Underground unmanned vehicle identification method and system based on multi-sensor fusion

Abstract

The invention belongs to the technical field of unmanned vehicle identification, and discloses an underground unmanned vehicle identification method and system based on multi-sensor fusion, and the method comprises the steps: carrying out the preprocessing of point cloud data obtained by a radar, obtaining noise reduction point cloud data, projecting the noise reduction point cloud data to an image collected by a camera, and obtaining a first target region; preprocessing an image acquired by a camera to obtain an enhanced image, and identifying the enhanced image to obtain a second target area; the correlation degree of the first target area and the second target area is judged, and if the correlation degree is smaller than a set value, the target position, the relative speed and the target type of the unmanned vehicle are output; and if the correlation degree is greater than or equal to a set value, secondary detection is carried out, so that the method can adapt to different sensor configurations and different environment changes, is suitable for sensing and positioning tasks of various unmanned vehicles, and reduces errors caused by a single sensor in a working process.

Description

Technical Field

[0001] This invention belongs to the field of unmanned vehicle recognition technology, and relates to a method and system for identifying unmanned underground vehicles based on multi-sensor fusion. Background Technology

[0002] With the rapid development of autonomous driving and deep learning technologies, deep learning-based target detection and image classification methods for unmanned vehicles have been widely applied in various fields. However, in special underground environments such as mines, there are often problems such as insufficient lighting, significant smoke and dust interference, complex spatial structures, and frequent reflections and occlusions. Traditional target recognition methods mostly rely on a single sensor for data acquisition, such as monocular cameras, lidar, and millimeter-wave radar, which are difficult to achieve stable and accurate perception. Monocular cameras are extremely sensitive to changes in lighting, and their detection accuracy fluctuates significantly under alternating light and dark conditions underground. LiDAR suffers severe signal attenuation in dusty and smoky environments and is prone to point cloud data loss due to tunnel bends, surface reflections, or partial occlusions. Although millimeter-wave radar has strong anti-interference capabilities, its low spatial resolution makes it difficult to reliably identify small or dense targets.

[0003] Multi-sensor fusion technology integrates complementary information from different sensors, such as cameras, lidar, and millimeter-wave radar, at different levels to improve the robustness and environmental adaptability of target recognition and localization. Existing fusion strategies are mostly based on ideal sensor data and lack the ability to handle common downhole issues like extreme noise, partial signal loss, and asynchronous data from different sources. Furthermore, maintaining continuous high-precision target monitoring is difficult in narrow tunnels. For example, vision-based recognition modules experience a sharp increase in error under low-light conditions, while lidar point clouds are prone to degradation due to complex wellbore structures and multipath reflections, further increasing the uncertainty of the fusion system. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for identifying unmanned underground vehicles based on multi-sensor fusion.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for identifying unmanned vehicles in underground mines based on multi-sensor fusion, comprising the following steps: preprocessing point cloud data acquired by radar to obtain denoised point cloud data; projecting the denoised point cloud data onto an image acquired by a camera to obtain a first target region; preprocessing the image acquired by the camera to obtain an enhanced image; identifying the enhanced image to obtain a second target region; judging the correlation between the first target region and the second target region; if the correlation is less than a set value, outputting the unmanned vehicle target position, relative speed, and target type; if the correlation is greater than or equal to the set value, performing a secondary detection.

[0006] Furthermore, the preprocessing of the image acquired by the camera to obtain the enhanced image includes: converting the image to the HSV color space to obtain the hue components. saturation component and brightness component Based on the Retinex fusion algorithm for luminance components Enhancement is performed to obtain the enhanced luminance component. Based on the enhanced luminance component For saturation components Correction is performed to obtain the corrected saturation components. ; to color components Corrected saturation components and enhanced luminance components The image is enhanced by converting from the HSV color space to RGB.

[0007] Furthermore, the fusion Retinex algorithm is as follows:

[0008] in, The total number of scales, It is a constant. The coefficients at each scale, For the final received image information, To incorporate the new center wrap function generated by the bilateral filter.

