A method and system for obstacle laser detection for unmanned mine cars

Abstract

The application relates to the technical field of mine card obstacle detection, and discloses a laser obstacle detection method and system for unmanned mine cards, which comprises the following steps: acquiring the signal-to-noise ratio of all detection echo signals, determining the adjustment strategy of the corresponding detection echo signals based on the signal-to-noise ratio, constructing a point cloud topology network based on point cloud data and performing clustering to determine a spatial feature cluster, extracting frequency spectrum features containing frequency spectrum data based on the spatial feature cluster, determining a spatial frequency spectrum feature cluster based on the synchronous change of the frequency spectrum features, determining effective echo frequency spectrum according to the frequency spectrum space and the frequency spectrum time of the spatial frequency spectrum feature cluster, comparing the effective echo frequency spectrum with a historical frequency spectrum database to determine whether there is obstacle information of the effective echo frequency spectrum, determining the obstacle information based on an obstacle recognition model and the effective echo frequency spectrum, and determining the effective obstacle of the driving route according to the obstacle information. The application ensures the reliability and recognition stability of obstacle detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mine truck obstacle detection, in particular to a laser obstacle detection method and system for unmanned mine trucks. BACKGROUND

[0002] With the acceleration of the intelligentization process of the mining industry, unmanned mine trucks have gradually become an important equipment for open-pit mining due to their advantages in improving operation efficiency, reducing labor costs, and reducing safety accidents. However, the complex obstacles in the mining area environment, such as rock rolling, cross traffic of other operation vehicles, and temporary accumulations, have higher requirements for the precision and accuracy of obstacle detection of unmanned mine trucks.

[0003] Chinese Patent No. CN111537994B discloses an unmanned mine truck obstacle detection method, which converts obstacle data obtained by laser radar and millimeter wave radar into corresponding vehicle coordinate systems respectively; adopts grid map height difference combined with neighborhood difference ground detection to draw a 0, 1 binary graph of "ground-elevated point"; adopts a multi-parameter model to cluster the elevated points; judges whether the clustering result is an obstacle that affects the normal driving of the vehicle according to the motion trajectory of the vehicle; detects whether the vehicle is in a drivable area; matches the obstacle data obtained by the millimeter wave radar with the obstacle data obtained by the laser radar, and outputs the final result. The present application is aimed at the actual application environment of mine dump trucks. As can be seen, when grid maps or simple density clustering are used, it is difficult to distinguish the spatial characteristics of ground protrusions and real obstacles, especially for irregular rock piles and low obstacles, which may result in over-clustering or under-clustering, thereby leading to false detection or missed detection. Moreover, the present application does not identify obstacles through the dynamic changes of the spectrum of the laser radar, which makes it difficult to accurately distinguish the differences between obstacles and the environment, resulting in limitations in matching obstacles.

[0004] Therefore, it is necessary to design a laser obstacle detection method and system for unmanned mine trucks to solve the problems in the current technology. SUMMARY

[0005] In view of this, the present application proposes a laser obstacle detection method and system for unmanned mine trucks, aiming to solve the problems in the current technology.

[0006] In one aspect, the present application proposes a laser obstacle detection method for unmanned mine trucks, comprising:

[0007] Based on the laser radar emitting radar detection signals to the driving route and determining all detection echo signals, the signal-to-noise ratio of all detection echo signals is obtained, the adjustment strategy of the corresponding detection echo signal is determined based on the signal-to-noise ratio, and the echo time, spectral data, and point cloud data are determined based on the adjusted all detection echo signals;

[0008] determining a constant travel strategy or a suspected obstacle strategy of the travel route according to the echo time, when determining the suspected obstacle strategy, constructing a point cloud topology network based on the point cloud data and determining a spatial feature cluster by clustering, extracting a spectral feature containing spectral data based on the spatial feature cluster, determining a spatial spectral feature cluster based on synchronous changes of the spectral feature;

[0009] determining an effective echo spectrum according to the spectral space and the spectral time of the spatial spectral feature cluster, comparing the effective echo spectrum with a historical spectrum database to determine whether there is obstacle information of the effective echo spectrum, when there is no obstacle information of the effective echo spectrum, obtaining an obstacle sample set and constructing an obstacle recognition model;

[0010] determining obstacle information based on the obstacle recognition model and the effective echo spectrum, and determining an effective obstacle of the travel route according to the obstacle information.

[0011] Further, when determining an adjustment strategy of a corresponding detection echo signal based on the signal-to-noise ratio, comprising:

[0012] when there is a signal-to-noise ratio greater than or equal to a signal-to-noise ratio threshold, performing phase compensation based on Doppler frequency;

[0013] when there is a signal-to-noise ratio less than the signal-to-noise ratio threshold, extracting a complex signal of each detection echo signal in a short time window, determining a conjugate product with a previous detection echo signal, and extracting a phase difference to determine a phase difference sequence, performing phase unwrapping and accumulation based on the phase difference sequence to determine an accumulated phase, and performing phase compensation based on the accumulated phase.

[0014] Further, when determining a constant travel strategy or a suspected obstacle strategy of the travel route according to the echo time, comprising:

[0015] obtaining a standard echo time interval, and comparing the standard echo time interval with the echo time;

[0016] when there is an echo time greater than the left boundary of the standard echo time interval and less than the right boundary of the standard echo time interval, determining that the travel route is a constant travel strategy, otherwise, determining that the travel route is a suspected obstacle strategy.

[0017] Further, when constructing a point cloud topology network based on the point cloud data and determining a spatial feature cluster by clustering, comprising:

[0018] removing isolated noise points in the point cloud data based on a bilateral filtering algorithm to determine effective point cloud data, and determining a neighborhood radius centered on each point of the effective point cloud data, the neighborhood radius being determined based on point cloud density;

[0019] searching all other points within a neighborhood radius of each point to determine whether the points are bidirectionally adjacent, if so, establishing an undirected edge between the point and the bidirectionally adjacent point, and determining the edge attribute of each undirected edge, otherwise, not establishing an undirected edge;

[0020] constructing all points and their corresponding undirected edges into the point cloud topology network, and determining the local connection density of each point, determining the initial clustering area based on the local connection density, and determining the reflection intensity gradient in each initial clustering area by traversing all undirected edges, and determining a plurality of reflection sub-areas with the same reflection intensity according to the relationship between the reflection intensity gradient and the average gradient;

[0021] determining the normal vector angle of all points in each reflection sub-area, and splitting the reflection sub-area based on the normal vector angle, and determining the splitting result as the spatial feature cluster.

