Downhole trackless rubber-tyred vehicle front perception domain attached type abnormal echo discrimination and control method

Abstract

The application discloses a kind of wellbore trackless rubber-tired vehicle forward sensing field attachment type abnormal echo discrimination and control method, first acquisition and pre-process laser radar point cloud, forward camera image and vehicle state information;Select forward key sensing field and divide into several sectors, extract sector stable echo distance and near end morphological characteristics;Establish real forward barrier and protective cover local attachment abnormal double hypothesis, combined with motion consistency, scanning angle locking feature discriminant sector state;Suspected sector executes local visual negative evidence review;Evaluate continuous credible braking corridor, output adaptive control instruction and dynamic update.The system includes corresponding functional module.The application solves the problem of abnormal echo misjudgment caused by local attachment of laser radar protective cover under wet working condition in wellbore, realizes reliable distinction between abnormal echo and real barrier, simultaneously combined with actual credible space output fine control strategy, improves driving safety and operation continuity of wellbore trackless rubber-tired vehicle.

Description

Technical Field

[0001] This invention belongs to the field of sensing and control technology for trackless rubber-tired vehicles in underground mines, and particularly relates to a method for identifying and controlling abnormal echoes of the forward sensing domain of trackless rubber-tired vehicles in underground mines. Background Technology

[0002] Trackless rubber-tired vehicles are core transportation equipment in underground mining processes, such as coal mines. The reliability of their forward collision avoidance sensing directly affects the safety of underground operations. These vehicles typically operate under harsh conditions, including wet conditions, dust suppression sprays, water accumulation in tunnels, and mud splashes. To achieve forward obstacle detection, they are usually equipped with multiple sensors such as lidar, forward-facing cameras, and millimeter-wave radar. Based on the environmental information collected by these sensors, they execute warnings, speed limits, or braking controls to ensure driving safety.

[0003] In existing technologies, solutions for handling sensor anomalies in underground trackless rubber-tired vehicles mainly fall into two categories: one type of solution addresses the problem of dust, mud, rainwater, and other contaminants adhering to the surface of the lidar protective cover, light-transmitting cover, or viewing window, determining whether the sensor is contaminated based on echo intensity, echo quantity, multi-frame statistical results, or the distribution of anomalies; the other type of solution focuses on multi-sensor fusion, ensuring driving safety by adjusting fusion weights, switching to a degradation mode, or directly limiting speed and braking when the performance of a certain sensor degrades.

[0004] However, the above solutions still have significant shortcomings in the wet operation environment of trackless rubber-tired vehicles underground. When mud adheres to the protective cover of a mechanically rotating lidar, it often doesn't mean the entire radar fails; rather, it results in a continuous, stable, and recurring abnormal echo within a specific local area that does not conform to the normal movement patterns of the vehicle. For forward collision avoidance sensing, such echoes are easily misinterpreted as real forward obstacles, leading to false emergency stops, repeated speed limits, or unreasonable vehicle downgrade control, affecting underground transportation efficiency and potentially causing secondary safety hazards due to frequent misoperations.

[0005] The main shortcomings of existing technologies include at least the following two aspects: First, most solutions can determine whether "the radar is dirty" or "a certain local area is abnormal", but it is difficult to further distinguish whether the local abnormality is the attached noise on the surface of the sensor protective cover or a real obstacle in the forward critical safety perception domain. Therefore, it is difficult to complete the filtering of misjudgments for driving decision-making. Second, after identifying perception degradation, most solutions usually adopt global, vehicle-level or experience-based processing, such as uniform downweighting, fixed speed limit or direct switching of downgrade mode. They do not combine the current vehicle speed, braking demand and the actual credibility of the forward critical perception domain for fine control, which can easily lead to overly conservative or unsafe control.

[0006] Therefore, how to reliably identify local fixed abnormal echoes in the forward critical sensing domain when mud adheres to the protective cover of the mechanical rotating lidar in the forward collision avoidance sensing of the trackless rubber-wheeled vehicle in the well, so as to avoid misjudging them as real forward obstacles, and output reasonable speed limiting or braking control strategies based on the actual credibility of the forward critical space, has become an urgent technical problem to be solved. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for identifying and adaptively controlling abnormal echoes in the forward key sensing domain of an underground trackless rubber-tired vehicle. This solves the problem of misjudging abnormal echoes caused by local adhesion of the lidar protective cover under underground wet working conditions, enables reliable differentiation between abnormal echoes and real obstacles, and improves driving safety and operational continuity.

[0008] Technical solution: The present invention provides a method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain, comprising the following steps:

[0009] S1. Real-time acquisition of lidar point cloud, forward-facing camera images, and vehicle status information, and unified preprocessing; the vehicle status information includes at least vehicle speed and yaw rate.

[0010] S2. Based on the vehicle's current driving direction, vehicle body envelope, forward collision avoidance requirements, and braking safety-related spatial range, select the forward key perception domain from the lidar scanning range.

[0011] S3. Divide the forward key sensing domain into several angular sectors, and extract stable echo distance and near-end echo morphology features in each sector; the near-end echo morphology features include near-end aggregation degree, radial thickness, multi-beam coverage and connectivity extension features with adjacent sectors.

[0012] S4. For each sector, simultaneously establish the assumption of real forward obstacles and the assumption of local attachment anomaly of the protective shield; under the assumption of real forward obstacles, predict the theoretical distance change of the sector based on the vehicle's motion state; under the assumption of local attachment anomaly of the protective shield, examine the characteristics of near-end echoes being fixed at ultra-close distances, repeating continuously for multiple frames, and not conforming to the vehicle's motion laws, and determine whether the anomaly repeatedly appears at similar angle positions in continuous scanning.