[0009] Furthermore, the preprocessing of the point cloud data acquired by the radar to obtain denoised point cloud data includes: The system acquires the actual measurement value of the radar-acquired UAV at the previous moment and the predicted value at the current moment; the actual measurement value at the previous moment is based on Kalman gain. Obtain the actual measured value at the current moment; update the error covariance matrix based on the predicted value and the actual measured value at the current moment. By updating the error covariance matrix Correcting prediction errors is used for data noise reduction.

[0010] Furthermore, the Kalman gain for:

[0011] in, Let be the error covariance matrix. This is the transformation matrix between the state estimate and the measured value at the previous time step. Let be the covariance matrix of the observation noise.

[0012] Furthermore, the covariance matrix for:

[0013] in, It is the identity matrix. For Kalman gain, Let be the transition matrix. Let be the error covariance matrix of the previous time step.

[0014] Furthermore, the determination of the correlation between the first target region and the second target region is as follows:

[0015] in, The overlapping area of ​​the region The total area of ​​the region. This is the set value.

[0016] Furthermore, before performing time and space calibration on the radar and camera, coordinate transformation is also included, wherein the radar coordinate system is transformed from the millimeter-wave radar coordinate system to the world coordinate system according to the Cartesian coordinate system; and the camera pixel coordinate system is transformed to the world coordinate system.

[0017] Furthermore, the predicted value at the current moment is :

[0018] in, Let be the transition matrix. This is the estimated radar target state vector.

[0019] This invention also provides an underground unmanned vehicle recognition system based on multi-sensor fusion, comprising: a first processing module for preprocessing point cloud data acquired by radar to obtain denoised point cloud data, and projecting the denoised point cloud data onto an image acquired by a camera to obtain a first target region; a second processing module for preprocessing the image acquired by the camera to obtain an enhanced image, and recognizing the enhanced image to obtain a second target region; a judgment module for judging the correlation between the first target region and the second target region, and if the correlation is less than a set value, outputting the unmanned vehicle target position, relative speed, and target type; and a detection module for performing secondary detection if the correlation is greater than or equal to a set value.

[0020] Compared with the prior art, the present invention has the following beneficial technical effects: This invention discloses a multi-sensor fusion-based method for identifying unmanned vehicles in underground mines. By fusing camera and radar data, it can improve target recognition accuracy under different environmental conditions, especially exhibiting stronger robustness in low visibility and complex working environments. Preprocessing the collected data through Kalman gain and the Retinex fusion algorithm enhances the accuracy and reliability of data fusion, reduces errors caused by single sensors during operation, and adapts to different sensor configurations and environmental changes, making it suitable for various unmanned vehicle perception and localization tasks. Attached Figure Description

[0021] Figure 1 This is a flowchart of a method for identifying unmanned underground vehicles based on multi-sensor fusion according to the present invention; Figure 2 This is a schematic diagram of a method for identifying unmanned underground vehicles based on multi-sensor fusion according to the present invention. Figure 3 This is a schematic diagram illustrating the principle of preprocessing images acquired by a camera in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] Example 1 A method for identifying unmanned underground vehicles based on multi-sensor fusion, such as Figure 1 As shown, the process includes the following steps: preprocessing the point cloud data acquired by the radar to obtain denoised point cloud data, projecting the denoised point cloud data onto the image acquired by the camera to obtain a first target region; preprocessing the image acquired by the camera to obtain an enhanced image, and identifying the enhanced image to obtain a second target region; judging the correlation between the first target region and the second target region, and if the correlation is less than a set value, outputting the target position, relative speed, and target type of the unmanned vehicle; if the correlation is greater than or equal to the set value, performing a secondary detection.

[0024] Specifically, the radar is first calibrated in time, and the camera is calibrated in space. In this embodiment, the radar is a millimeter-wave radar. Based on the coordinate system, the point cloud data collected after radar calibration is preprocessed and used to generate a first target region on the image collected by the camera. The location, distance, and speed information of obstacles are also obtained. After camera calibration, the collected video and images are processed by image noise reduction. A second target region is generated by YOLOv11 vision detection, and the target category information is obtained. The correlation between the first target region and the second target region is calculated. If the correlation is less than or equal to a set value, the target type, relative speed, and relative position are output. If the correlation is greater than the set value, the Adaboost classifier is used for re-detection to check whether the target has been identified. Here, the target is an unmanned vehicle.