[0022] Further, when extracting the spectral feature containing spectral data based on the spatial feature cluster, and determining the spatial spectral feature cluster based on the synchronous change of the spectral feature, comprising:

[0023] extracting the segment of the spectral data corresponding to all points in the spatial feature cluster, and comparing the acquisition time of each segment with the acquisition time of the corresponding point, eliminating the segments with time mismatch, and determining the overall main frequency offset, high frequency energy and main frequency change rate according to the remaining segments;

[0024] determining the synchronism in the spatial coordinate graph, the spatial distribution graph and the spatial position graph according to the overall main frequency offset, the high frequency energy and the main frequency change rate respectively, extracting the segment with the highest synchronism, and splitting the spatial feature cluster according to whether the corresponding points form a continuous region in space, and determining the splitting result as the spatial spectral feature cluster.

[0025] Further, when determining the effective echo spectrum according to the spectral space and spectral time of the spatial spectral feature cluster, comprising:

[0026] determining the connection relationship of all points in each connected component based on the point cloud topology network of the spatial spectral feature cluster, extracting the segment of the spectral data synchronously collected by all points in the spatial spectral feature cluster, and verifying whether the spectral feature of each segment in its connected component satisfies the transitivity association;

[0027] Extract all fragments that meet the transitivity association to construct a spectrum fragment set, extract the reflection intensity and high frequency energy proportion of each point according to the corresponding connection relationship of the connected component corresponding to the spectrum fragment set, and count the association direction of the reflection intensity and high frequency energy proportion of each point, determine the main frequency change trend based on the association direction, and retain the fragments that are consistent with the main frequency change trend and the point cloud motion direction to construct an effective spectrum set, and determine the effective echo spectrum based on the effective spectrum set.

[0028] Further, when determining whether there is obstacle information of the effective echo spectrum by comparing the effective echo spectrum with a historical spectrum database, it includes:

[0029] The historical spectrum database includes a plurality of historical effective echo spectra and a plurality of historical obstacle information, and each historical effective echo spectrum corresponds to a historical obstacle information;

[0030] When there is a historical effective echo spectrum identical to the effective echo spectrum in the historical spectrum database, the historical obstacle information corresponding to the historical effective echo spectrum is determined as the obstacle information, and it is determined that there is obstacle information of the effective echo spectrum;

[0031] When there is no historical effective echo spectrum identical to the effective echo spectrum in the historical spectrum database, it is determined that there is no obstacle information of the effective echo spectrum.

[0032] Further, when there is no obstacle information of the effective echo spectrum, an obstacle sample set is obtained and an obstacle recognition model is constructed, including:

[0033] Divide the obstacle sample set into a training set and a test set, and pre-select a convolutional neural network model;

[0034] Train the convolutional neural network model using the training set, and determine the prediction accuracy by substituting the test set into the trained convolutional neural network model;

[0035] When the prediction accuracy is greater than or equal to a prediction accuracy threshold, the trained convolutional neural network model is determined as the obstacle recognition model.

[0036] Further, when determining obstacle information based on the obstacle recognition model and the effective echo spectrum, and determining the effective obstacle of the driving route according to the obstacle information, it includes:

[0037] Substitute the effective echo spectrum into the obstacle recognition model to determine the obstacle information;

[0038] Determine the effective obstacle of the driving route according to the relationship between the obstacle information and the size of the unmanned mine gauge.

[0039] Compared with the prior art, the present application has the beneficial effects that: by adjusting the echo signal of the probe, the interference of noise is effectively filtered, laying a data foundation for subsequent analysis and processing, combining point cloud topology network clustering and spectral feature analysis, accurately distinguishing ground protrusions from real obstacles, solving the risk of over-clustering or under-clustering of irregular rock piles and low obstacles, thereby reducing the false detection and missed detection rates, ensuring the stability of effective identification of complex obstacles in the mining area, introducing spectral data for dynamic analysis, using the synchronous change of spectral features to construct a spatial spectral feature cluster, accurately capturing the differences between obstacles and the environment by comparing with the historical spectral database, dealing with dynamic obstacle scenarios such as rock rolling and temporary accumulation, and complex working conditions such as multi-vehicle intersection, determining obstacle information through effective echo spectrum, and accurately identifying effective obstacles, ensuring the driving safety of unmanned mining vehicles in complex environments, while reducing unnecessary deceleration or detour caused by misjudgment.

[0040] On the other hand, the present application also provides a kind of obstacle laser detection system for unmanned mining vehicle, for applying the above-mentioned obstacle laser detection method for unmanned mining vehicle, comprising:

[0041] The acquisition adjustment module is configured to emit radar probe signals to the driving route based on the laser radar and determine all probe echo signals, obtain the signal-to-noise ratio of all probe echo signals, determine the adjustment strategy of the corresponding probe echo signal based on the signal-to-noise ratio, and determine the echo time, spectral data and point cloud data based on the adjusted all probe echo signals;

[0042] The spatial spectrum module is configured to determine the constant driving strategy or the suspected obstacle strategy of the driving route according to the echo time, construct a point cloud topology network based on the point cloud data and determine a spatial feature cluster when the suspected obstacle strategy is determined, extract a spectral feature containing spectral data based on the spatial feature cluster, and determine a spatial spectral feature cluster based on the synchronous change of the spectral feature;

[0043] The obstacle analysis module is configured to determine an effective echo spectrum according to the spectral space and spectral time of the spatial spectral feature cluster, compare the effective echo spectrum with a historical spectral database to determine whether there is obstacle information of the effective echo spectrum, and obtain an obstacle sample set and construct an obstacle recognition model when there is no obstacle information of the effective echo spectrum;

[0044] The obstacle determination module is configured to determine obstacle information based on the obstacle recognition model and the effective echo spectrum, and determine the effective obstacle of the driving route according to the obstacle information.

[0045] It can be understood that the above-mentioned obstacle laser detection method and system for unmanned mining vehicles have the same beneficial effects, which will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0046] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments and are not meant to limit the present application. Furthermore, the same reference numerals in different drawings identify the same elements. In the drawings:

[0047] Figure 1 A flowchart of an obstacle laser detection method for unmanned mining vehicles provided by an embodiment of the present application;

[0048] Figure 2 A functional block diagram of an obstacle laser detection system for unmanned mining vehicles provided by an embodiment of the present application. DETAILED DESCRIPTION

[0049] Exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0050] Reference Figure 1 As shown in the drawings, in some embodiments of the present application, an obstacle laser detection method for unmanned mining vehicles includes:

[0051] S100: Based on the laser radar, radar detection signals are emitted to the driving route, all detection echo signals are determined, the signal-to-noise ratio of all detection echo signals is obtained, the adjustment strategy of the corresponding detection echo signal is determined based on the signal-to-noise ratio, and the echo time, spectral data and point cloud data are determined based on the adjusted all detection echo signals.

[0052] S200: Determine the constant driving strategy or suspected obstacle strategy of the driving route according to the echo time, when the suspected obstacle strategy is determined, construct the point cloud topology network based on the point cloud data and determine the spatial feature cluster by clustering, extract the spectral feature containing the spectral data based on the spatial feature cluster, and determine the spatial spectral feature cluster based on the synchronous change of the spectral feature.

