A Differentiated Collision Avoidance Control Method and System for Sprinkler Trucks Based on Target Classification

Abstract

This invention belongs to the field of automation and control technology for municipal sanitation vehicles, specifically relating to a differentiated obstacle avoidance control method and system for sprinkler trucks based on target classification. The control method includes simultaneously acquiring image data and point cloud data of the environment surrounding the sprinkler truck; constructing a target detection and classification model, and transmitting the image data to the trained target detection and classification model; processing the target detection and classification model and inputting it into a pre-fusion algorithm with the point cloud data; when a pedestrian is detected, analyzing whether there is an effective occupant; when a second type of target is detected, initiating pre-collaborative safety; when both first and second types of targets are detected simultaneously, generating corresponding avoidance strategies in parallel and independently, and if there are conflicts between avoidance strategies for different targets, fusion is performed according to an arbitration principle. The system of this invention possesses a deep understanding of the environment and refined decision-making capabilities, can simulate human driver judgment, and has a high degree of intelligence.

Description

Technical Field

[0001] This invention belongs to the field of automation and control technology of municipal sanitation vehicles, specifically relating to a differentiated avoidance control method and system for sprinkler trucks based on target classification. Background Technology

[0002] Water trucks are key equipment for cleaning and suppressing dust on urban roads. To achieve automation, existing water trucks have begun to integrate basic pedestrian detection functions.

[0003] However, existing sprinkler truck systems mostly employ a single strategy: stopping spraying as soon as a pedestrian is detected. This leads to unnecessary and frequent interruptions to spraying operations in scenarios where there is no actual risk of splashing, such as when pedestrians are behind bus stops or trees where there is no actual risk of splashing, significantly reducing operational efficiency. Therefore, the system lacks sufficient intelligence; it lacks a deep understanding of the environment and refined decision-making capabilities, failing to simulate the situational judgments of human drivers, such as recognizing obstacles that can provide shelter or anticipating the avoidance needs of different targets. Furthermore, current technologies primarily focus on pedestrians, neglecting the large number of bicycles, electric bikes, and other two- or three-wheeled vehicles on the road. These road users are also susceptible to water splashes and may lose control due to sudden spraying, causing secondary accidents. Current avoidance logic fails to cover these critical scenarios.

[0004] In view of this, a complete method and system is proposed that can execute differentiated, non-single avoidance strategies in sprinkler truck operation scenarios based on target classification results. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a differentiated avoidance control method and system for sprinkler trucks based on target classification. It is specifically designed for sprinkler truck operation scenarios. It uses multi-sensor fusion data and target classification technology to identify pedestrians and non-motorized vehicles on the road and executes a differentiated sprinkler avoidance method based on the target identification type.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: On the one hand, the present invention provides a differentiated avoidance control method for sprinkler trucks based on target classification, including: Image and point cloud data of the surrounding environment of the sprinkler truck are collected synchronously at a set frequency. An object detection and classification model was built based on the YOLOv5 algorithm, and the model was trained by setting a dataset. Image data was then transmitted to the trained object detection and classification model for processing. The target bounding box, category, and confidence score of each category in the image coordinate system obtained after processing by the target detection and classification model are fused with the point cloud data before inputting the point cloud data to obtain the position coordinates and velocity vectors with at least the labels of the first and second types of targets. The first type of target is pedestrians. When a pedestrian is detected, the system analyzes whether there are any effective obstructions. If so, it triggers cooperative passage and then determines whether the distance between the water truck and the pedestrian reaches the verification distance. If the target is reached, the system will identify whether the pedestrian shows a valid intention to avoid the obstacle. If the pedestrian shows a valid intention to avoid the obstacle, the system will maintain cooperative passage and the pedestrian will pass through the section of road. Otherwise, the water truck will stop spraying and give way. If there is no effective cover, the water truck will stop spraying and give way. The second type of target is non-motorized vehicles. When the second type of target is identified, the pre-collaborative safety mechanism is activated. When both types of targets are identified simultaneously, the corresponding avoidance strategies are generated in parallel and independently. Then, it is determined whether there is a conflict between the avoidance strategies of different targets. If there is, the strategy fusion is performed according to the arbitration principle; otherwise, the corresponding avoidance strategy is executed.

[0007] Furthermore, the set frequency is a frame rate / frequency greater than or equal to 30Hz.