[0025] Before performing time and space calibration on the radar and camera, coordinate transformation is also included. In the millimeter-wave radar coordinate system transformation stage: based on the right-hand Cartesian coordinate system, the positional relationship between the world coordinates and the millimeter-wave radar coordinate system is established. First, the raw scan data of the millimeter-wave radar in the two-dimensional polar coordinate plane is acquired. This data typically includes the distance between the target point and the radar. and azimuth Secondly, the installation position parameters of the radar in the vehicle coordinate system are set. Taking the radar's mounting point on the vehicle as a reference, its longitudinal offset in the vehicle coordinate system is defined. and height offset The original polar coordinate data of the radar Transform to 3D coordinates in vehicle coordinate system .

[0026]

[0027]

[0028]

[0029] Points in the radar scanning plane are mapped to a vehicle-centered world coordinate system, where The horizontal position is the horizontal position. The longitudinal position of the vehicle. This refers to the position in the height direction.

[0030] The transformed point cloud data is combined with temporal filtering methods such as Kalman filtering to achieve dynamic noise reduction through state estimation and error covariance update, thereby improving the accuracy and stability of target perception.

[0031] The camera's pixel coordinate system must also be transformed from the world coordinate system accordingly. The origin is defined as the center of the camera's optical lens. Each coordinate axis points parallel to the world coordinate system, and the camera's optical axis is... Establish the camera coordinate system (axis). , , , Then the coordinates of a point in the world coordinate system space are (). , , Transform to the camera coordinate system, where the coordinates are represented as follows: The conversion relationship between the two is as follows:

[0032] in, The overall transformation matrix, i.e., the rotation and translation matrix between two coordinate systems, includes the orthogonal rotation matrix. Translation matrix T.

[0033] The coordinates in the camera Coordinates in a two-dimensional image coordinate system The conversion relationship is as follows:

[0034] in, This refers to the camera's focal length.

[0035] In addition, in the pixel coordinate system The Middle Okay, number The pixels of a column are represented in this coordinate system as And image coordinate system Since millimeters are used as the basic unit, coordinate transformation is also required. The transformation relationship is as follows:

[0036] In summary, pixel coordinate system Points in Points in the world coordinate system , , The conversion relationship between them is:

[0037] in, , , These consist of a camera projection matrix, an extrinsic parameter matrix, and an intrinsic parameter matrix. Additionally, the camera needs to be calibrated to prepare for accuracy measurements and to meet the requirements of target detection and recognition tasks.

[0038] To obtain the image coordinates after lens distortion correction, the image needs to be modified to obtain the corrected image coordinates. :

[0039] in, , , The radial distortion coefficient is... The tangential distortion coefficient is... This represents the distance from a pixel to the center of the image.

[0040] After completing the joint calibration, the time information of the camera and radar is synchronized, and the same time point of the two sensors is used as the starting point for fusion.

[0041] First, the target scanned by the millimeter-wave radar is projected onto the image plane:

[0042] in, , Image coordinates of the left diagonal vertex of the region of interest from the millimeter-wave radar. The intersection of the diagonals To perceive the aspect ratio of the target, This is the scaling factor between physical and pixel dimensions. The radar's first target area is set according to the preset target mission.

[0043] Millimeter-wave radar can still operate normally under adverse weather conditions and environments, is less affected by lighting conditions, and has a long operating range. However, it is prone to missed detections and false detections when detecting moving targets, and the raw data collected from downhole may contain random noise. Therefore, after obtaining the raw radar data, it is necessary to perform noise reduction processing on the radar point cloud data to reduce noise interference. Kalman filtering is an effective method for state estimation of linear dynamic systems. It accurately estimates the internal state of the process through a series of measured observations and effectively removes noise effects when updating the target state. Considering that the relationship between the position, velocity, and other states reflected in millimeter-wave radar data and the actual measured values ​​is often nonlinear, this invention selects extended Kalman filtering, which handles the state estimation problem of nonlinear systems, to perform noise reduction processing on the radar point cloud data.

