A mine grid large ore three-dimensional positioning system and method based on fusion of vision and point cloud

Abstract

The application discloses a mine grid large ore three-dimensional positioning system and method based on vision and point cloud fusion, and solves the problem of depth information loss of a pure vision scheme.The system comprises an image acquisition module, an image processing and recognition module, a three-dimensional point cloud acquisition module, a sensor joint calibration module and a vision-point cloud fusion positioning module.The method comprises the following steps: the image acquisition module and the three-dimensional point cloud acquisition module synchronously acquire the overhead image of the ore and the three-dimensional point cloud data of the grid area respectively; the image processing and recognition module acquires the two-dimensional bounding box information; the vision-point cloud fusion positioning module determines the two-dimensional center point of the ore based on the two-dimensional bounding box information; the sensor joint calibration module calibrates the space transformation relationship between the image acquisition module and the three-dimensional point cloud acquisition module; and the vision-point cloud fusion positioning module reversely projects the two-dimensional center point to the three-dimensional space and demarcates the corresponding three-dimensional search area, and calculates the three-dimensional space center coordinates of the target ore.

Description

Technical Field

[0001] This invention belongs to the fields of mining machinery automation, machine vision and multi-sensor fusion technology, specifically involving a three-dimensional positioning system and method for large blocks of ore in mine grids based on vision and point cloud fusion. Background Technology

[0002] In mining and mineral processing, primary crushing is a crucial step. When ore passes through a screen (also known as a "grid screen" or "grate"), ore of suitable size falls directly into the next process, while larger pieces remain on the screen and require secondary crushing by crushing equipment (such as hydraulic breakers, robotic arms, etc.).

[0003] Currently, the identification and positioning of large pieces of ore on the grid mainly relies on manual operation. Operators manually control the crushing equipment to carry out operations by monitoring videos or observing on-site. This method has the following significant drawbacks: (1) Low efficiency: The slow reaction speed of manual identification and operation makes it impossible to achieve continuous and efficient crushing operations; (2) High labor intensity: The working environment is harsh, requiring high concentration from operators, which easily leads to fatigue; (3) Poor positioning accuracy: Inaccurate manual visual positioning leads to the crushing equipment hitting empty or hitting multiple times, increasing energy consumption and equipment wear; (4) Insufficient safety: There are potential safety hazards for personnel, and it is impossible to achieve unmanned and intelligent operation of the entire process.

[0004] While some image processing methods exist for ore particle size identification in existing technologies, they mostly focus on particle size analysis of ore after crushing or on conveyor belts. They lack a complete solution for real-time, precise positioning of stationary large ore pieces in specific scenarios such as on a grid, directly serving the crushing actuator. In particular, how to accurately convert pixel positions in an image into world coordinate system positions usable by the mechanical actuator, and adapt to the complex environment of mines with varying lighting and dust interference, remains a challenge that existing technologies have not yet effectively solved.

[0005] In addition, existing vision-based 2D positioning schemes have inherent limitations: (1) lack of depth information: 2D images cannot provide the vertical distance (Z-axis coordinate) between the ore and the camera, while crushing operations need to know the actual height of the ore surface in order to adjust the impact force or prevent collisions; (2) potential size misjudgment: In 2D images, a large ore lying flat and a small ore piled up may have similar projected areas, and filtering based solely on pixel size may lead to misjudgment; (3) strong dependence on the assumption of the grid plane: Vision-based schemes usually assume that the ore is located on the grid plane (Z=0). When the ore piles up or the grid is deformed due to force, this assumption does not hold, resulting in errors in the XY coordinates mapped onto the plane.

[0006] To address the challenge of depth information acquisition, existing technologies employ solutions using 3D cameras or LiDAR alone. However, these methods are costly and less intuitive and efficient than 2D vision in identifying semantic information such as "what constitutes ore." Therefore, efficiently and cost-effectively integrating the powerful recognition capabilities of 2D images with the precise ranging capabilities of 3D sensors has become a key technological bottleneck for achieving high-precision, intelligent ore crushing operations. Summary of the Invention

[0007] The purpose of this invention is to provide a three-dimensional positioning system for large ore blocks in mine grids based on vision and point cloud fusion, so as to solve the problem of lack of depth information in pure vision solutions, achieve centimeter-level accurate acquisition of the three-dimensional coordinates (X, Y, Z) of the ore space, and more accurately determine the true volume of the ore, so as to provide more reliable data support for intelligent crushing operations.

