monitoring module analyzes the data collected by the visual sensor in real time and outputs sensor abnormality signals;

[0117] The switching control module switches the visual sensor to the joint navigation mode of the laser radar and infrared sensor according to the abnormal signal of the sensor;

[0118] The blind walking path planning module first obtains distance information, temperature data and heat data;

[0119] S10. Generate three-dimensional point cloud data through distance information, and pre-process the three-dimensional point cloud data, temperature data and heat data to obtain a standard first data set;

[0120] S20. Obtaining a mine-environment model by spatially matching the data in the standard first data set; identifying fixed obstacles, dynamic obstacles and abnormal areas in the mine-environment model in combination with the underground mine map;

[0121] S30. Based on fixed obstacles, locate the transport robot and generate multiple candidate paths;

[0122] S40. Perform a safety assessment on each candidate path to obtain a safety factor of the candidate path; and obtain the optimal path in combination with historical path information;

[0123] S50. Enter the blind walking state according to the optimal path; monitor the laser radar and infrared sensor data in real time, update the candidate path when deviating from the predetermined path and / or new risks arise, and re-evaluate the safety of the candidate path;

[0124] S60. When the visual sensor returns to normal and / or the transport robot reaches the target area, the blind walking state is exited.

[0125] Further, the underground mine map includes fixed obstacles, mine structure and safe passage information;

[0126] The visual sensor is used to collect image information of the internal environment of the mine;

[0127] The distance information is collected by the laser radar, and the temperature data and heat data are collected by the infrared sensor.

[0128] Furthermore, the specific steps of obtaining the abnormal signal of the sensor are:

[0129] The data collected by the visual sensor in real time is image data collected continuously in the mine at a predetermined frame rate; each frame of the image data is accompanied by a timestamp and collection location information;

[0130] Preprocessing the image data to obtain standard image data;

[0131] Performing image quality and state analysis on the standard image data to obtain an image data set; the image data set includes image brightness data, image contrast data, image clarity data and image occlusion detection data;

[0132] An image quality index is obtained by comprehensively measuring the image data set and compared with a preset threshold; when the image quality index is inconsistent with the preset threshold range, a sensor abnormality signal is obtained.

[0133] Furthermore, the calculation formula of the image quality index is:

[0134] IQI=ω 1 Q b +ω 2 Q c +ω 3 Q f +ω 4 (1-Q o );

[0135] Where IQI is the image quality index, ω 1 is the image brightness quality score weight, Q b is the image brightness quality score, ω 2 is the image contrast quality score weight, Q c is the image contrast quality score, ω 3 is the image clarity quality score weight, Q f is the image clarity quality score, ω 4 is the image occlusion ratio weight, Q o is the image occlusion ratio.

[0136] Image brightness quality score Q b It is obtained by the following method: convert the collected image data into a grayscale image and calculate the average grayscale value L of all pixels avg ; According to the preset brightness lower limit B min and upper limit B max , get the image brightness quality score The preset brightness lower limit B min and upper limit B max Determined by statistical analysis of image data under historical working conditions;

[0137] Image contrast quality score Q c It is obtained by the following method: Calculate the standard deviation σ of the image grayscale as a direct indicator of image contrast, and then obtain Among them C max It is the preset ideal maximum contrast ratio, which is statistically determined by collecting image data under normal working conditions;

[0138] Image clarity quality score Q f It is obtained by applying the Laplacian operator to the image to extract the edge information of the image and calculating the variance V of the gradient image. lap ,get where F max It is the preset ideal clarity index value, which is determined by measuring the Laplace variance of the clear image under experimental conditions;

[0139] Image occlusion ratio Q o It is obtained by the following method: using Otsu adaptive threshold segmentation algorithm to segment the image into foreground and background, and using morphological processing to extract continuous occlusion areas; counting the area A of the occlusion area o With the total image area A t The ratio of

[0140] Furthermore, the specific steps of obtaining the mine-environment model are as follows:

[0141] S201. Preliminary registration of the three-dimensional point cloud data in the standard first data set with the pre-stored underground mine map, using the RANSAC algorithm based on feature extraction to achieve coarse matching and determine the initial relative position;

[0142] S202. Use an iterative closest point algorithm to perform fine matching on the data after preliminary registration, and adjust the position and posture between the point cloud data and the pre-stored map;

[0143] S203. By comparing the difference between the matching result and the pre-stored map, the difference area is divided using the region growing and cluster analysis method, where:

[0144] S2031. A fixed obstacle is an obstacle that always exists during multiple consecutive matching processes and is consistent with the corresponding position in the pre-stored map;

[0145] S2032. A dynamic obstacle is an obstacle that is significantly displaced from the fixed obstacle position in the pre-stored map and / or only appears locally in the current three-dimensional point cloud data;

[0146] S2033. The abnormal area is an area where the local light, temperature or heat is abnormal;

[0147] S204. Storing the information of the fixed obstacles, the dynamic obstacles and the abnormal areas in a structured form to form a complete mine-environment model.