[0044] The measurement vector of the extended Kalman filter is the position and velocity state of the radar-detected target. One detection target In time The state vector can be represented as:

[0045] in, The longitudinal distance from which the radar reaches the target. The lateral distance at which the radar reaches the target. Indicates the target's movement speed. This represents the target's azimuth angle relative to the radar. During prediction, the EFK extended Kalman filter describes the process model of the state vector. This is used to represent changes in the target's distance, speed, and azimuth.

[0046] in, This represents the time interval between the states before and after the movement.

[0047] After Jacobi linearization, the state transition matrix is ​​as follows:

[0048] The error covariance is obtained from the state transition matrix as follows:

[0049] in This represents system noise. The predicted value for the current time step is obtained based on the transition matrix and the actual measurement value from the previous time step.

[0050]

[0051] The transformation matrix is ​​obtained based on the predicted and estimated values ​​at the same time:

[0052] The transition matrix is ​​based on the error covariance matrix. Obtain Kalman gain

[0053] Error covariance matrix for:

[0054] in, It is the identity matrix. For Kalman gain, Let be the transition matrix. Let be the error covariance matrix of the previous time step.

[0055] Kalman gain for:

[0056] in, Let be the error covariance matrix. This is the transformation matrix between the state estimate and the measured value at the previous time step. Let be the covariance matrix of the observation noise.

[0057] By updating the error covariance matrix Correcting prediction errors is used for data noise reduction.

[0058] The image acquired by the camera is preprocessed to obtain an enhanced image, including: converting the image to the HSV color space to obtain hue components. saturation component and brightness component Based on the Retinex fusion algorithm for luminance components Enhancement is performed to obtain the enhanced luminance component. Based on the enhanced luminance component For saturation components Correction is performed to obtain the corrected saturation components. ; to color components Corrected saturation components and enhanced luminance components The image is enhanced by converting from the HSV color space to RGB.

[0059] Image and video data from underground mines suffers from low illumination and dust, resulting in low brightness, uneven illumination, color distortion, and loss of detail. This invention employs an image enhancement method based on the Retinex algorithm to improve image quality and suppress phenomena such as halo and edge blurring. Specifically: First, the image data is converted from RGB space to HSV space, generating a space with three components: saturation, brightness, and hue. While ensuring the hue component remains unchanged, the algorithm is used to enhance the brightness component. Then, the saturation component is calibrated and corrected. Finally, the enhanced HSV space image is converted back to RGB space to generate an enhanced and denoised RGB image. Figure 3 As shown.

[0060] The Retinex algorithm can be expressed as follows:

[0061] in, For the final received image information, and These are the reflection component and illuminance component of light emitted by an object.

[0062] The reflection component of the object itself is obtained by converting it to the number domain:

[0063] To improve image fidelity, a multi-scale Retinex algorithm needs to be derived:

[0064] in, The total number of scales, The coefficients at each scale.

[0065] Then, the multi-scale Retinex algorithm is combined with an improved bilateral filter. First, the original image is logarithmically transformed to obtain:

[0066] Since the reflection component of the object itself can be obtained by subtracting the illuminance component from the original image, the formula for estimating the illuminance component is:

[0067] in, For the center-wrap function, Let be the scale function of the central illusion, and be the standard deviation of the low-pass Gaussian filter function. Different The value has a significant impact on the image enhancement effect; values ​​that are too large or too small will cause blurring and distortion of the image.

[0068] The low-pass Gaussian filter function is replaced with a bilateral filter function, and the bilateral filter function is re-convolved with the original image to estimate the reflection component. To prevent the center wrap function between two pixels with similar gray values ​​from changing due to coordinate differences and affecting the final illumination component, a correction function is added to the bilateral filter function. The weight factor of the corrected bilateral filter function is used as the center wrap function for the multi-scale Retinex algorithm.

[0069] in, and These are the coordinates and grayscale value of the image center point, respectively. Let be the standard deviation of the Gaussian function in the spatial domain. Let be the standard deviation over the range of the Gaussian function. For correction functions, if the difference between the grayscale value of a pixel and the center point is less than or equal to... ,but

[0070] otherwise, =1.