[0053] S300: Determine the effective echo spectrum according to the spectrum space and spectrum time of the spatial spectrum feature cluster, compare the effective echo spectrum with the historical spectrum database to determine whether there is obstacle information of the effective echo spectrum, when there is no obstacle information of the effective echo spectrum, obtain the obstacle sample set and construct the obstacle recognition model.

[0054] S400: Determine the obstacle information based on the obstacle recognition model and the effective echo spectrum, and determine the effective obstacle of the driving route according to the obstacle information.

[0055] Specifically, the FMCW laser radar is deployed on the unmanned mining truck, and after the laser radar transmits radar detection signals to the driving route of the unmanned mining truck, it will receive all the detection echo signals reflected from the front. At this time, the signal quality is evaluated by determining the signal-to-noise ratio of the detection echo signal. The signal-to-noise ratio reflects the intensity ratio of the effective signal and the noise. For example, in the dust environment of the mining area, the echo of the distant rock may be scattered, resulting in a signal-to-noise ratio below 10dB, while the signal-to-noise ratio of the close-range vehicle can reach more than 30dB. Based on the difference in signal-to-noise ratio, the adjustment strategy is determined, so as to suppress the noise of the detection echo signal with low signal-to-noise ratio, and fine-tune the detection echo signal with high signal-to-noise ratio to avoid signal saturation. Finally, three types of key data are extracted from the adjusted detection echo signal: echo time, spectral data and point cloud data. The echo time reflects the distance between the unmanned mining truck and the detection target. For example, when the echo time increases by 1μs, the corresponding distance increases by 0.15m. The spectral data reflects the properties of the detection target, such as the material. For example, the spectral main frequency of metal equipment is concentrated in 5-10kHz, while the rock is in 10-20kHz. The point cloud data reflects the spatial form of the detection target, and each point contains three-dimensional coordinate information. Through the adjustment of the detection echo signal, the risk of signal quality decline caused by dust and strong light in the mining area is avoided, thereby ensuring the reliability of the subsequent analysis data and avoiding the interference of noise or distorted signals on the detection result.The driving strategy is determined based on the echo time. If it is said that the echo time of a plurality of continuous frames is within the allowed braking time range, it is indicated that there may be an obstacle in front of the unmanned mine truck. For example, the front 50 m of the unmanned mine truck is a flat ground, and the echo time of the flat ground is stable at 333 μs, corresponding to a detection distance of 50 m in front. If a certain frame of echo time suddenly decreases to 100 μs, the distance corresponds to 15 m, and the echo time decreases by 233 μs, indicating that the distance of the unmanned mine truck decreases by 35 m. It may be reflected by the obstacle such as the rock pile in front of 15 m. When it is determined that there may be an obstacle, a suspected obstacle strategy is executed. At this time, the point cloud data is combined to form a topological network reflecting the spatial connection relationship, and then the spatial feature cluster is obtained by clustering. Compared with the traditional grid map or density clustering, the topological network can more accurately distinguish continuous targets (such as rock piles) and discrete points (such as ground protrusions). For example, the point cloud of a rock pile is clustered into a continuous spatial feature cluster by the topological network, and the isolated points of the ground protrusion are clustered separately to avoid under-clustering or over-clustering. Then the spectral features of the spatial feature cluster are extracted, the spatial spectral feature cluster is determined by analyzing the synchronous change of the spectrum, and the targets with different materials but similar spatial forms are further distinguished. Through the spatio-temporal binding relationship between the point cloud data and the spectral data, the problem of distinguishing targets with similar spatial features but different materials is effectively solved. For example, the spatial features of the rock pile and the soil pile may be similar, but the high-frequency energy proportion of the rock is higher than that of the soil, and the synchronous change trend of the main frequency is more stable. Through spectral analysis, they can be accurately distinguished to avoid misjudgment of the soil pile as an obstacle, thereby improving the reliability and stability of obstacle detection.

[0056] Specifically, according to the spectrum space and spectrum time of the spatial spectrum feature cluster, the effective echo spectrum is screened out, the effective echo spectrum is a spectrum feature set of the detection echo signal detected by the laser radar, which can truly reflect the physical characteristics (such as material, surface state, and motion trend) of the obstacle and is verified after screening, is the core spectrum information retained after removing noise, interference and abnormal signals from the original spectrum data, and the effective echo spectrum is compared with the historical spectrum database. If the matching is successful, the obstacle information (category, size, etc.) is directly determined according to the historical spectrum database. If the matching fails, it indicates that the historical spectrum database does not exist obstacle information of the effective echo spectrum. Then, an obstacle sample set is obtained, which contains at least 1000 groups of sample data, covering the effective echo spectrum and the corresponding obstacle information under different dust conditions. Because the types of obstacles in the mining area are various (rocks, metal equipment, soil piles, pipelines, rolling stones, etc.), and new targets (such as new building materials temporarily stacked) may appear, although the obstacle sample set contains at least 1000 groups of sample data to cover common obstacles, there are still differences in the same type of obstacles (such as broken rocks and complete rocks), which makes it difficult for the obstacle sample set to cover all forms. Therefore, by learning the common characteristics of the same type of obstacles through a model, new targets or targets with changed forms can be generalized and recognized. For example, even if the sample set does not contain the effective echo spectrum of a certain type of pipeline, the model can determine it as a metal obstacle according to its main frequency and high-frequency energy ratio, thereby effectively determining the obstacle information.

[0057] It can be understood that, by topological network clustering and spectrum synchronization analysis, the risk of false detection and missed detection of irregular and low obstacles is avoided, the spatial features and material features are recognized by utilizing the spectrum dynamic change of the laser radar, the reliability of distinguishing rocks and soil, metal and non-metal is ensured, and whether the obstacle is passable and avoidable is determined according to the obstacle information, for example, low stones can be directly driven past, and stone piles on the ground are avoided. In addition, the obstacle information is determined according to the obstacle recognition model and the effective echo spectrum, the adaptive recognition of unknown obstacles is realized, the real-time demand of obstacle detection in the complex environment of the mining area is met, and a reliable guarantee is provided for the safe driving of the unmanned mining vehicle.

[0058] In some embodiments of the present application, when determining the adjustment strategy of the corresponding detection echo signal based on the signal-to-noise ratio, it includes: when the signal-to-noise ratio is greater than or equal to the signal-to-noise ratio threshold, performing phase compensation based on the Doppler frequency; when the signal-to-noise ratio is less than the signal-to-noise ratio threshold, extracting the complex signal of each detection echo signal in a short time window, determining the conjugate product of the previous detection echo signal, and extracting the phase difference to determine the phase difference sequence, performing phase unwrapping based on the phase difference sequence and accumulation to determine the accumulated phase, and performing phase compensation based on the accumulated phase.