[0013] S5. Based on motion consistency, scanning angle locking features and proximal morphological features, the attachment anomaly is identified for each sector, the sector state is divided, and the state upgrade is achieved through a state machine.

[0014] S6. For sectors that are located within the overlapping area of ​​the forward camera's field of view and are in a suspected attachment state or where attachment and obstacles are difficult to distinguish, perform local visual negative evidence verification when the image quality meets the conditions.

[0015] S7. Calculate the minimum safe braking distance based on the vehicle's current speed, system delay, and available braking capacity, and assess whether there is a continuous and reliable braking corridor that meets the vehicle width and braking requirements within the forward critical perception domain.

[0016] S8. Based on the effective depth, effective width and whether there are real obstacles inside the continuous reliable braking corridor that cannot be eliminated, output normal operation, speed limit downgrade or emergency braking control commands.

[0017] S9. Repeat steps S1 to S8 to dynamically update the forward key sensing domain, sector status, continuous reliable braking corridor and control results, forming an adaptive sensing and control closed loop.

[0018] Furthermore, in step S1, point cloud preprocessing includes invalid point removal, outlier removal, initial screening of nearby stray points, and time synchronization; image preprocessing includes brightness checking, sharpness checking, and glare or overexposure area identification; and vehicle status data undergoes time alignment and smoothing processing.

[0019] Furthermore, in step S3, the stable echo distance is extracted as follows: first, locally dense near-end candidate point clusters are screened within the sector, and then the stable echo distance is calculated from the candidate point clusters; the stable echo distance is any one of the median distance, truncated mean distance, or quantile distance.

[0020] Furthermore, in step S4, the determination of motion consistency is achieved by calculating the motion consistency residual; the motion consistency residual is the absolute value of the stable echo distance of the sector at the current moment and the theoretical distance under the assumption of a real stationary obstacle; the theoretical distance is the distance obtained after vehicle motion compensation.

[0021] Furthermore, in step S4, the method for determining the scanning angle locking feature is as follows: statistical analysis is performed on the near-end echo subdivision angle position of the suspected sector in multiple consecutive scanning cycles. When the dispersion of the angle position sequence is less than a preset angle dispersion threshold, or the proportion of scanning cycles falling into the preset angle window is greater than a preset proportion threshold, the suspected sector is determined to have scanning angle locking features.

[0022] Furthermore, in step S6, the method for reviewing local visual negative evidence is as follows: the spatial region corresponding to the suspected sector is projected onto the image to form a local region of interest, and the region is checked to see if there are occlusion boundaries, entity outlines or texture interruptions that should be present in a real obstacle; if the image maintains continuous texture and there are no clear obstacle boundaries, the attachment anomaly judgment is enhanced; if there are stable entity outlines or occlusion relationships, the real obstacle judgment is maintained.

[0023] Furthermore, in step S7, the determination condition for a continuous reliable braking corridor is: from the front of the vehicle to the boundary of the minimum safe braking distance, each depth layer has a laterally continuous reliable interval, and the available width of the reliable interval is not less than the sum of the vehicle width and the lateral safety margin, while adjacent depth layers maintain spatial connectivity.

[0024] Furthermore, in step S7, the formula for calculating the minimum safe braking distance is:

[0025]

[0026] Where D is the minimum safe braking distance. The vehicle's current speed. The total delay in perception, decision-making, and execution. For the currently available deceleration, For additional safety margin; this calculation formula is used to illustrate the minimum forward depth requirement that a continuously reliable braking corridor must achieve.

[0027] This invention also discloses a forward sensing domain adhesion-type abnormal echo discrimination and control system for trackless rubber-tired vehicles in underground mines, comprising:

[0028] LiDAR is used to collect point cloud data of the area in front of and around the vehicle.

[0029] A forward-facing camera is used to capture image information of the area in front of the vehicle;

[0030] The vehicle status acquisition unit is connected to the vehicle chassis system and is used to acquire vehicle speed, yaw rate, and optional braking status, steering status, or wheel speed information.

[0031] The data preprocessing unit is used to perform time alignment, filtering, and validity processing on point clouds, images, and vehicle states;

[0032] The forward key perception domain selection module is used to select the spatial area directly related to the forward driving safety of the current vehicle from the LiDAR scanning range;

[0033] The sector observation and morphological feature extraction module is used to divide the forward key sensing domain into sectors and extract stable echo distance, near-endpoint aggregation degree, radial thickness, multi-beam coverage, and connectivity extension features with adjacent sectors.

[0034] The motion consistency prediction module is used to predict the theoretical distance change of the sector based on the vehicle's motion state under the assumption of real obstacles.

[0035] The scanning angle lock recognition module is used to determine whether a proximal abnormality repeatedly appears at a similar angle position in continuous scanning;

[0036] An attachment anomaly state discrimination module is used to integrate motion consistency, scan angle locking features and proximal morphological features, output sector status and realize state upgrade through state machine;

[0037] The image quality gating module is used to determine whether an image meets the local verification conditions;

[0038] The local visual negative evidence verification module is used to verify the continuous texture, occlusion boundary and local entity evidence of the image region corresponding to the suspected sector;

[0039] The Continuous Reliable Braking Corridor Assessment Module is used to construct a forward brakeable passage corridor based on the current spatial distribution of reliable sectors and to determine whether the continuous reliable braking corridor criteria are met.

[0040] The control decision module is used to output normal operation, speed limit, or braking control commands based on the continuous reliable braking corridor and the actual obstacle conditions.

[0041] Furthermore, the lidar is a mechanically rotating lidar; the forward key perception domain preferably covers the braking channel in front of the vehicle and its safety margin, excluding the side and rear areas that are less related to the current forward decision.