[0071] The improved bilateral filtering described above is combined with the original multi-scale Retinex algorithm to obtain the fused Retinex algorithm:

[0072] in, The total number of scales, It is a constant. The coefficients at each scale, For the final received image information, To incorporate the new center wrap function generated by the bilateral filter.

[0073] Furthermore, the original image is converted to HSV space using the following formula:

[0074] in, and These represent the maximum and minimum values ​​in the RGB color space; luminance component. The maximum value in RBG; saturation component It is a proportional value; For hue components; , , They represent the three colors, red, green, and blue, and are spaced apart on the color wheel. complementary hue difference .

[0075] The fused Retinex algorithm is applied to the luminance component while keeping the hue component unchanged. Finally, the saturation component of the image is corrected based on the enhancement change of the luminance component, thus completing the enhancement of the low-light image.

[0076]

[0077] The illuminance component estimated in the luminance space; This refers to the center-wrap function proposed in the improved bilateral filtering algorithm described above; The scale number; These are the weighting coefficients.

[0078] Since enhancing the luminance component will also cause a corresponding change in saturation, it is necessary to correct the saturation component.

[0079] and These represent the enhanced saturation and brightness components, respectively. It is a constant; This is for adjusting the coefficient.

[0080]

[0081] To enhance the point position coordinates; and These are the locations of the enhancement points. The average brightness and saturation of all points within a given area; To enhance the brightness variance of the points; The variance of saturation; These are the coordinates of the pixels within the neighborhood.

[0082] Finally, complete the image conversion from HSV space back to RGB space:

[0083] ; ; ; parameters , ; Modulo operation is represented. This is the base symbol.

[0084] Determine the correlation between the first target region and the second target region:

[0085] in, The overlapping area of ​​the region The total area of ​​the region. This is the set value.

[0086] If the correlation is less than the set value, the output will only show the target position, relative speed, and target type of the unmanned vehicle. Further decision fusion is needed. Given the complex working environment in mines, conflicts may occur between DS evidence, requiring improvement and the establishment of a recognition framework. ,definition Let be the evidence vector, where elements in For sensors Given The confidence level of the target class, where , Thus, the evidence set Additionally, the Pearson coefficient is used to establish the distance between the two vectors to determine their deviation. It is used to characterize the difference between two individual spatial vectors.

[0087]

[0088] When the Pearson coefficient is between -1 and 1, the correlation coefficient is greater than 0 when two pieces of evidence are positively correlated, and the higher the degree of correlation, the greater the distance between the evidence vectors. The smaller the coefficient, the lower the correlation; conversely, when two pieces of evidence are negatively correlated, the coefficient is less than 0, indicating a lower degree of correlation and a smaller distance between the evidence vectors. The larger.

[0089] Construct a distance matrix D to indicate the overall relevance and conflict of evidence.

[0090]

[0091] Normalization process yields the relative degree of conflict of evidence :

[0092] Evidence weighting coefficients are defined to describe the importance of evidence in the fusion process, i.e. for:

[0093] The original probability assignment is When, the probability of redistribution is

[0094]

[0095] The redistributed probabilities are then substituted into the DS evidence theory decision formula for fusion, thus completing the improvement process.

[0096] Establish a decision fusion model combining millimeter-wave radar and visual cameras: This model combines the relative position and velocity information of targets ahead of the autonomous vehicle provided by millimeter-wave radar with the target type information collected by the camera using a previous data-level fusion method, overcoming the limitations of data acquired from a single sensor. After clarifying the task requirements, an identification framework is established. This process involves clearly identifying all target outcomes encountered during environmental perception. Then, the results of data-level fusion (various target types and confidence information) are weighted and redistributed to determine the target types and basic probability allocation functions. Finally, this function is combined with DS evidence theory to obtain a final accurate assessment of the target type and confidence level. Figure 2 As shown.

[0097] Example 2 The present invention also provides an underground unmanned vehicle identification system based on multi-sensor fusion, comprising: a first processing module, a second processing module, a judgment module, and a detection module.