[0059] Specifically, the phase of the probe echo signal will produce linear deviation due to the movement of the obstacle and the unmanned mine car, and will also produce random disturbance due to mine dust and clutter. The phase distortion will directly affect the extraction accuracy of spectral features such as main frequency and high-frequency energy ratio. The signal-to-noise ratio threshold is set to 15 dB, and the specific signal-to-noise ratio threshold can be dynamically set according to the scale and environmental pollution degree of the mine. In the mine environment, when the signal-to-noise ratio is greater than or equal to 15 dB, the effective component of the probe echo signal accounts for about 97%, while the noise and other interference (dust scattering, clutter) accounts for about 3%. The Doppler frequency (a key parameter reflecting target motion) can be stably extracted, and when the signal-to-noise ratio is less than 15 dB, the noise ratio will exceed 3%, and the extraction error of the Doppler frequency will increase, which cannot be directly used for compensation. At this time, the length of the short-time window is determined according to the sampling period of the laser radar, and the complex signal (retaining phase information) of the short-time window is intercepted from each frame of probe echo signal, so as to calculate the conjugate product of the previous frame signal to offset the common noise component, and then obtain the phase difference sequence. After phase unwrapping and accumulation, the phase wrapping situation is solved, so as to recover the phase change trend from the noise, and then offset the phase jitter caused by random noise. The differential phase compensation eliminates the phase distortion caused by target motion and noise interference, ensures the accuracy of the spectral data, and the differential phase compensation takes into account the compensation accuracy under different signal-to-noise ratios. In high signal-to-noise ratio, it is used to improve the compensation efficiency, and in low signal-to-noise ratio, it enhances the anti-noise ability, so that the spectral data after phase compensation can truly reflect the obstacle, and provide a reliable foundation for subsequent spatial spectral feature cluster construction and obstacle identification, and improve the robustness of laser detection in complex mine environment.

[0060] In some embodiments of the present application, in the constant driving strategy or suspected obstacle strategy according to the echo time to determine the driving route, the method comprises: obtaining a standard echo time interval, and comparing the standard echo time interval with the echo time. When there is an echo time greater than the left boundary of the standard echo time interval and less than the right boundary of the standard echo time interval, it is determined that the driving route is a constant driving strategy, otherwise, it is determined that the driving route is a suspected obstacle strategy.

[0061] Specifically, the standard echo time interval is determined according to the unmanned mine truck on the flat road without obstacles and the braking distance of the vehicle itself. If there is no obstacle in front of the unmanned mine truck on the driving path, the radar detection signal of the laser radar to the echo time is mainly determined by the reflection of the ground. At this time, the echo time obtained is the shortest or typical echo delay in normal driving, which reflects the detection distance under the condition that the front of the vehicle is unobstructed. The minimum safe parking distance of the unmanned mine truck in actual operation needs to avoid obstacles, which is determined by the vehicle speed, braking performance, road conditions and the like. If the distance corresponding to the echo time is less than this braking distance, the vehicle may not be able to avoid obstacles even if it is braked urgently, so the braking distance needs to be taken as a lower limit reference, thereby forming the standard echo time interval, the left boundary of which is determined by the echo time converted from the braking distance, and the right boundary of which is determined by the detection delay of the flat road without obstacles. Although the road surface in the mine area has natural undulations, the change of echo time is continuous and small under normal circumstances, and the presence of obstacles will cause a step change in echo time, and the interval boundary can capture this mutation. By determining the standard echo time interval, it is possible to avoid excessive reaction to the slight undulations of the road surface, and to accurately identify the abnormality caused by obstacles. When the driving route is determined as a constant driving strategy, the driving can continue, and when the driving route is determined as a suspected obstacle strategy, it indicates that there may be a suspected obstacle on the road surface, which needs to be further analyzed based on the point cloud data to ensure the reliability and stability of the obstacle detection.

[0062] In some embodiments of the present application, when the point cloud topology network is constructed based on the point cloud data and the spatial feature cluster is determined by clustering, the following steps are included: removing isolated noise points in the point cloud data based on a bilateral filtering algorithm to determine valid point cloud data, and determining a neighborhood radius based on the valid point cloud data, the neighborhood radius being determined based on the point cloud density; searching all other points within the neighborhood radius of each point to determine whether they are bidirectional neighbors, and if so, establishing an undirected edge between the point and the bidirectional neighbor, and determining the edge attribute of each undirected edge, otherwise, not establishing an undirected edge; constructing the point cloud topology network by connecting all points and their corresponding undirected edges, and determining the local connection density of each point; determining the initial clustering area based on the local connection density, and determining the reflection intensity gradient in each initial clustering area by traversing all undirected edges; determining a plurality of reflection sub-areas with the same reflection intensity according to the relationship between the reflection intensity gradient and the average gradient; determining the normal vector angle of all points in each reflection sub-area, and splitting the reflection sub-area based on the normal vector angle; and determining the splitting result as a spatial feature cluster.

[0063] Specifically, in the mining environment, the point cloud data of the laser radar can contain isolated noise points caused by dust scattering and sensor noise. These isolated noise points are not only spatially isolated, but also have no features with surrounding points. The range kernel of the bilateral filter can capture this difference and remove it, ensuring the reliability and stability of the effective point cloud data. The neighborhood radius is determined for each point in the effective point cloud data, which is a region radius centered on a point for searching neighboring points. The neighborhood radius is dynamically adjusted according to the point cloud density. The neighborhood radius is smaller when the point cloud density is higher, so as to avoid including irrelevant points. The neighborhood radius is larger when the point cloud density is lower, so as to ensure that enough neighboring points are captured. The point cloud density refers to the number of points per unit volume, which is an important attribute feature of the point cloud and reflects the spatial distribution and density of the point cloud. Generally, for a region with a high point cloud density (100 points per cubic meter), the domain radius is 0.08 m, and there are about 8 points within the domain radius. For a region with a low point cloud density (50 points per cubic meter), the domain radius is 0.11 m, and there are about 9 points within the neighborhood radius. By dynamically adjusting the domain radius, the neighboring points can be searched adaptively, ensuring the accuracy of the search. For each point, all other points within its neighborhood radius are searched to determine whether they are bidirectional neighbors. If point A is within the neighborhood of point B, and point B is also within the neighborhood of point A, it means that the spatial correlation between the two points is strong and they are not accidental neighbors. Then, it is determined that they are bidirectional neighbors and an undirected edge is established between them. The edge attributes are recorded, including the reflectance intensity and the normal vector angle, which lay the data foundation for subsequent analysis. Unidirectional neighbors may be accidental (e.g., a point falls on the edge of another point cloud), while bidirectional neighbors can ensure the stability of the connection. The point-cloud topology network reflects the spatial correlation through the point-edge relationship, providing a structured basis for clustering.