[0008] Another objective of this invention is to provide a three-dimensional positioning method for large blocks of ore in mine grids based on the fusion of vision and point cloud.

[0009] The technical solution of this invention is: (one) A three-dimensional positioning system for large ore blocks (ore blocks that cannot fall through the grid into the ore pass) in mines, based on vision and point cloud fusion, includes: Image acquisition module, used to acquire top view images of the grid and the ore on it; The image processing and recognition module is connected to the image acquisition module and is used to run a deep learning object detection model to identify mineral targets in the image and output their two-dimensional bounding box information. The 3D point cloud acquisition module is used to acquire 3D point cloud data of the grid area; The sensor joint calibration module is connected to the image acquisition module and the 3D point cloud acquisition module respectively, and is used to calibrate the spatial transformation relationship between the image acquisition module and the 3D point cloud acquisition module. The vision-point cloud fusion localization module, connected to the image processing and recognition module and the 3D point cloud acquisition module, is used for: determining the 2D center point of the ore based on 2D bounding box information; projecting the 2D center point back into 3D space using calibrated spatial transformation relationships, and delineating the corresponding 3D search area in the 3D point cloud data; performing cluster analysis on the point cloud within the 3D search area to determine the point cloud clusters belonging to the target ore; and calculating the 3D spatial center coordinates of the target ore based on the point cloud clusters.

[0010] As a further improvement of the present invention, the visual-point cloud fusion positioning module is also used to calculate the three-dimensional size or volume of the target ore based on the point cloud cluster, so as to verify or correct the size screening results based on two-dimensional image recognition.

[0011] As a further improvement of the present invention, it also includes a point cloud data processing module. The input end of the point cloud data processing module is connected to the three-dimensional point cloud acquisition module, and the output end is connected to the vision-point cloud fusion positioning module. It is used to perform at least one preprocessing operation on the raw point cloud data acquired by the three-dimensional point cloud acquisition module, including filtering and denoising, downsampling, and region of interest extraction, so as to generate concise and high-quality point cloud data and send it to the vision-point cloud fusion positioning module.

[0012] As a further improvement of the present invention, a control output module is also included, which is connected to the vision-point cloud fusion positioning module and the external crushing equipment control system, respectively, for sending the calculated three-dimensional spatial center coordinates to the crushing equipment control system to guide it to perform crushing operations.

[0013] As a further improvement of the present invention, the deep learning object detection model running in the image processing and recognition module is the YOLOv10 model.

[0014] As a further improvement of the present invention, the three-dimensional point cloud acquisition module is a lidar or a depth camera.

[0015] As a further improvement of the present invention, a light source module is also included to provide illumination for the image acquisition module. (two) A method for three-dimensional localization of large ore blocks in mine grids based on vision and point cloud fusion, employing the aforementioned three-dimensional localization system for large ore blocks in mine grids based on vision and point cloud fusion, includes the following steps: A. The image acquisition module and the 3D point cloud acquisition module simultaneously acquire the top view image of the grid and the ore on it, as well as the 3D point cloud data of the grid area. B. The image processing and recognition module identifies ore targets in images using a deep learning object detection model and obtains their two-dimensional bounding box information; the visual-point cloud fusion localization module determines the two-dimensional center point of the ore based on the two-dimensional bounding box information. C. The sensor joint calibration module calibrates the spatial transformation relationship between the image acquisition module and the 3D point cloud acquisition module; the vision-point cloud fusion positioning module uses the pre-calibrated spatial transformation relationship to project the 2D center point in reverse to the 3D space and delineate the corresponding 3D search area in the 3D point cloud data. D. The vision-point cloud fusion localization module performs cluster analysis on the point cloud within the three-dimensional search area to determine the point cloud clusters belonging to the target ore; E. The vision-point cloud fusion positioning module calculates the three-dimensional spatial center coordinates of the target ore based on point cloud clusters.

[0017] Furthermore, in step E, the vision-point cloud fusion localization module calculates the three-dimensional size or volume of the target ore based on the point cloud cluster, and combines the three-dimensional size or volume with the size screening results based on two-dimensional image recognition to finally determine whether it is a large piece of ore.