[0042] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0043] 1. This invention does not simply determine whether the lidar is dirty, nor does it directly perform global degradation processing after detecting degradation. Instead, it focuses on the forward critical sensing domain and distinguishes the abnormal echoes caused by local mud adhesion from the actual forward obstacles online. This significantly reduces the risk of false emergency stops and false speed limits caused by local mud adhesion, ensuring the continuous operating efficiency of the trackless rubber-wheeled vehicle in the well.

[0044] 2. The anomaly detection of the present invention no longer relies solely on a single distance residual, but instead combines motion consistency, scanning angle locking characteristics, and near-end spatial morphology characteristics for joint judgment. This is more in line with the actual physical mechanism of local adhesion of mechanical rotating lidar protective covers, and is more targeted and accurate than general dirt detection or conventional motion compensation judgment.

[0045] 3. The visual information in this invention adopts a condition-triggered local negative evidence verification method instead of performing conventional image detection on all sectors. This avoids the structural contradiction of a single forward-facing camera covering the entire radar sector, improves the practical usability under weak visual conditions downhole, and reduces the redundant processing cost of visual information.

[0046] 4. The control strategy in this invention no longer relies on the simple effective sector ratio, but makes decisions based on whether there is a continuous and reliable braking corridor that meets the requirements of vehicle width and braking depth. This makes the control action directly correspond to the vehicle's current forward traffic safety, reducing unnecessary conservative degradation while ensuring safety. It is more suitable for application scenarios where underground trackless rubber-tired vehicles are sensitive to continuous operation and forward collision avoidance. Attached Figure Description

[0047] Figure 1 This is a block diagram of the system structure of the present invention;

[0048] Figure 2 This is a flowchart of the method of the present invention;

[0049] Figure 3 This is a schematic diagram of the forward critical sensing domain and continuous reliable braking corridor of the present invention.

[0050] Figure 4 This is a schematic diagram of the projection of the abnormal sector to a local area of ​​the image in this invention. Detailed Implementation

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0052] Figure 2 The flowchart shown is a diagram of the method of this invention, clearly illustrating the nine steps of the method, which are sequentially connected to form a closed-loop control. Step S1 is data acquisition and preprocessing, providing a foundation for subsequent discrimination; steps S2-S5 are the core process of abnormal echo discrimination, enabling the differentiation between attachment anomalies and real obstacles; step S6 is visual verification, further validating the discrimination results; steps S7-S8 are control decisions, outputting adaptive control commands; and step S9 is dynamic updating, ensuring the system adapts to changes in operating conditions.

[0053] The first step involves the system acquiring LiDAR point clouds, forward-facing camera images, and vehicle status information in real time, followed by unified preprocessing. Vehicle status information includes at least vehicle speed and yaw rate. Point cloud preprocessing includes invalid point removal, outlier removal, initial screening of nearby stray points, and time synchronization; image preprocessing includes brightness checks, sharpness checks, and glare or overexposure area identification; vehicle status data undergoes time alignment and smoothing. The purpose of this step is not merely general filtering, but to establish a unified spatiotemporal reference for subsequently distinguishing between "real forward obstacles" and "protective shield-attached anomalous echoes."

[0054] The second step involves selecting a key forward perception domain within the lidar scanning range. This domain does not make a uniform judgment across the entire field of view. Instead, it considers the vehicle's current driving direction, vehicle envelope, forward collision avoidance requirements, and the spatial range related to braking safety. This key forward perception domain preferably covers the braking path in front of the vehicle and its safety margin, while excluding the rearward and side areas, which are less relevant to the current forward decision-making, from the core discrimination range. This reduces the interference of sidewalls, support structures, and lateral clutter on forward decision-making.

[0055] The third step involves dividing the forward key sensing domain into several angular sectors, and extracting stable echo distance and near-end echo morphology features within each sector. Preferably, dense near-end candidate point clusters are first selected from the sector point cloud, and then the stable echo distance is calculated from these candidate clusters to avoid occasional jump-point interference caused by directly taking the minimum value. Simultaneously, the near-end aggregation degree, radial thickness, multi-beam coverage, and connectivity extension features with adjacent sectors are extracted to reflect whether the sector echo is closer to a "spatial solid obstacle" or a "near-end attached anomaly."

[0056] The fourth step involves establishing both a real forward obstacle assumption and a local attachment anomaly assumption for each sector. Under the real forward obstacle assumption, based on the vehicle's current speed, yaw rate, and sector observations from the previous moment or several moments prior, the theoretical distance change that the sector should exhibit at the current moment should a real stationary obstacle exist. Under the local attachment anomaly assumption, the focus is on examining whether the near-end echo exhibits characteristics such as ultra-close-range fixation, repeated occurrence across multiple frames, and non-compliance with vehicle motion patterns. For mechanically rotating lidar, further examination is conducted to determine whether the anomaly repeatedly appears at similar subdivision angle positions during continuous scanning, in order to identify the scan angle locking phenomenon unique to attachment anomalies.

[0057] The fifth step involves determining attachment anomalies in each sector based on motion consistency, scan angle locking characteristics, and near-end morphological features. If the measured stable distance of a sector consistently deviates significantly from the theoretical distance under the assumption of a real obstacle, and the near-end echo of that sector exhibits very short distances, minimal fluctuations, concentration in a local angular domain, weak expansion of adjacent sectors, and repeated appearances at similar angular positions in continuous scanning, then that sector is more likely to be a local attachment anomaly of the protective cover. To avoid misjudgments caused by single-frame jitter, a sector state machine is preferably set up to divide the sector state into normal state, suspected attachment state, confirmed attachment state, and real obstacle state; only when the attachment anomaly criteria are consistently met within several consecutive sampling periods is it upgraded to the confirmed attachment state.