[0064] It can be understood that the local connection density is the number of undirected edges of a point divided by the volume of the neighborhood, which reflects the degree of dense connection of the point in space. The higher the value, the stronger the correlation with the surrounding points, and the more likely it belongs to the same obstacle entity. In the open-pit mine environment, the average value of the local connection density of the rock pile is 12 per cubic meter, and the minimum value is 8 per cubic meter. The average value of the local connection density of the isolated rock is 6 per cubic meter, and the minimum value is 4 per cubic meter (because the point cloud is sparse, but it is still higher than the ground 3 per cubic meter). The average value of the local connection density of the metal vehicle (truck, excavator) is 14 per cubic meter because the metal surface is smooth. Through the experimental data and simulation verification of the mine, the local connection density of different obstacle types can be determined, and the region where the point is located is divided according to the value of the local connection density of the point. Thus, an initial clustering region is formed. The initial clustering region reflects that it may belong to the same obstacle entity. In the initial clustering region, all undirected edges are traversed to determine the reflection intensity gradient. The reflection intensity gradient is the difference between the reflection intensities of the undirected edges divided by the distance between the two points, which reflects the reflection intensity change rate along the undirected edge direction. In the initial clustering region, the region where the reflection intensity gradient is less than or equal to twice the average gradient of the region is divided, such as the metal and rock interface. Thus, the reflection sub-region with the same reflection intensity is obtained, which ensures the consistency of the material in the reflection sub-region. In each reflection sub-region, the reflection sub-region is further split according to the angle between the normal vectors of the points. The normal vector angle reflects the surface flatness. A larger angle indicates that the surface is uneven, such as rock, and a smaller angle indicates that the surface is flat, such as a metal plate. If the angle between the normal vectors of the two points is greater than 30°, the region is further split, and the split result is determined as a spatial feature cluster. Otherwise, the spatial feature cluster is directly determined. The spatial feature cluster can reflect the spatial position, reflection characteristics and surface morphology of the obstacle, which lays a foundation for subsequent analysis combined with frequency spectrum data and ensures the reliability of obstacle detection.

[0065] In some embodiments of the present application, when the spatial feature cluster is extracted based on the spatial feature cluster, the frequency spectrum data is extracted, and the spatial frequency spectrum feature cluster is determined based on the synchronous change of the frequency spectrum feature. It includes: extracting the segment of the frequency spectrum data corresponding to all points in the spatial feature cluster, and comparing the acquisition time of each segment with the acquisition time of the corresponding point. The segments with time mismatch are removed, and the main frequency offset, high frequency energy and main frequency change rate are determined according to the remaining segments. The synchronicity in the spatial coordinate graph, spatial distribution graph and spatial position graph is determined according to the main frequency offset, high frequency energy and main frequency change rate. The segment with the highest synchronicity is extracted, and the spatial feature cluster is split according to whether the corresponding points form a continuous region in space. The split result is determined as a spatial frequency spectrum feature cluster.

[0066] Specifically, the spectral data contains the frequency characteristics (such as the main frequency and the like) of the radar echo signal. Due to the possible slight time difference between the collection of the point cloud data of the laser radar and the spectral data, by comparing the collection time of the segment and the collection time of the corresponding point, the mispositioned data (such as the segment of a certain spectrum actually corresponds to the point one second ago, which is irrelevant to the current point cloud) is eliminated, ensuring the spatiotemporal consistency of subsequent analysis. The main frequency offset refers to the difference between the main peak frequency extracted from the spectral data at a certain time and the determined reference frequency. The reference frequency is determined by real-time extraction of the stable component of the detection echo signal after adjusting the signal-to-noise ratio. The offset reflects the relative motion of the obstacle with respect to the sensor in the radial direction or the dynamic change of the obstacle surface / structure. The main frequency offset of a stationary or slowly changing obstacle is close to zero or fluctuates slightly, while the offset of an obstacle with significant radial velocity or structural change is larger. High-frequency energy refers to the total energy in a certain segment within a pre-set high-frequency band. The high-frequency band is usually set to 1 kHz or above, and can be adjusted according to the area and complexity of the mining area. The main frequency change rate reflects the speed of change of the main frequency over time. The change rate of a uniform motion object is stable between ±2 Hz / s, while the change rate of a variable speed object fluctuates greatly. The main frequency offset of all spectra in the spatial feature cluster is labeled on the spatial coordinate graph of the corresponding point cloud (with the x-axis as the point cloud horizontal coordinate and the y-axis as the main frequency offset), forming a scatter plot distribution of the offset and the spatial position. If the scatter points show a continuous one-way trend (such as gradually rising or falling from left to right without abrupt jumps), the synchronization is high. If the scatter points are randomly distributed (irregular up and down fluctuations), the synchronization is low. The high-frequency energy is labeled on the spatial distribution graph of the corresponding point cloud (the color depth represents the size of the high-frequency energy, usually red for higher high-frequency energy and blue for lower high-frequency energy). If the color shows a continuous gradual change area (such as gradually changing from red to blue from the cluster center to the edge without sudden red-blue alternation), the synchronization is high. If the color is randomly distributed (red and blue are mixed in the same small area), the synchronization is low.Draw a trend line of the main frequency change rate and the spatial position of the corresponding point cloud (each point corresponds to a main frequency change rate and a spatial position, connected by a broken line), if the trend line shows an orderly change rule (such as the main frequency change rate continuously increases or remains stable as the coordinate of the spatial position increases), the synchronization is high, if the trend line fluctuates sharply (sudden rise and sharp drop, irregular change), the synchronization is low, for the same spatial feature cluster, three types of high synchronization fragments are screened out, and whether the corresponding point cloud of the fragments can form a continuous region in space (such as the point cloud coordinates are not broken, forming a complete block / strip region) is observed, if a continuous region can be formed, the spatial feature cluster is retained and recorded as a spatial spectrum feature cluster, otherwise, the spatial feature cluster is split, and the split spatial feature cluster is determined as a spatial spectrum feature cluster, based on the spectrum data, further avoiding misjudgment of discrete entities as entities of the same obstacle, through space-time matching, multi-dimensional feature synchronization and spatial continuity verification, realizing comprehensive clustering from morphology clustering to morphology and spectrum joint clustering, and improving the fineness and accuracy of obstacle recognition in complex mine scenes.

[0067] In some embodiments of the present application, when determining the effective echo spectrum according to the spectrum space and spectrum time of the spatial spectrum feature cluster, the connection relationship of all points in each connected component is determined based on the point cloud topology network of the spatial spectrum feature cluster, the frequency spectrum data fragments of all points in the spatial spectrum feature cluster are extracted, and the frequency spectrum features of each fragment in the connected component are verified to determine whether they meet the transitivity correlation, all fragments meeting the transitivity correlation are extracted to construct a spectrum fragment set, the reflection intensity and high frequency energy proportion of each point in the connected component corresponding to the spectrum fragment set are extracted according to the corresponding connection relationship, and the correlation direction of the reflection intensity and the high frequency energy proportion of each point is counted, the main frequency change trend is determined based on the correlation direction, and the fragments consistent with the main frequency change trend and the point cloud motion direction are retained to construct an effective spectrum set, and the effective echo spectrum is determined based on the effective spectrum set.