[0098] The first processing module preprocesses the point cloud data acquired by the radar to obtain denoised point cloud data, and projects the denoised point cloud data onto the image acquired by the camera to obtain the first target region; the second processing module preprocesses the image acquired by the camera to obtain an enhanced image, and identifies the enhanced image to obtain the second target region; the judgment module judges the correlation between the first target region and the second target region, and if the correlation is less than a set value, outputs the target position, relative speed and target type of the unmanned vehicle; the detection module performs secondary detection if the correlation is greater than or equal to the set value.

[0099] The unmanned underground vehicle identification system based on multi-sensor fusion provided by this invention can implement the same method steps as the above method, so it will not be described again.

[0100] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

Claims

1. A method for identifying unmanned underground vehicles based on multi-sensor fusion, characterized in that, Includes the following steps: The point cloud data acquired by the radar is preprocessed to obtain noise-reduced point cloud data, and the noise-reduced point cloud data is projected onto the image captured by the camera to obtain the first target area. The image acquired by the camera is preprocessed to obtain an enhanced image, and the enhanced image is then identified to obtain a second target region; The correlation between the first target area and the second target area is judged. If the correlation is less than a set value, the target position, relative speed and target type of the unmanned vehicle are output. If the correlation is greater than or equal to the set value, a second detection is performed.

2. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 1, characterized in that, The process of preprocessing the images acquired by the camera to obtain enhanced images includes: Convert the image to the HSV color space to obtain the hue components. saturation component and brightness component ; Luminance component based on the fusion Retinex algorithm Enhancement is performed to obtain the enhanced luminance component. ; Based on the enhanced luminance component For saturation components Correction is performed to obtain the corrected saturation components. ; Tone components Corrected saturation components and enhanced luminance components The image is enhanced by converting from the HSV color space to RGB.

3. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 2, characterized in that, The fusion Retinex algorithm is as follows: in, The total number of scales, It is a constant. The coefficients at each scale, For the final received image information, To incorporate the new center wrap function generated by the bilateral filter.

4. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 1, characterized in that, The preprocessing of the point cloud data acquired by the radar to obtain denoised point cloud data includes: Obtain the actual measurement value of the radar-acquired UAV at the previous moment and the predicted value at the current moment; The actual measurement value at the previous moment is based on Kalman gain. Obtain the actual measurement value at the current moment; Update the error covariance matrix based on the predicted and actual measured values ​​at the current moment. ; By updating the error covariance matrix Correcting prediction errors is used for data noise reduction.

5. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 4, characterized in that: Kalman gain for: in, Let be the error covariance matrix. This is the transformation matrix between the state estimate and the measured value at the previous time step. Let be the covariance matrix of the observation noise.

6. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 5, characterized in that: The error covariance matrix for: in, It is the identity matrix. For Kalman gain, Let be the transition matrix. Let be the error covariance matrix of the previous time step.

7. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 6, characterized in that: The determination of the correlation between the first target region and the second target region is as follows: in, The overlapping area of ​​the region The total area of ​​the region. This is the set value.

8. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 1, characterized in that: Before performing time and space calibration on the radar and camera, coordinate transformation is also included. The radar coordinate system is transformed from the millimeter-wave radar coordinate system to the world coordinate system based on the Cartesian coordinate system; the camera pixel coordinate system is transformed to the world coordinate system.

9. The method for identifying unmanned underground vehicles based on multi-sensor fusion according to claim 4, characterized in that, The predicted value at the current moment is : in, Let be the transition matrix. This is the estimated radar target state vector.

10. A multi-sensor fusion-based identification system for unmanned underground vehicles, characterized in that: First processing module: used to preprocess the point cloud data acquired by radar to obtain noise-reduced point cloud data, and project the noise-reduced point cloud data onto the image captured by the camera to obtain the first target area; The second processing module is used to preprocess the image acquired by the camera to obtain an enhanced image, and to identify the enhanced image to obtain a second target region. Judgment module: used to judge the correlation between the first target area and the second target area. If the correlation is less than the set value, it outputs the target position, relative speed and target type of the unmanned vehicle. Detection module: Used to perform secondary detection if the correlation is greater than or equal to a set value.

Images