[0058] Step 6: For sectors located within the overlapping field of view of the forward-facing camera and suspected of being attached or where attachment is difficult to distinguish from obstacles, a local visual review is performed. This visual review does not perform routine target detection uniformly on all sectors. Instead, provided the image quality meets the requirements, the spatial region corresponding to the suspected sector is projected onto the image to form a region of interest (ROI). Priority is given to checking whether this region contains occlusion boundaries, entity contours, or texture interruptions that should be present in a real obstacle. If the texture of the tunnel floor or tunnel wall remains continuous in the image and there are no clear obstacle boundaries, the visual result is used to enhance the attachment anomaly judgment. If there are stable entity contours, occlusion relationships, or local target evidence in the image, the sector is maintained as a real obstacle and no anomaly is removed. For sectors where the image is unusable, the field of view does not overlap, or the projection is unreliable, the visual review is not enforced, and the main discrimination path based on the spatiotemporal features of the point cloud is retained.

[0059] The seventh step involves calculating the minimum safe braking distance based on the vehicle's current speed, system latency, and available braking capacity, and assessing within the forward critical perception domain whether a continuous, reliable braking corridor exists that meets the vehicle's width and braking requirements. Unlike simply counting the number of effective sectors, this invention focuses on whether a continuous, reliable forward space still exists in front of the vehicle, covering the vehicle's passage width and extending beyond the current braking distance. Even if some edge sectors fail, the vehicle can still maintain restricted operation as long as a continuous, reliable space in the center meets the passage and braking requirements; conversely, if the continuous, reliable space in front of the central axle is cut off, even if there are many remaining effective sectors, the current forward perception is not considered safe.

[0060] Step 8: Based on the effective depth and width of the continuous reliable braking corridor, and whether there are any real obstacles within it that cannot be removed, output normal operation, speed limit downgrade, or emergency braking control. If there is a continuous reliable braking corridor in the forward direction that meets the minimum safe braking distance requirement, and no real obstacles are identified within the corridor, the vehicle maintains normal operation; if the continuous reliable braking corridor is insufficient to support safe braking at the current speed, but can still support lower speeds, output speed limit downgrade control; if the continuous reliable braking corridor is interrupted, its effective depth is severely insufficient, or there are real obstacles within the corridor that cannot be removed, output emergency braking or stopping control.

[0061] The ninth step involves continuously repeating the above process. As vehicle speed, direction, local attachment status of the protective cover, and image availability change, the forward key perception domain, sector status, continuous reliable braking corridor, and control results are all dynamically updated, thus forming an online adaptive perception and control closed loop for harsh downhole environments.

[0062] The system of the present invention is as follows Figure 1The system includes: a lidar, a forward-facing camera, a vehicle status acquisition unit, a data preprocessing unit, a forward key perception domain selection module, a sector observation and morphological feature extraction module, a motion consistency prediction module, a scan angle lock-in recognition module, an attachment anomaly status discrimination module, an image quality gating module, a local visual negative evidence verification module, a continuous reliable braking corridor evaluation module, and a control decision module.

[0063] The lidar is used to collect point cloud data of the environment in front of and around the vehicle; the forward-facing camera is used to collect image information in front of the vehicle; the vehicle status acquisition unit is connected to the vehicle chassis system to acquire vehicle speed, yaw rate, and optional braking status, steering status, or wheel speed information; the data preprocessing unit is used to perform time alignment, filtering, and validity processing on the point cloud, image, and vehicle status.

[0064] The forward key perception domain selection module is used to select spatial areas directly related to the forward driving safety of the current vehicle from the LiDAR scanning range; the sector observation and morphological feature extraction module is used to divide the area into sectors and extract stable echo distance, near-end point aggregation degree, radial thickness, multi-beam coverage, and connectivity extension features with adjacent sectors; the motion consistency prediction module is used to predict the theoretical distance change of the sector based on the vehicle's motion state under the assumption of real obstacles; the scan angle lock recognition module is used to determine whether a near-end anomaly repeatedly appears at a similar angle position in continuous scanning; the attachment anomaly state discrimination module is used to output the sector state by integrating the above information; the image quality gating module is used to determine whether the image meets the local verification conditions; the local visual negative evidence verification module is used to verify the continuous texture, occlusion boundary, and local entity evidence of the image area corresponding to the suspected sector; the continuous reliable braking corridor evaluation module is used to construct a forward brakeable passage corridor based on the current reliable sector spatial distribution; and the control decision module is used to output normal operation, speed limit, or braking control commands based on the continuous reliable braking corridor and the real obstacle situation.

[0065] Figure 1 The diagram shown is a system structure block diagram of the present invention. The system includes a lidar 1, a forward-facing camera 2, a vehicle status acquisition unit 3, a data preprocessing unit 4, a forward key perception domain selection module 5, a sector observation and morphological feature extraction module 6, a motion consistency prediction module 7, a scan angle lock-in recognition module 8, an attachment anomaly state discrimination module 9, an image quality gating module 10, a local visual negative evidence verification module 11, a continuous reliable braking corridor evaluation module 12, a control decision module 13, and a vehicle actuator 14. All modules work collaboratively to achieve attachment-type anomaly echo discrimination and adaptive control. The output of the control decision module 13 is connected to the vehicle actuator 14 and is used to control the vehicle to perform normal operation, speed limiting, or braking actions.