[0068] Specifically, in the point cloud topology network, the connected component refers to the largest continuous point set connected by undirected edges (bidirectional adjacent relationship), and the largest continuous point set is an independent area composed of points associated with each other in the point cloud, and each connected component corresponds to a spatially continuous entity, such as a complete rock or the body of a mine car. For example: in the point cloud topology network, points A-B-C are connected by undirected edges, points D-E are connected by undirected edges, and A is not connected to D, then {A, B, C} and {D, E} are two independent connected components, respectively corresponding to two spatially separated entities. The transitive association reflects the indirect association of the spectral features within the connected component, for example: if the spectral features of point A and point B are similar (such as a main frequency offset difference less than 5 Hz), and the spectral features of point B and point C are similar, then the spectral features of point A and point C must also be similar (transitive association through B). This transitivity reflects the physical law of spatially continuous distribution of spectral features of the same entity (the spectral characteristics of different parts of the same object do not suddenly jump). For a spectral segment within a connected component, the verification process is as follows: select a point P in the connected component as a reference, record its spectral features, and traverse all directly adjacent points (points connected by undirected edges) of point P. If the spectral features of adjacent point Q and point P differ within a threshold range (such as a main frequency offset difference less than 5 Hz), then mark point Q as directly associated with point P. If all points in the connected component satisfy the consistency of any two points through transitive association, then the spectral features of the segment satisfy the transitive association. If there is a point whose transitive difference with the reference point exceeds the threshold range, then it is determined not to satisfy. By verifying the transitive association, false segments that are locally similar in spectrum but globally contradictory (such as accidental dust spectrum similar to rock, but cannot satisfy global transitivity) can be removed, further ensuring that the segment comes from the same entity. The high-frequency energy proportion is represented as the proportion of energy in the high-frequency band (above 1 kHz) in the total energy in the spectrum, reflecting the roughness of the surface of the obstacle (rock is greater than metal, rough surface scatters more high-frequency signals). According to the undirected edge connection relationship of the connected component, the reflection intensity and high-frequency energy proportion of each point are extracted in turn to form a correspondence between points, reflection intensity, and high-frequency energy, ensuring that the data comes from continuous points in the same connected component.

[0069] Specifically, the correlation direction refers to the trend relationship between the two, which has two typical cases, positive correlation: when the reflection intensity increases, the high-frequency energy proportion also increases (such as rough metal surface, higher reflection intensity and more high-frequency scattering), negative correlation: when the reflection intensity increases, the high-frequency energy proportion decreases (such as smooth metal surface, high reflection intensity but less high-frequency scattering), for the points in the connected component, the correlation coefficient of the reflection intensity and the high-frequency energy proportion is determined based on the Pearson correlation algorithm, when the correlation coefficient is greater than 0.6, it is determined as positive correlation direction, otherwise it is determined as negative correlation direction, the main frequency trend is the overall change direction of the main frequency in the connected component over time (increasing / decreasing / stable), from the correlation direction, if it is positive correlation, the main frequency of the rough surface is more significantly affected by the movement, if the mining card is close to the obstacle, the main frequency offset will increase over time, if it is negative correlation, the main frequency of the smooth surface is relatively stable, but when moving, if the mining card moves away from the obstacle, the main frequency offset will decrease over time. The movement direction of the mining card is determined by the difference calculation between point cloud frames (such as the coordinate change of two consecutive point clouds), if the main frequency trend matches the movement direction (such as the mining card moving forward, the main frequency offset increases over time), the frequency spectrum segment comes from the real entity of the movement, if it does not match (such as the mining card moving forward, but the main frequency offset decreases), it may be noise or false signal, thereby constructing an effective frequency spectrum set, the effective frequency spectrum set retains all the trend-consistent segments, and the comprehensive features (mean value of the frequency spectrum features) are the effective echo spectrum. The effective echo spectrum can truly reflect the physical properties and movement state of the obstacle.

[0070] It can be understood that, by removing the noise segments that are consistent locally but contradictory globally, it is ensured that the corresponding frequency spectrum comes from the same obstacle entity, so that the effective echo spectrum can accurately reflect the material, roughness and movement state of the obstacle, and the risk of obstacle identification error is reduced in a complex mine environment.

[0071] In some embodiments of the present application, when determining whether there is obstacle information of the effective echo spectrum by comparing the effective echo spectrum with the historical spectrum database, the method comprises: the historical spectrum database comprises a plurality of historical effective echo spectrums and a plurality of historical obstacle information, and each historical effective echo spectrum corresponds to a historical obstacle information, when there is a historical effective echo spectrum identical to the effective echo spectrum in the historical spectrum database, the historical obstacle information corresponding to the historical effective echo spectrum is determined as the obstacle information, and it is determined that there is the obstacle information of the effective echo spectrum, when there is no historical effective echo spectrum identical to the effective echo spectrum in the historical spectrum database, it is determined that there is no obstacle information of the effective echo spectrum.

[0072] Specifically, the historical spectrum database is a spectrum feature archive of common obstacles in the mining area (such as rocks, mine cars, metal equipment, soil piles, etc.), and each historical effective echo spectrum corresponds to unique historical obstacle information (such as type, size, material, etc.). The effective echo spectrum contains the core features of the current obstacle (main frequency offset, high frequency energy ratio, main frequency trend, etc.), and these features have a unique spectrum for each obstacle (such as rocks and mine cars). The spectrum features of different obstacles (such as rocks and mine cars) are different, and the spectrum features of the same type of obstacle are highly similar. Therefore, if there is a record in the historical spectrum database that is the same as the current effective echo spectrum, it means that the unmanned mine truck has encountered a known obstacle, and the historical information can be directly reused. If not, it is a new and unknown obstacle that needs to be processed by the subsequent model. Through the comparison of the historical spectrum database, there is no need to perform complex model reconstruction on each obstacle, thereby adapting to the real-time needs of the unmanned mine truck in the mining area. Moreover, the historical spectrum database has been verified for a long time, avoiding the risk of repeated identification errors of similar obstacles. When an unknown obstacle appears, its effective echo spectrum and confirmed obstacle information are stored in the historical spectrum database. Through data-driven means, the content of the historical spectrum database is continuously enriched, realizing rapid identification of obstacles and ensuring the accuracy and reliability of detection.

[0073] In some embodiments of the present application, when there is no obstacle information of the effective echo spectrum, the obstacle sample set is obtained and the obstacle recognition model is constructed, including: dividing the obstacle sample set into a training set and a test set, preselecting a convolutional neural network model, training the convolutional neural network model using the training set, and substituting the test set into the trained convolutional neural network model to determine the prediction accuracy. When the prediction accuracy is greater than or equal to the prediction accuracy threshold, the trained convolutional neural network model is determined as the obstacle recognition model.