[0018] The beneficial effects of this invention are: 1. Achieve true three-dimensional positioning: Output complete three-dimensional coordinates including depth (Z-axis) information, upgrading the control of crushing equipment from "two-dimensional planar positioning" to "three-dimensional spatial positioning", greatly improving the accuracy and safety of operation.

[0019] 2. Improve the accuracy of size judgment: By combining point cloud data to calculate the true three-dimensional volume, it can effectively distinguish between "large but thin" and "small but thick" ores, avoiding misjudgments based on two-dimensional projected area, and making the "large block" screening standard more scientific.

[0020] 3. Enhanced System Robustness: When visual recognition is temporarily obstructed or interfered with by reflections, point cloud data can serve as supplementary verification; conversely, when the point cloud is sparse and features are not obvious, visual recognition can provide semantic guidance. Combining the two approaches leverages their strengths and compensates for their weaknesses, resulting in stronger overall system stability.

[0021] 4. Innovative Fusion Strategy: The fusion paradigm of "YOLO rapid initial localization + point cloud fine-tuning for depth determination" proposed in this invention is significantly innovative. It avoids the huge computational overhead of directly processing massive 3D point clouds for target recognition, efficiently locks the target region using YOLOv10, and then performs fine calculations within a small local point cloud, achieving a perfect balance between accuracy and efficiency.

[0022] 5. Provides a data foundation for advanced intelligent control: Three-dimensional coordinates and volume information make it possible to realize advanced functions such as adaptive crushing parameter adjustment (adjusting the impact force according to the size and height of the ore) and optimal crushing path planning (considering obstacle avoidance and efficiency in three-dimensional space). Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the structure of a three-dimensional positioning system for large blocks of ore in a mine grid based on the fusion of vision and point cloud according to the present invention. Figure 2 This is a flowchart of a three-dimensional positioning method for large blocks of ore in a mine grid based on the fusion of vision and point cloud.

[0024] In the diagram: 1-Image acquisition module; 2-Image processing and recognition module; 3-3D point cloud acquisition module; 4-Sensor joint calibration module; 5-Vision-point cloud fusion positioning module; 6-Point cloud data processing module; 7-Control output module; 8-Crushing equipment control system. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Example 1 like Figure 1 As shown, a three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion includes: Image acquisition module 1 is used to acquire a top view image of the grid and the ore on it; Image processing and recognition module 2, connected to image acquisition module 1, is used to run a deep learning object detection model, identify mineral targets in the image and output their two-dimensional bounding box information; 3D point cloud acquisition module 3 is used to acquire 3D point cloud data of the grid area; The sensor joint calibration module 4 is connected to the image acquisition module 1 and the three-dimensional point cloud acquisition module 3 respectively, and is used to calibrate the spatial transformation relationship between the image acquisition module 1 and the three-dimensional point cloud acquisition module 3. The vision-point cloud fusion positioning module 5 is connected to the image processing and recognition module 2 and the 3D point cloud acquisition module 3, respectively. It is used for: determining the 2D center point of the ore based on the 2D bounding box information; projecting the 2D center point back into 3D space using the calibrated spatial transformation relationship, and delineating the corresponding 3D search area in the 3D point cloud data; performing cluster analysis on the point cloud within the 3D search area to determine the point cloud clusters belonging to the target ore; calculating the 3D spatial center coordinates of the target ore based on the point cloud clusters; and calculating the 3D size or volume of the target ore based on the point cloud clusters to verify or correct the size screening results based on 2D image recognition.

[0027] It also includes a point cloud data processing module 6, whose input end is connected to the 3D point cloud acquisition module 3 and whose output end is connected to the vision-point cloud fusion positioning module 5. It is used to perform at least one preprocessing operation on the raw point cloud data acquired by the 3D point cloud acquisition module 3, including filtering and denoising, downsampling, and region of interest extraction, to generate concise and high-quality point cloud data, and send it to the vision-point cloud fusion positioning module 5.

[0028] It also includes a control output module 7, which is connected to the vision-point cloud fusion positioning module 5 and the external crushing equipment control system 8, respectively, and is used to send the calculated three-dimensional spatial center coordinates to the crushing equipment control system 8 to guide it to perform crushing operations.

[0029] The deep learning object detection model running in the image processing and recognition module 2 is the YOLOv10 model.

[0030] The 3D point cloud acquisition module 3 is a lidar or depth camera.