[0066] Figure 3The diagram shows the forward key perception domain and continuous reliable braking corridor of the present invention. A lidar is installed on the vehicle body 21 (radar installation position 22). The forward key perception domain 23 is the area in front of the vehicle from -60° to +60° and with a depth of 0 to 30 m. This area is divided into ordinary sectors 24 and suspected attachment anomaly sectors 25. The continuous reliable braking corridor 26 is a continuous spatial area within the forward key perception domain that meets the requirements of vehicle width and braking depth. The minimum safe braking distance boundary 27 is the forward boundary of the continuous reliable braking corridor, ensuring that the vehicle has sufficient braking space.

[0067] Figure 4 The diagram shows the projection of the abnormal sector to a local area of ​​the image according to the present invention. The spatial area 31 corresponding to the suspected abnormal sector is located within the field of view 32 of the forward-facing camera. The spatial area is projected onto the image plane 33 to form a local region of interest (ROI) 34. The ROI is divided into a continuous texture area 36 and a physical boundary or obstacle evidence area 35. By determining whether there is physical boundary or obstacle evidence in the area, the local visual negative evidence verification is achieved.

[0068] Key technical details:

[0069] (1) Regarding the extraction of stable echo distance within a sector. This invention does not recommend directly using the minimum distance within a sector as the criterion. Instead, it is preferable to first screen locally dense candidate point clusters located near the edge of the sector, and then calculate the stable echo distance from these candidate point clusters. The stable echo distance can be achieved using any one of the median distance, truncated mean distance, or quantile distance. The purpose of this is to suppress the amplification effect of individual outlier noise points, occasional reflection points, and scattering points on the discrimination results.

[0070] (2) Regarding the core difference between real obstacles and attachment anomalies. This invention argues that real forward obstacles are physical targets in the scene space, and their distance evolution should be consistent with the vehicle's movement, and they usually have a certain spatial expansion in adjacent sectors; while the local attachment of mud to the protective cover is a sensor-related anomaly, and its echo is more likely to show ultra-close range, small fluctuations, local angular concentration, weak expansion in adjacent sectors, and repeated appearance at similar angular positions during mechanical rotation scanning. This difference is the basis for the mechanism discrimination in this invention.

[0071] (3) Regarding motion consistency residuals. In a preferred embodiment, the stable echo distance of the k-th sector at the current time can be denoted as... The theoretical distance of this sector under the assumption of a real stationary obstacle is denoted as... and with As a motion consistency residual, among which This is the current observation distance. This is the theoretical distance obtained after vehicle motion compensation. In practical applications, this single residual is not required to be used as the sole criterion. Instead, it is used in conjunction with scan angle locking features and proximal morphological features to avoid making rigid judgments based on only a single threshold.

[0072] (4) Regarding scanning angle locking characteristics. For mechanically rotating lidar, when mud adheres to a fixed position of the protective cover, abnormal echoes will usually reappear at similar subdivision angle positions in consecutive scanning cycles. Therefore, this invention preferably records the trend of subdivision angle position changes of the near-end echo of the suspected sector and uses "repeated appearance at similar angle positions in multiple consecutive cycles" as important evidence of adhesion anomalies. Even if a real forward obstacle is observed in a certain angular domain, its position will usually evolve with the movement of the vehicle and changes in spatial relationships, making it difficult to maintain this angle position lock in the sensor coordinate system for a long time. In a preferred embodiment, the subdivision angle positions of the near-end echo of the suspected sector in multiple consecutive scanning cycles can be statistically analyzed, and the presence of scanning angle locking characteristics can be determined based on the dispersion of the subdivision angle position sequence. Specifically, the subdivision angle positions of the corresponding near-end echo in N consecutive scanning cycles can be used as the angle position sequence. When the dispersion of the angle position sequence is less than a preset angle dispersion threshold, or the proportion of scanning cycles falling into a preset angle window is greater than a preset proportion threshold, the suspected sector is determined to have scanning angle locking characteristics. The angular discrete threshold and preset angular window can be set according to the lidar angular resolution, installation posture, protective cover structure, and on-site calibration results. The purpose of this approach is to further transform the "repeated appearance of multiple consecutive loops at similar angular positions" into a quantifiable engineering criterion, thereby enhancing the stability of identifying local attachment anomalies of the protective cover, without limiting the scanning angle lock to a single fixed algorithm form.

[0073] (5) Regarding the visual verification method. In this invention, the forward-facing camera is not used as a unified primary criterion for all sectors. Instead, local negative evidence verification is performed when the image quality meets the conditions and the suspected sector is in the field of view overlap area. The so-called negative evidence verification prioritizes the search for image evidence that "should appear if there is a real obstacle, but does not appear at present," such as obvious occlusion boundaries, entity outlines, texture interruptions, or local structural abrupt changes. If the corresponding image area maintains continuous texture and lacks entity boundary evidence, it is more supportive of the attachment anomaly judgment.

[0074] (6) Regarding Continuous Reliable Braking Corridors. Unlike simply statistically analyzing the percentage of effective sectors, this invention defines a continuous, reliable spatial region within the forward critical sensing domain that meets the vehicle width and braking depth requirements as a continuous reliable braking corridor. As long as this corridor exists, it indicates that the current forward sensing still has the capability to support restricted operation; conversely, if the corridor is destroyed in front of the central axis, it means that the forward space truly directly related to vehicle braking safety is no longer reliable. In a preferred embodiment, a continuous reliable braking corridor requires not only meeting the vehicle width requirement at a single depth section, but also forming a connected region composed of continuous reliable sectors from the front of the vehicle to the minimum safe braking distance boundary. Specifically, within the forward critical sensing domain, the distribution of reliable sectors at each depth layer can be searched along the forward depth direction; when each depth layer has a laterally continuous reliable interval from the front of the vehicle to the minimum safe braking distance boundary, and the available width of this reliable interval is not less than the sum of the vehicle width and the lateral safety margin, while adjacent depth layers maintain spatial connectivity, a continuous reliable braking corridor is determined to exist. If only a local depth layer meets the width condition, but continuous connectivity cannot be maintained between the preceding and following depth layers, then a continuous and reliable braking corridor is not considered to have been formed. The purpose of this approach is to avoid misjudging "partial passability" as "overall safe braking passability," ensuring that control decisions directly correspond to the true state of the vehicle's forward continuous braking space.