[0074] Specifically, when the effective echo spectrum has no matching historical effective echo spectrum in the historical spectrum database, an obstacle sample set is obtained, which contains a large number of effective echo spectra of obstacles and related information (such as type, size, material, etc.) of the corresponding obstacles, and is divided into a training set with a proportion of 70%-80% and a test set with a proportion of 20%-30%. The training set is used to let the model learn the correlation between the spectrum features and the information of the obstacles, and the test set is used to verify the generalization ability of the model. A convolutional neural network model (CNN) is selected as the original model, which is good at processing data with local correlation (such as frequency distribution of spectrum, energy change trend). The spectrum features of the spectrum are extracted through the convolution layer, the redundant information is compressed through the pooling layer, and finally the prediction result of the obstacle information is output through the fully connected layer. In the training process, the training set is used to repeatedly iterate the convolutional neural network model, and the model adjusts the parameters (such as the size of the convolution kernel, the number of neurons) to minimize the prediction error; and after each training is completed, the test set is used to evaluate the prediction accuracy, and the prediction accuracy threshold is preferably 0.9. When the prediction accuracy is greater than or equal to the preset prediction accuracy threshold, it means that the model has mastered a stable identification rule, and it is determined as the obstacle identification model. Through the model learning the spectrum features of the obstacles, the problem of not being able to identify without historical records is solved, thereby covering new types of obstacles that suddenly occur in the mining area, such as rare rock shapes and temporarily intruding equipment. Moreover, taking the convolutional neural network model as the original model can automatically capture subtle features in the spectrum that are difficult to detect, such as small fluctuations in high-frequency energy, reduce the subjectivity of human feature extraction, and ensure the stability of the obstacle detection and identification results.

[0075] In some embodiments of the present application, when the obstacle information is determined based on the obstacle identification model and the effective echo spectrum, and the effective obstacle of the driving route is determined according to the obstacle information, it includes: substituting the effective echo spectrum into the obstacle identification model to determine the obstacle information, and determining the effective obstacle of the driving route according to the relationship between the obstacle information and the size of the unmanned mining truck.

[0076] Specifically, after the effective echo spectrum is input into the obstacle identification model, the model outputs obstacle information of the obstacle, including type (rock, mine car) and size (length, width, height) and other parameters. These information is the basic data for judging whether the obstacle affects driving. Based on the inherent size of the unmanned mine truck, a constraint condition is established. If the size of the obstacle invades the driving path of the unmanned mine truck, or its height exceeds the safety threshold for passing through the mine truck (for example, the obstacle is 4.2m high, and the chassis of the unmanned mine truck cannot pass from above), it is determined that the obstacle is effective. Otherwise, if the size of the obstacle is small (for example, a small stone with a diameter of 0.5m) and located on the roadside, it does not affect the passing of the mine truck, and is excluded as an invalid obstacle. According to the relationship between the obstacle information and the size of the unmanned mine truck, the effective obstacle of the driving route is determined, thereby filtering out invalid obstacles (such as small stones and low grass), thereby flexibly adapting the stability of detecting obstacles in complex environments such as mine areas, and ensuring the reliability and identification stability of obstacle detection.

[0077] In summary, the beneficial effects of the present application are that the signal-to-noise ratio of the detected echo signal is adjusted, effectively filtering out the interference of noise, laying a data foundation for subsequent analysis and processing, combining point cloud topology network clustering and spectral feature analysis, accurately distinguishing between ground protrusions and real obstacles, solving the risk of over-clustering or under-clustering of irregular rock piles and low obstacles, thereby reducing the false detection and missed detection rates, ensuring the stability of effective identification of complex obstacles in mine areas, introducing spectral data for dynamic analysis, using the synchronous change of spectral features to construct a spatial spectral feature cluster, accurately capturing the differences between obstacles and the environment by comparing with the historical spectral database, dealing with dynamic obstacle scenarios such as rock rolling and temporary accumulation, and complex working conditions such as multiple vehicle crossing, determining obstacle information through effective echo spectrum, and accurately identifying effective obstacles, ensuring the driving safety of unmanned mine trucks in complex environments, while reducing unnecessary deceleration or detour caused by misjudgment.

[0078] In another preferred mode based on the above embodiment, referring to Figure 2 The present embodiment provides an obstacle laser detection system for an unmanned mine truck for applying the above obstacle laser detection method for an unmanned mine truck, comprising:

[0079] The acquisition adjustment module is configured to emit a radar detection signal to the driving route based on the laser radar and determine all detection echo signals, obtain the signal-to-noise ratio of all detection echo signals, determine the adjustment strategy of the corresponding detection echo signal based on the signal-to-noise ratio, and determine the echo time, spectral data and point cloud data based on the adjusted all detection echo signals.

[0080] The spatial spectrum module is configured to determine a constant driving strategy or a suspected obstacle strategy of the driving route according to the echo time, construct a point cloud topology network based on the point cloud data and determine a spatial feature cluster by clustering when the suspected obstacle strategy is determined, extract a spectrum feature containing spectrum data based on the spatial feature cluster, and determine a spatial spectrum feature cluster based on the synchronous change of the spectrum feature.

[0081] The obstacle analysis module is configured to determine an effective echo spectrum according to the spectrum space and spectrum time of the spatial spectrum feature cluster, compare the effective echo spectrum with a historical spectrum database to determine whether there is obstacle information of the effective echo spectrum, and obtain an obstacle sample set and construct an obstacle recognition model when there is no obstacle information of the effective echo spectrum.

[0082] The obstacle determination module is configured to determine the obstacle information based on the obstacle recognition model and the effective echo spectrum, and determine an effective obstacle of the driving route according to the obstacle information.

[0083] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer usable program code.

[0084] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one or more flows and / or blocks. Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0085] These computer program instructions can also be stored in a computer readable storage medium that can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one or more flows and / or blocks. Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0086] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0087] Finally, it should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the technical solutions of the present application. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.

Claims

1. A method for obstacle laser detection for unmanned mine cars, characterized in that, The method comprises: emitting radar detection signals to a driving route based on a laser radar and determining all detection echo signals, obtaining signal-to-noise ratios of all detection echo signals, determining an adjustment strategy of a corresponding detection echo signal based on the signal-to-noise ratios, and determining echo time, spectral data and point cloud data based on the adjusted all detection echo signals; determining a constant driving strategy or a suspected obstacle strategy of the driving route according to the echo time, constructing a point cloud topology network based on the point cloud data and determining a spatial feature cluster when the suspected obstacle strategy is determined, extracting spectral features containing spectral data based on the spatial feature cluster, and determining a spatial spectral feature cluster based on synchronous changes of the spectral features; determining an effective echo spectrum according to the spectral space and spectral time of the spatial spectral feature cluster, comparing the effective echo spectrum with a historical spectrum database to determine whether there is obstacle information of the effective echo spectrum, and obtaining an obstacle sample set and constructing an obstacle recognition model when there is no obstacle information of the effective echo spectrum; determining obstacle information based on the obstacle recognition model and the effective echo spectrum, and determining an effective obstacle of the driving route according to the obstacle information; when constructing a point cloud topology network based on the point cloud data and determining a spatial feature cluster, comprising: determining effective point cloud data by removing isolated noise points in the point cloud data based on a bilateral filtering algorithm, and determining a neighborhood radius based on each point of the effective point cloud data, wherein the neighborhood radius is determined based on point cloud density; searching all other points within the neighborhood radius of each point to determine whether they are bidirectional neighbors, if so, establishing an undirected edge between the point and the bidirectional neighbor, and determining the edge attribute of each undirected edge, otherwise, not establishing an undirected edge; constructing the point cloud topology network by connecting all points with their corresponding undirected edges, determining the local connection density of each point, determining the initial clustering area based on the local connection density, and determining the reflection intensity gradient in each initial clustering area by traversing all undirected edges, and determining a plurality of reflection sub-regions with the same reflection intensity according to the relationship between the reflection intensity gradient and the average gradient; determining the normal vector angle of all points in each reflection sub-region, splitting the reflection sub-region based on the normal vector angle, and determining the splitting result as the spatial feature cluster.