[0031] It also includes a light source module that provides illumination for the image acquisition module 1.

[0032] In this embodiment, the image acquisition module 1 uses a high-resolution (2-megapixel), high dynamic range industrial camera, equipped with a dustproof, waterproof, and shockproof housing, and is fixedly installed 4 meters above the grid. The light source module uses LED supplementary lighting to provide stable and uniform illumination for the image acquisition module 1, suppressing shadows and reflections and enhancing the contrast between the ore and the grid background. The 3D point cloud acquisition module 3 uses a LiDAR, installed directly above the grid, and fixed next to the industrial camera.

[0033] The core of the image processing and recognition module 2 includes a pre-trained YOLOv10 deep learning object detection neural network. The image processing and recognition module 2 receives images from the image acquisition module 1, runs the YOLOv10 network, quickly identifies all ore targets in the image, and outputs the bounding box information for each ore. The bounding box information includes the pixel coordinates (x1, y1, x2, y2) of the upper left and lower right corners, the confidence score, and the category label. Based on a preset "large block" size pixel threshold, it filters out target ores whose size exceeds the limit.

[0034] The 3D point cloud acquisition module 3 acquires 3D point cloud data of the grid area. Each point in the point cloud data contains spatial coordinates (X, Y, X). l Y l Z l ).

[0035] The sensor joint calibration module 4 calibrates the extrinsic matrix (rotation matrix R and translation vector T) between the image acquisition module 1 and the 3D point cloud acquisition module 3, as well as their respective intrinsic parameters, to achieve precise spatial synchronization of the pixel coordinate system, camera coordinate system, and radar coordinate system / point cloud coordinate system.

[0036] The point cloud data processing module 6 receives the raw point cloud data and performs preprocessing such as filtering (removing outlier noise points) and downsampling.

[0037] The visual-point cloud fusion positioning module 5 is the core innovative module, and its functions include: (1) Coordinate synchronization: Using joint calibration parameters, the center point (c) of the two-dimensional pixels output by YOLOv10 is synchronized. x , c y The image is projected back into three-dimensional space to form a line-of-sight vector originating from the optical center of the camera. (2) Point cloud region extraction: Using the line of sight vector as the center, delineate a three-dimensional search region in the point cloud data (such as a small cylinder or cube region centered on the projection point). (3) Solving for the 3D center point: Within the search area, perform cluster analysis on all point cloud data to find point cloud clusters belonging to the same target ore. Calculate the 3D geometric center or weighted center of this point cloud cluster as the final 3D spatial center coordinates (X, Y, X) of the ore. f Y f Z f ).

[0038] (4) Volume / size verification: Based on this point cloud cluster, the actual three-dimensional dimensions (length, width, height) and volume of the ore can be accurately calculated, which can be used to perform secondary verification and correction of the two-dimensional size screening results of YOLOv10, and more accurately determine whether it is a "large piece".

[0039] The control output module 7 is connected to the vision-point cloud fusion positioning module 5 and the external crushing equipment control system 8 to convert the physical coordinates (X) into position. f Y f Z f According to the agreed communication protocol, the system sends a message to the crushing equipment control system 8, guiding it to move to the corresponding coordinate point to perform the crushing operation.

[0040] The image processing and recognition module 2, sensor joint calibration module 4, vision-point cloud fusion positioning module 5, point cloud data processing module 6, and control output module 7 are all hosted and run by a computing platform, which can be an industrial control computer or an embedded AI computing device. This embodiment uses an industrial control computer (equipped with a GPU), deployed in a control room. The industrial camera is connected to the industrial control computer via a gigabit network cable, and the industrial control computer communicates with the breaker's controller via a switch.

[0041] A method for three-dimensional localization of large ore blocks in mine grids based on vision and point cloud fusion, employing the aforementioned three-dimensional localization system for large ore blocks in mine grids based on vision and point cloud fusion, such as... Figure 2 As shown, it includes the following steps: S1. System initialization and multi-sensor joint calibration: S11. Fixed installation of image acquisition module 1 and 3D point cloud acquisition module 3; S12. Perform camera intrinsic parameter calibration and radar intrinsic parameter calibration; S13. Perform joint extrinsic parameter calibration of the camera and radar to obtain the precise spatial transformation relationship between them.