[0148] Furthermore, generating multiple candidate travel paths includes the following steps:

[0149] S301. Discretizing the fixed obstacles and channel information in the mine-environment model into a navigation map, wherein the nodes of the navigation map correspond to channel intersections, endpoints and key turning points, and the edges correspond to walkable channels between the nodes;

[0150] S302. Weights are assigned to each edge in the navigation graph based on the distance between nodes, channel width, and safety distance from fixed obstacles. The weights of the edges are calculated based on the path length, obstacle risk, and turning angle.

[0151] S303. Using a graph search algorithm in the navigation map, searching from the transport robot positioning point to the predetermined target point, and generating multiple candidate travel paths.

[0152] S40. Perform a safety assessment on each candidate path to obtain a safety factor of the candidate path; and obtain the optimal path in combination with historical path information;

[0153] Furthermore, the specific calculation formula of the safety factor is:

[0154]

[0155] Among them, S i is the safety factor of the i-th candidate path, α is the obstacle spacing weight, d i is the obstacle spacing value of the i-th candidate path, β is the thermal risk value weight, r i is the thermal risk value of the i-th candidate path, γ is the path smoothness value weight, s i is the smoothness value of the i-th candidate path, δ is the weight of the historical path matching value, m i is the historical path matching value of the i-th candidate path, d meas is the average distance between the key node of the i-th candidate path and the nearest fixed obstacle, d ref Preset the upper limit of the safe distance for the i-th candidate path, r meas is the thermal risk value along the candidate path obtained by the infrared sensor for the i-th candidate path (the higher the thermal risk value, the greater the risk), r ref is the maximum thermal risk reference value preset for the i-th candidate path, obtained through a large amount of data statistics, s meas is the smoothness index of the i-th candidate path, s refis the reference value of the ideal smooth path of the ith candidate path, and the historical path matching value of the ith candidate path is obtained through correlation analysis.

[0156] S50. Enter the blind walking state according to the optimal path; monitor the laser radar and infrared sensor data in real time, update the candidate path when deviating from the predetermined path and / or new risks arise, and re-evaluate the safety of the candidate path;

[0157] S60. When the visual sensor returns to normal and / or the transport robot reaches the target area, exit the blind walking state. For the specific process, refer to Figure 3 .

[0158] The present invention achieves the beneficial effects of high-precision positioning and intelligent navigation in an underground mine environment. Through multi-sensor data acquisition and real-time status monitoring, the coordinated work of vision, laser radar and infrared sensors is realized, ensuring that other sensor modes can be quickly switched when the visual data is abnormal, effectively avoiding positioning errors caused by harsh conditions such as low light and dust. The laser radar is used to generate a three-dimensional point cloud, and combined with the underground mine map, a fine mine-environment model is constructed through the RANSAC and ICP algorithms to accurately identify fixed obstacles, dynamic obstacles and abnormal areas, and provide accurate data support for subsequent path planning. On this basis, the system generates multiple candidate passage paths, and introduces a comprehensive safety factor for risk assessment to ensure that the final selected path has high safety and passage efficiency. In addition, in the blind walking state, the system can monitor environmental changes in real time, dynamically update candidate paths, ensure that the robot performs tasks continuously and stably, and significantly reduce the risk of collisions and accidents. In summary, the present invention has obvious advantages in improving positioning accuracy, enhancing navigation robustness, optimizing path selection and improving system safety, and effectively meets the high standard requirements of complex underground mine operations for intelligent navigation systems.

[0159] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A visual intelligent navigation system for a handling robot based on high-precision positioning, characterized in that: include: The data acquisition module acquires underground mine maps and historical path information; and sets visual sensors, laser radars, and infrared sensors on the handling robot; The status monitoring module analyzes the data collected by the visual sensor in real time and outputs sensor abnormality signals; The switching control module switches the visual sensor to the joint navigation mode of the laser radar and infrared sensor according to the abnormal signal of the sensor; Blind walking path planning module, obtains distance information, temperature data and heat data; S10. Generate three-dimensional point cloud data through distance information, and pre-process the three-dimensional point cloud data, temperature data and heat data to obtain a standard first data set; S20. Obtaining a mine-environment model by spatially matching the data in the standard first data set; identifying fixed obstacles, dynamic obstacles and abnormal areas in the mine-environment model in combination with the underground mine map; S30. Based on fixed obstacles, locate the transport robot and generate multiple candidate paths; S40. Perform a safety assessment on each candidate path to obtain a safety factor of the candidate path; and obtain the optimal path in combination with historical path information; S50. Enter the blind walking state according to the optimal path; monitor the laser radar and infrared sensor data in real time, update the candidate path when deviating from the predetermined path and / or new risks arise, and re-evaluate the safety of the candidate path; S60. When the visual sensor returns to normal and / or the transport robot reaches the target area, the blind walking state is exited.