2. The method for obstacle laser detection for unmanned mine cars of claim 1, wherein, when determining the adjustment strategy of the corresponding detection echo signal based on the signal-to-noise ratio, comprising: when there is a signal-to-noise ratio greater than or equal to a signal-to-noise ratio threshold, performing phase compensation based on Doppler frequency; when there is a signal-to-noise ratio less than the signal-to-noise ratio threshold, extracting a complex signal of each detection echo signal in a short time window, determining a conjugate product with a previous detection echo signal, extracting a phase difference to determine a phase difference sequence, performing phase unwrapping based on the phase difference sequence and accumulation to determine an accumulated phase, and performing phase compensation based on the accumulated phase.

3. The method for obstacle laser detection for unmanned mine cars of claim 2, wherein, when determining the constant driving strategy or the suspected obstacle strategy of the driving route according to the echo time, comprising: obtaining a standard echo time interval, and comparing the standard echo time interval with the echo time; When there is an echo time greater than the left boundary of the standard echo time interval and less than the right boundary of the standard echo time interval, it is determined that the driving route is a constant driving strategy, otherwise, it is determined that the driving route is a suspected obstacle strategy.

4. The method for obstacle laser detection for unmanned mine cars of claim 3, wherein, In the method, the spatial feature cluster is extracted based on the spatial feature cluster, and the spatial spectrum feature cluster is determined based on synchronous changes of the spectrum features. The spectrum data segments corresponding to all points in the spatial feature cluster are extracted, and the collection time of each segment is compared with the collection time of the corresponding point. The segments with time mismatch are removed, and the main frequency offset, high frequency energy and main frequency change rate are determined based on the remaining segments. The synchronicity in the spatial coordinate graph, spatial distribution graph and spatial position graph is determined based on the main frequency offset, high frequency energy and main frequency change rate, the segment with the highest synchronicity is extracted, the spatial feature cluster is split based on whether the corresponding points form a continuous region in space, and the split result is determined as the spatial spectrum feature cluster.

5. The method for obstacle laser detection for unmanned mine cars of claim 4, wherein, In the method, the effective echo spectrum is determined based on the spectrum space and spectrum time of the spatial spectrum feature cluster. Based on the point cloud topology network of the spatial spectrum feature cluster, the connection relationship of all points in each connected component is determined, the spectrum data segments of all points in the spatial spectrum feature cluster are extracted, and the spectrum features of each segment in the connected component are verified to meet the transitivity association. The spectrum segment set is constructed by extracting all segments meeting the transitivity association, the reflection intensity and high frequency energy proportion of each point are extracted based on the corresponding connection relationship of the connected component of the spectrum segment set, the association direction of the reflection intensity and high frequency energy proportion of each point is counted, the main frequency change trend is determined based on the association direction, the segments consistent with the main frequency change trend and the point cloud motion direction are retained to construct the effective spectrum set, and the effective echo spectrum is determined based on the effective spectrum set.

6. The method for obstacle laser detection for unmanned mine cars of claim 5, wherein, In the method, the effective echo spectrum is compared with the historical spectrum database to determine whether there is obstacle information of the effective echo spectrum. The historical spectrum database includes a plurality of historical effective echo spectrums and a plurality of historical obstacle information, and each historical effective echo spectrum corresponds to a historical obstacle information. When there is a historical effective echo spectrum identical to the effective echo spectrum in the historical spectrum database, the historical obstacle information corresponding to the historical effective echo spectrum is determined as the obstacle information, and it is determined that there is obstacle information of the effective echo spectrum. When there is no historical effective echo spectrum identical to the effective echo spectrum in the historical spectrum database, it is determined that there is no obstacle information of the effective echo spectrum.

7. The method for obstacle laser detection for unmanned mine cars of claim 6, wherein, When there is no obstacle information of the effective echo spectrum, an obstacle sample set is obtained and an obstacle recognition model is constructed, including: The obstacle sample set is divided into a training set and a test set, and a convolutional neural network model is preselected; The convolutional neural network model is trained using the training set, and the test set is substituted into the trained convolutional neural network model to determine the prediction accuracy; When the prediction accuracy is greater than or equal to a prediction accuracy threshold, the trained convolutional neural network model is determined as the obstacle identification model.

8. The method for obstacle laser detection for unmanned mine cars of claim 7, wherein, When the obstacle information is determined based on the obstacle identification model and the effective echo spectrum, and the effective obstacle of the driving route is determined according to the obstacle information, comprising: The effective echo spectrum is substituted into the obstacle identification model to determine the obstacle information. The effective obstacle of the driving route is determined according to the relationship between the obstacle information and the size of the unmanned mine truck.

9. A barrier laser detection system for a driverless mine truck for applying the barrier laser detection method for a driverless mine truck as claimed in any one of claims 1 to 8, characterized in that, Comprising: The acquisition adjustment module is configured to emit radar detection signals to the driving route based on the laser radar and determine all detection echo signals, obtain the signal-to-noise ratio of all detection echo signals, determine the adjustment strategy of the corresponding detection echo signal based on the signal-to-noise ratio, and determine the echo time, spectrum data and point cloud data based on the adjusted all detection echo signals; The spatial spectrum module is configured to determine the constant driving strategy or the suspected obstacle strategy of the driving route according to the echo time, when the suspected obstacle strategy is determined, construct a point cloud topology network based on the point cloud data and perform clustering to determine a spatial feature cluster, extract a spectrum feature containing spectrum data based on the spatial feature cluster, and determine a spatial spectrum feature cluster based on the synchronous change of the spectrum feature; The obstacle analysis module is configured to determine the effective echo spectrum according to the spectrum space and spectrum time of the spatial spectrum feature cluster, compare the effective echo spectrum with a historical spectrum database to determine whether there is obstacle information of the effective echo spectrum, when there is no obstacle information of the effective echo spectrum, obtain an obstacle sample set and construct an obstacle identification model; The obstacle determination module is configured to determine the obstacle information based on the obstacle identification model and the effective echo spectrum, and determine the effective obstacle of the driving route according to the obstacle information.

Images