[0042] Calibration process: Using a calibration board of known dimensions, images are taken at multiple locations on the grid plane. OpenCV calibration tools are used to obtain the camera's intrinsic parameters and extrinsic parameters relative to the grid plane, generating a mapping matrix H. A composite calibration board with a special ArUco code pattern and 3D geometric features is used for joint calibration. Simultaneously, camera images and radar point clouds are acquired. By identifying the features of the calibration board in both datasets, an eye-on-hand calibration algorithm is used to solve for the rigid transformation matrix between the camera and radar.

[0043] S2. Multi-source data synchronous acquisition: Image acquisition module 1 and 3D point cloud acquisition module 3 respectively acquire top view images of the grid and the ore on it and 3D point cloud data of the grid area. The image acquisition module 1 and 3D point cloud acquisition module 3 are controlled to be synchronously triggered or strictly synchronized based on timestamps (the camera acquires images at a frequency of 5Hz) to ensure that each frame of image and point cloud data acquired corresponds to the on-site state at the same time.

[0044] S3, Image Processing and Recognition Module 2: Ore identification and 2D center point extraction based on YOLOv10: S31. Preprocess the image acquired by the image acquisition module 1, including noise reduction and contrast enhancement; S32. Input the preprocessed image into the trained YOLOv10 neural network model and set the confidence threshold to 0.3; S33. Obtain the two-dimensional bounding box information of all ore targets output by the model; S34. Based on the pixel area of ​​the bounding box (x2-x1)*(y2-y1) or the longest side of the box, compare it with a preset threshold, set the area threshold T=5000 pixels, filter out "large" ore, and output the center pixel coordinates (c) of the detection box for each large ore. x c y ).

[0045] Model training process: Thousands of grid images with different time periods and ore distributions were collected, and the ore bounding boxes were labeled using the LabelImg tool. The YOLOv10s model was trained and optimized using this dataset within the PyTorch framework, resulting in the final weight file best.pt.

[0046] S4. Point cloud data filtering and denoising: The point cloud data processing module 6 performs filtering and denoising, voxel downsampling and region of interest extraction on the raw point cloud acquired by the 3D point cloud acquisition module 3 to generate a high-quality point cloud, which is then sent to the vision-point cloud fusion localization module 5.

[0047] S5, the sensor joint calibration module 4 calibrates the spatial transformation relationship between the image acquisition module 1 and the 3D point cloud acquisition module 3, and sends it to the vision-point cloud fusion positioning module 5.

[0048] S6. Visual-point cloud fusion positioning module 5 calculates the 3D center point: S61, Spatial Projection: For each pixel center point (c) obtained by YOLOv10, xi c yi Using camera intrinsic parameters and camera-radar extrinsic parameters, the approximate projection line of the camera in the radar coordinate system is calculated. S62, Point Cloud Association: In point cloud data, search for spatial points within a preset three-dimensional neighborhood along the projection line direction; S63. Point cloud clustering and filtering: Perform clustering algorithms on the searched points to find the point cloud clusters that are most likely to correspond to the visually recognized ore. S64. Three-dimensional coordinate calculation: Calculate the average coordinates of all points in the cloud cluster, or take its centroid to obtain the precise three-dimensional center coordinates (X) of the ore after fusion. i Y i Z i ); S65. Volume-assisted decision: Based on the point cloud cluster, calculate the approximate three-dimensional bounding box and volume of the ore, compare this volume information with the preset three-dimensional volume threshold, and use it as one of the decision bases for the final determination of "whether it needs to be crushed", thereby enhancing reliability.

[0049] S7. Three-Dimensional Coordinate Output and Control: The control output module 7 sends the calculated three-dimensional coordinate list to the PLC of the crushing equipment control system 8 via the Modbus TCP protocol. The PLC controls the breaker hammer to move to the (X, Y, Z) coordinates and execute one crushing action. The above process is repeated until there are no large pieces of ore that meet the conditions in the image. The crushing equipment control system 8 can not only locate the horizontal position (X, Y), but also optimize the falling stroke or impact angle of the crushing tool based on the height (Z) information, achieving more intelligent and safer operation.