[0075] (7) Regarding the minimum safe braking distance. In a preferred embodiment, the minimum safe braking distance can be calculated by multiplying the current vehicle speed by the total system delay, and then adding the braking distance and safety margin under the current braking capacity. The expression can be written as:

[0076]

[0077] Where D is the minimum safe braking distance. The vehicle's current speed. The total delay in perception, decision-making, and execution. For the currently available deceleration, For additional safety margin. This expression describes the minimum forward depth requirement that a continuously reliable braking corridor must achieve.

[0078] (8) Regarding parameter settings. The distance threshold, angular position discrete threshold, state transition duration frames, image quality threshold, and braking margin in this invention are not limited to unique fixed values, but can be set according to vehicle size, radar installation location, roadway width, vehicle speed range, braking performance, and on-site calibration results. This ensures that the solution has clear engineering applicability and avoids unnecessarily limiting the scope of patent protection to overly narrow numerical conditions.

[0079] Example: Theoretical calculation example under typical downhole conditions

[0080] This embodiment uses a trackless rubber-tired vehicle with a width of 2.8m and a lateral safety margin of 0.5m, and a set of mechanically rotating lidar as examples. Under the assumption of typical underground wet operation conditions, the calculation process, discrimination process, and control output of the method of the present invention are explained. The parameters in the embodiment are exemplary parameters used to illustrate the implementation of the technical solution of the present invention in typical scenarios, and can be adjusted according to vehicle model, sensor configuration, installation location, and on-site calibration results.

[0081] Assume the vehicle is currently traveling straight ahead. According to the current "Coal Mine Safety Regulations" regarding the operation of trackless rubber-tired vehicles underground, the operating speed of these vehicles should not exceed 25 km / h when transporting personnel, and should not exceed 40 km / h when transporting materials. The safe operating distance between vehicles traveling in the same direction should not be less than 50 m. This embodiment illustrates the calculation process of the present invention under typical low-speed underground conditions, exemplarily taking the vehicle's current speed v = 2.5 m / s (approximately 9 km / h), which is lower than the aforementioned upper speed limit for trackless rubber-tired vehicles transporting personnel and materials underground.

[0082] Meanwhile, let the total system delay be... = 0.6 s, currently available deceleration = 2.0 m / s², additional safety margin = 3.0 m, the key forward perception domain of the lidar is selected as the area in front of the vehicle from -60° to +60° and with a depth of 0 to 30m.

[0083] S1 Camera Calibration and Image Correction, Point Cloud and Vehicle Status Preprocessing. The system acquires the LiDAR point cloud, forward-facing camera image, and vehicle status information at the current time t, where the vehicle speed is 2.5 m / s and the yaw rate is approximately 0 rad / s. After time synchronization, filtering, and basic correction of the point cloud, image, and vehicle status, a unified data input that can be used for subsequent discrimination is obtained.

[0084] S2 selects the key forward perception domain. Combining vehicle envelope, forward collision avoidance requirements, and braking safety requirements, the key forward perception domain is selected from the full field of view of the LiDAR, namely the area in front of the vehicle from -60° to +60° and with a depth of 0 to 30 m. This area mainly covers the vehicle's current braking path and its safety margin range, while areas to the sides and rear that are less relevant to the current forward safety decision are not included in the core discrimination range.

[0085] S3 Sector Observation and Proximal Morphological Feature Extraction. The forward key sensing domain is divided into 12 sectors, each with an angle width of 10°. For each sector, locally dense candidate proximal point clusters are first selected, and then the stable echo distance is calculated from these candidate point clusters, preferably using the median distance as the stable echo distance. Simultaneously, the proximal point clustering degree, radial thickness, and connectivity extension features of adjacent sectors are extracted. Taking the k = 7th sector located near the front as an example, the central angle of this sector is approximately +5°, and the stable echo distance at the current moment is denoted as... = 1.2 m. This sector has a high degree of aggregation near the end point, a small radial thickness, and weak extension to adjacent sectors, showing preliminary characteristics of a near-end attachment anomaly.

[0086] S4 establishes a dual-hypothesis discrimination basis. For each sector, both the "real forward obstacle hypothesis" and the "protective shield local attachment anomaly hypothesis" are established simultaneously. Under the real forward obstacle hypothesis, based on the vehicle's current speed, yaw rate, and the sector's observation results from the previous moment, the theoretical distance change for that sector at the current moment if a real stationary obstacle exists is predicted. Taking sector k=7 as an example, under the simplified conditions of an approximately zero yaw rate and a stationary forward target corresponding to that sector, if the observed distance at the previous moment was 1.8 m and the sampling period was 0.1 s, the theoretical distance after motion compensation can be written as:

[0087] = 1.8 - 2.5 × 0.1 = 1.55 m

[0088] Under the assumption of localized abnormal attachment of the protective cover, the focus is on examining whether the sector exhibits echo characteristics that are extremely close-range fixed, with small fluctuations, locally concentrated, and do not conform to the normal movement patterns of the vehicle.