2. According to claim 1, a transport robot visual intelligent navigation system based on high-precision positioning is characterized by: The underground mine map includes fixed obstacles, mine structure and safe passage information; The visual sensor is used to collect image information of the internal environment of the mine; The distance information is collected by the laser radar, and the temperature data and heat data are collected by the infrared sensor.

3. According to claim 1, a transport robot visual intelligent navigation system based on high-precision positioning is characterized by: The specific steps of obtaining the abnormal signal of the sensor are: The data collected by the visual sensor in real time is image data collected continuously at a predetermined frame rate in the mine; each frame of the image data is accompanied by a timestamp and collection location information; Preprocessing the image data to obtain standard image data; Performing image quality and state analysis on the standard image data to obtain an image data set; the image data set includes image brightness data, image contrast data, image clarity data and image occlusion detection data; An image quality index is obtained by comprehensively measuring the image data set and compared with a preset threshold; when the image quality index is inconsistent with the preset threshold range, a sensor abnormality signal is obtained.

4. According to claim 3, a transport robot visual intelligent navigation system based on high-precision positioning is characterized in that: The calculation formula of the image quality index is: IQIQω1Q b +ω2Q c +ω3Q f +ω4(1-Q o )4 Among them, IQI is the image quality index, ω1 is the image brightness quality score weight, Q b is the image brightness quality score, ω2 is the image contrast quality score weight, Q c is the image contrast quality score, ω3 is the image clarity quality score weight, Q f is the image clarity quality score, ω4 is the image occlusion ratio weight, Q o is the image occlusion ratio.

5. According to claim 1, a transport robot visual intelligent navigation system based on high-precision positioning is characterized by: The specific steps for obtaining the mine-environment model are as follows: S201. Preliminary registration of the three-dimensional point cloud data in the standard first data set with the pre-stored underground mine map, using the RANSAC algorithm based on feature extraction to achieve coarse matching and determine the initial relative position; S202. Use an iterative closest point algorithm to perform fine matching on the data after preliminary registration, and adjust the position and posture between the point cloud data and the pre-stored map; S203. By comparing the difference between the matching result and the pre-stored map, the difference area is divided using the region growing and cluster analysis method, where: S2031. A fixed obstacle is an obstacle that always exists during multiple consecutive matching processes and is consistent with the corresponding position in the pre-stored map; S2032. A dynamic obstacle is an obstacle that is significantly displaced from the fixed obstacle position in the pre-stored map and / or only appears locally in the current three-dimensional point cloud data; S2033. The abnormal area is an area where the local light, temperature or heat is abnormal; S204. Storing the information of the fixed obstacles, the dynamic obstacles and the abnormal areas in a structured form to form a complete mine-environment model.

6. The visual intelligent navigation system for a handling robot based on high-precision positioning according to claim 1 is characterized in that: Generating multiple candidate travel paths includes the following steps: S301. Discretizing the fixed obstacles and channel information in the mine-environment model into a navigation map, wherein the nodes of the navigation map correspond to channel intersections, endpoints and key turning points, and the edges correspond to walkable channels between the nodes; S302. Weights are assigned to each edge in the navigation graph based on the distance between nodes, channel width, and safety distance from fixed obstacles. The weights of the edges are calculated based on the path length, obstacle risk, and turning angle. S303. Using a graph search algorithm in the navigation map, searching from the transport robot positioning point to the predetermined target point, and generating multiple candidate travel paths.

7. The visual intelligent navigation system for a handling robot based on high-precision positioning according to claim 1 is characterized in that: The specific calculation formula of the safety factor is: S i =(d i a*r i β *s i γ *m i d) 1 / (α+β+γ+δ) ; Among them, S i is the safety factor of the i-th candidate path, α is the obstacle spacing weight, d i is the obstacle spacing value of the i-th candidate path, β is the thermal risk value weight, r i is the thermal risk value of the i-th candidate path, γ is the path smoothness value weight, s i is the smoothness value of the i-th candidate path, δ is the weight of the historical path matching value, m i is the historical path matching value of the i-th candidate path.

Images