Claims

1. A three-dimensional positioning system for large blocks of ore in mine grids based on vision and point cloud fusion, characterized in that: include: Image acquisition module (1) is used to acquire a top view image of the grid and the ore on it; The image processing and recognition module (2) is connected to the image acquisition module (1) and is used to run a deep learning target detection model to identify the ore target in the image and output its two-dimensional bounding box information. The three-dimensional point cloud acquisition module (3) is used to acquire three-dimensional point cloud data of the grid area; The sensor joint calibration module (4) is connected to the image acquisition module (1) and the three-dimensional point cloud acquisition module (3) respectively, and is used to calibrate the spatial transformation relationship between the image acquisition module (1) and the three-dimensional point cloud acquisition module (3); The vision-point cloud fusion positioning module (5) is connected to the image processing and recognition module (2) and the three-dimensional point cloud acquisition module (3), respectively, and is used to: determine the two-dimensional center point of the ore based on the two-dimensional bounding box information; project the two-dimensional center point back to the three-dimensional space using the calibrated spatial transformation relationship, and delineate the corresponding three-dimensional search area in the three-dimensional point cloud data; perform cluster analysis on the point cloud in the three-dimensional search area to determine the point cloud cluster belonging to the target ore; and calculate the three-dimensional spatial center coordinates of the target ore based on the point cloud cluster.

2. The three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion as described in claim 1, characterized in that: The visual-point cloud fusion positioning module (5) is also used to calculate the three-dimensional size or volume of the target ore based on the point cloud cluster, so as to verify or correct the size screening results based on two-dimensional image recognition.

3. The three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion as described in claim 1, characterized in that: It also includes a point cloud data processing module (6), the input end of which is connected to the three-dimensional point cloud acquisition module (3) and the output end is connected to the vision-point cloud fusion positioning module (5). It is used to perform at least one preprocessing operation on the raw point cloud data acquired by the three-dimensional point cloud acquisition module (3), including filtering and denoising, downsampling, and region of interest extraction, so as to generate concise and high-quality point cloud data and send it to the vision-point cloud fusion positioning module (5).

4. The three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion as described in claim 1, characterized in that: It also includes a control output module (7), which is connected to the vision-point cloud fusion positioning module (5) and the crushing equipment control system (8) respectively, and is used to send the calculated three-dimensional spatial center coordinates to the crushing equipment control system (8).

5. A three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion according to any one of claims 1-4, characterized in that: The deep learning target detection model running in the image processing and recognition module (2) is the YOLOv10 model.

6. A three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion according to any one of claims 1-4, characterized in that: The three-dimensional point cloud acquisition module (3) is a lidar or depth camera.

7. A three-dimensional positioning system for large blocks of ore in a mine grid based on vision and point cloud fusion according to any one of claims 1-4, characterized in that: It also includes a light source module that provides illumination for the image acquisition module (1).

8. A method for three-dimensional positioning of large blocks of ore in mine grids based on vision and point cloud fusion, characterized in that: The three-dimensional positioning system for large blocks of ore in a mine grid, based on vision and point cloud fusion as described in any one of claims 1-7, includes the following steps: A. The image acquisition module (1) and the three-dimensional point cloud acquisition module (3) synchronously acquire the top view image of the grid and the ore on it and the three-dimensional point cloud data of the grid area, respectively. B. Image processing and recognition module (2) identifies the ore target in the image through a deep learning target detection model and obtains its two-dimensional bounding box information; visual-point cloud fusion positioning module (5) determines the two-dimensional center point of the ore based on the two-dimensional bounding box information; C. Sensor joint calibration module (4) calibrates the spatial transformation relationship between image acquisition module (1) and three-dimensional point cloud acquisition module (3); visual-point cloud fusion positioning module (5) uses the pre-calibrated spatial transformation relationship to project the two-dimensional center point in reverse to the three-dimensional space and delineate the corresponding three-dimensional search area in the three-dimensional point cloud data. D. The visual-point cloud fusion positioning module (5) performs cluster analysis on the point cloud within the three-dimensional search area to determine the point cloud cluster belonging to the target ore; E. Visual-point cloud fusion positioning module (5) calculates the three-dimensional spatial center coordinates of the target ore based on the point cloud cluster.

9. A three-dimensional positioning method for large blocks of ore in a mine grid based on vision and point cloud fusion as described in claim 8, characterized in that: In step E, the visual-point cloud fusion positioning module (5) calculates the three-dimensional size or volume of the target ore based on the point cloud cluster, and combines the three-dimensional size or volume with the size screening results based on two-dimensional image recognition to finally determine whether it is a large piece of ore.

Images