[0089] S5 Attachment Anomaly Detection (including scan angle locking criterion). Calculate motion consistency residuals for sector k = 7:

[0090]

[0091] This residual indicates a significant deviation between the current observed distance and the theoretical distance under the assumption of a truly stationary obstacle. Simultaneously, the subdivided angular positions of the near-end echoes in this sector are statistically analyzed over N = 5 consecutive scan cycles. An exemplary angular position sequence could be:

[0092] [5.1°, 5.0°, 5.2°, 5.1°, 5.0°]

[0093] The aforementioned angular position sequence exhibits low dispersion, with an exemplary standard deviation of approximately 0.08°, which is less than the preset angular dispersion threshold of 0.5°. The percentage of scan cycles falling within the preset angular window (e.g., ±0.5°) is 100%, exceeding the preset percentage threshold of 80%. Simultaneously, the proximal morphological features of this sector also satisfy conditions such as close proximity, low fluctuation, and weak expansion of adjacent sectors.

[0094] Based on the joint judgment of motion consistency residuals, scan angle locking features, and near-end morphological features, the sector state machine can upgrade sector k = 7 from "suspected attachment state" to "confirmed attachment state". The remaining sectors remain in normal state or actual obstacle state according to their respective observation characteristics. The above thresholds are only exemplary parameters in this embodiment and can be adjusted according to radar angular resolution, installation attitude, vehicle size, and on-site calibration results.

[0095] S6 Local Visual Negative Evidence Verification. Since sector k = 7 is located within the overlapping area of ​​the forward-facing camera's field of view and has entered a suspected or confirmed attachment state, local visual negative evidence verification is triggered. The system projects the spatial region corresponding to this sector into the image, forming a local region of interest (ROI), and performs a local inspection of this region.

[0096] In this embodiment, the roadway surface texture within the ROI area remains continuous, with no obvious occlusion boundaries, interruptions in entity outlines, or abrupt changes in local structure, lacking substantial evidence to support the existence of a real obstacle. Therefore, this local visual result serves as negative evidence, further enhancing the judgment of local attachment anomalies of the protective cover and preventing the anomalous echo from being considered a real forward obstacle in subsequent safety decisions.

[0097] S7 constructs a continuous and reliable braking corridor. First, based on the minimum safe braking distance calculation formula:

[0098]

[0099] Substituting the parameters into this embodiment, we get:

[0100] D = 2.5 × 0.6 + (2.5)² / (2 × 2.0) + 3 = 1.5 + 1.5625 + 3 = 6.0625 m

[0101] That is, the minimum safe braking distance is approximately 6.06 m.

[0102] Subsequently, within the forward critical sensing domain, the distribution of reliable sectors at each depth layer within the range of 0–6.06 m is searched along the depth direction. Based on the lateral projection range of the corresponding reliable sectors at each depth layer, a laterally continuous reliable interval is determined. For example, the following results can be obtained:

[0103] Within the depth range of 0–2 m, there exists a laterally continuous reliable interval with an available width of approximately 3.8 m; within the depth range of 2–4 m, there exists a laterally continuous reliable interval with an available width of approximately 3.9 m; and within the depth range of 4–6.06 m, there exists a laterally continuous reliable interval with an available width of approximately 3.7 m.

[0104] The lateral continuity and reliability intervals of each of the aforementioned depth layers are all greater than the sum of the vehicle width and the lateral safety margin, i.e., 2.8 + 0.5 = 3.3 m, and adjacent depth layers maintain spatial connectivity without significant interruptions. Therefore, it can be determined that a continuous and reliable braking corridor exists within the current forward critical perception domain that meets the requirements for vehicle passage width and braking depth.

[0105] S8 outputs control commands. Since there is a continuous and reliable braking corridor in the forward direction that meets the minimum safe braking distance requirement, and no real obstacles that cannot be eliminated are identified in the corridor, the control decision module outputs a normal operation command, and the vehicle continues to travel at the current speed without triggering speed limit or braking control.

[0106] S9 Dynamic Updates. In the next instant, as the vehicle continues to move forward and the protective cover attachment status may change, the forward key perception domain, sector status, continuous reliable braking corridor, and control results are all updated in real time, thus forming an online adaptive perception and control closed loop.

[0107] As can be seen from the above theoretical calculation examples, when local mud adhesion to the protective cover causes a fixed abnormal echo in the sector directly in front, the present invention can identify this type of adhesion-type abnormal echo through scanning angle locking discrimination, local visual negative evidence verification, and continuous reliable braking corridor judgment, and output corresponding control results accordingly, thereby demonstrating that the method of the present invention is feasible and engineering applicable under this typical downhole working condition.

Claims

1. A method for identifying and controlling anomaly echoes of adhesion type in the forward sensing domain of an underground trackless rubber-tired vehicle, characterized in that, Includes the following steps: S1. Real-time acquisition of lidar point cloud, forward-facing camera images, and vehicle status information, and unified preprocessing; the vehicle status information includes at least vehicle speed and yaw rate. S2. Based on the vehicle's current driving direction, vehicle body envelope, forward collision avoidance requirements, and braking safety-related spatial range, select the forward key perception domain from the lidar scanning range. S3. Divide the forward key sensing domain into several angular sectors, and extract stable echo distance and near-end echo morphology features in each sector; The near-end echo morphological characteristics include the near-end point aggregation degree, radial thickness, multi-beam coverage, and connectivity and extension characteristics with adjacent sectors; S4. For each sector, simultaneously establish the assumption of real forward obstacles and the assumption of local attachment anomaly of the protective shield; under the assumption of real forward obstacles, predict the theoretical distance change of the sector based on the vehicle's motion state; under the assumption of local attachment anomaly of the protective shield, examine the characteristics of near-end echoes being fixed at ultra-close distances, repeating continuously for multiple frames, and not conforming to the vehicle's motion laws, and determine whether the anomaly repeatedly appears at similar angle positions in continuous scanning. S5. Based on motion consistency, scan angle locking features and proximal morphological features, the attachment anomaly is identified for each sector, the sector state is divided, and the state upgrade is achieved through a state machine. S6. For sectors that are located within the overlapping area of ​​the forward camera's field of view and are in a suspected attachment state or where attachment and obstacles are difficult to distinguish, perform local visual negative evidence verification when the image quality meets the conditions. S7. Calculate the minimum safe braking distance based on the vehicle's current speed, system delay, and available braking capacity, and assess whether there is a continuous and reliable braking corridor that meets the vehicle width and braking requirements within the forward critical perception domain. S8. Based on the effective depth, effective width and whether there are real obstacles inside the continuous reliable braking corridor that cannot be eliminated, output normal operation, speed limit downgrade or emergency braking control commands. S9. Repeat steps S1 to S8 to dynamically update the forward key sensing domain, sector status, continuous reliable braking corridor and control results, forming an adaptive sensing and control closed loop.

2. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S1, point cloud preprocessing includes invalid point removal, outlier removal, initial screening of nearby stray points, and time synchronization; image preprocessing includes brightness checking, sharpness checking, and glare or overexposure area identification. Vehicle status data undergoes time alignment and smoothing.

3. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S3, the stable echo distance is extracted as follows: first, locally dense near-end candidate point clusters are screened within the sector, and then the stable echo distance is calculated from the candidate point clusters; the stable echo distance is any one of the median distance, truncated mean distance, or quantile distance.

4. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S4, the determination of motion consistency is achieved by calculating the motion consistency residual; the motion consistency residual is the absolute value of the stable echo distance of the sector at the current moment and the theoretical distance under the assumption of a real stationary obstacle; the theoretical distance is the distance obtained after vehicle motion compensation.

5. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S4, the method for determining the scanning angle locking feature is as follows: statistical analysis is performed on the near-end echo subdivision angle position of the suspected sector in multiple consecutive scanning cycles. When the dispersion of the angle position sequence is less than the preset angle dispersion threshold, or the proportion of scanning cycles falling into the preset angle window is greater than the preset proportion threshold, the suspected sector is determined to have the scanning angle locking feature.

6. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S6, the method for reviewing local visual negative evidence is as follows: the spatial region corresponding to the suspected sector is projected onto the image to form a local region of interest, and the region is checked to see if there are occlusion boundaries, entity outlines or texture interruptions that should be present in a real obstacle; if the image maintains continuous texture and there are no clear obstacle boundaries, the attachment anomaly judgment is enhanced; if there are stable entity outlines or occlusion relationships, the real obstacle judgment is maintained.

7. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S7, the determination condition for a continuous reliable braking corridor is: from the front of the vehicle to the boundary of the minimum safe braking distance, each depth layer has a laterally continuous reliable interval, and the available width of the reliable interval is not less than the sum of the vehicle width and the lateral safety margin, while adjacent depth layers maintain spatial connectivity.

8. The method for identifying and controlling anomaly echoes of a trackless rubber-tired vehicle in the forward sensing domain according to claim 1, characterized in that, In step S7, the formula for calculating the minimum safe braking distance is: ； Where D is the minimum safe braking distance. The vehicle's current speed. The total delay in perception, decision-making, and execution. For the currently available deceleration, For additional safety margin; this calculation formula is used to illustrate the minimum forward depth requirement that a continuously reliable braking corridor must achieve.

9. A forward sensing domain adhesion-type abnormal echo discrimination and control system for underground trackless rubber-tired vehicles, characterized in that, To implement the method as described in claim 1, comprising: LiDAR is used to collect point cloud data of the area in front of and around the vehicle. A forward-facing camera is used to capture image information of the area in front of the vehicle; The vehicle status acquisition unit is connected to the vehicle chassis system and is used to acquire vehicle speed, yaw rate, and optional braking status, steering status, or wheel speed information. The data preprocessing unit is used to perform time alignment, filtering, and validity processing on point clouds, images, and vehicle states; The forward key perception domain selection module is used to select the spatial area directly related to the forward driving safety of the current vehicle from the LiDAR scanning range; The sector observation and morphological feature extraction module is used to divide the forward key sensing domain into sectors and extract stable echo distance, near-endpoint aggregation degree, radial thickness, multi-beam coverage, and connectivity extension features with adjacent sectors. The motion consistency prediction module is used to predict the theoretical distance change of the sector based on the vehicle's motion state under the assumption of real obstacles. The scanning angle lock recognition module is used to determine whether a proximal abnormality repeatedly appears at a similar angle position in continuous scanning; An attachment anomaly state discrimination module is used to integrate motion consistency, scan angle locking features and proximal morphological features, output sector status and realize state upgrade through state machine; The image quality gating module is used to determine whether an image meets the local verification conditions; The local visual negative evidence verification module is used to verify the continuous texture, occlusion boundary and local entity evidence of the image region corresponding to the suspected sector; The Continuous Reliable Braking Corridor Assessment Module is used to construct a forward brakeable passage corridor based on the current spatial distribution of reliable sectors and to determine whether the continuous reliable braking corridor criteria are met. The control decision module is used to output normal operation, speed limit, or braking control commands based on the continuous reliable braking corridor and the actual obstacle conditions.

10. The system according to claim 9, characterized in that, The lidar is a mechanically rotating lidar; the forward key perception domain preferably covers the braking channel in front of the vehicle and its safety margin, excluding the side and rear areas that are less related to the current forward decision.

Images