[0008] Furthermore, the specific process for analyzing whether an effective shield exists is as follows: The pre-fusion algorithm can also obtain the position coordinates and velocity vectors with the third type of target label, where the third type of target is the occluder. After identifying pedestrians and occluders, a semantic segmentation model or an instance segmentation model is used to perform pixel-level segmentation of the pedestrian region and the occluder region in the image to generate an accurate contour mask. The pedestrian contour mask is then projected onto a virtual plane perpendicular to the line vector connecting the vehicle's centroid and the pedestrian's centroid, and the width of the maximum bounding rectangle of the projection is calculated. Similarly, project the outline mask of the occluder onto the same virtual plane as the pedestrian, and calculate the maximum continuous width that the occluder projection can cover. ; Calculate pedestrian position under projection Location of the shelter Connect the points and calculate the distance between this line and the pedestrian's line of sight from the water truck. The included angle θ; like > * K and θ are less than the set threshold If the width of the obstruction is determined to be valid, then K is considered a redundancy coefficient greater than 1, which can be taken as 1.2. It is 30 degrees Celsius.

[0009] Furthermore, the collaboration is specifically achieved through: The system controls the audible and visual alarm to sound an alert, while simultaneously controlling the water pressure regulating mechanism to reduce the water pressure to 50%~70% of the calibrated value, and controlling the nozzle angle adjusting motor to deflect the main water stream 10~15 degrees away from pedestrians.

[0010] Furthermore, the verification distance Calculated dynamically using the following formula: in, This refers to the instantaneous speed of the water truck. The total response time of the water truck's differentiated obstacle avoidance control system is set to 200-500 milliseconds. The effective water jet radius under the current pump pressure; The initial velocity of the water flow at the nozzle. P is the water spray pressure. The density of water is given, but in practice, due to nozzle shape and frictional losses, a velocity coefficient is introduced. Adjustments are made, with values ​​ranging from 0.85 to 0.98; The elevation angle of the sprinkler head; The fixed safety buffer distance is 2-5 meters; h is the distance from the sprinkler head to the ground.

[0011] Furthermore, identifying whether a pedestrian exhibits an effective avoidance intention includes: focusing on a real-time image sequence of the pedestrian area, and quantifying the pedestrian's avoidance intention by analyzing the following features. If all of the following features are met, the pedestrian is considered to have exhibited an effective avoidance intention; otherwise, the pedestrian has no avoidance intention, and the sprinkler truck will stop spraying and give way. Feature 1: By tracking the point cloud with millimeter-wave radar, it is calculated whether the centroid of a pedestrian has produced a continuous displacement toward the direction of an effective shield, wherein the continuous displacement is equal to or greater than 0.5 meters; Feature 2: The point cloud tracking data is converted to the camera coordinate system. The displacement distance of the pedestrian's centroid in two adjacent frames is obtained through millimeter-wave radar, thereby obtaining the pedestrian's instantaneous velocity vector. It is then determined whether the angle θ between the instantaneous velocity direction and the line connecting the pedestrian (centroid) and the occluder (centroid) is less than 30 degrees. If it is less than 30 degrees, it is considered that the pedestrian has an effective avoidance tendency.

[0012] Furthermore, the aforementioned pre-collaborative security specifically refers to: When the dynamic distance between the sprinkler truck and the second type of target reaches the warning distance When the time comes, the sound and light alarm will sound a warning, and the current vehicle speed will be reduced by 5% to 10% or decreased by 3 to 5 km / h. The dynamic safety distance was then measured. When the dynamic distance between the water truck and the second target reaches At that time, the water truck stopped spraying and gave way.

[0013] Furthermore, the warning distance The calculation formula is: in, For the speed of the water truck, The buffer time is set to 2-3 seconds; The dynamic safety distance The calculation formula is: in, This refers to the speed of the water truck. The total response time of the water truck's differentiated obstacle avoidance control system is set to 200-500 milliseconds. The effective water jet radius under the current pump pressure. A fixed safety buffer distance of 2 to 3 meters is recommended.

[0014] Furthermore, the strategy integration in accordance with the arbitration principle includes: when coordinated passage, cessation of spraying and avoidance, and prior coordinated safety exist simultaneously, then cessation of spraying and avoidance are executed; when coordinated passage and prior coordinated safety exist simultaneously, then they are executed in parallel.

[0015] On the other hand, the present invention also provides a target classification-based differentiated avoidance control system for sprinkler trucks, used in the differentiated avoidance control method for sprinkler trucks as described above, comprising: The environmental perception module consists of multiple sensor combinations evenly installed on the sprinkler truck. Each sensor combination includes at least a visual camera and a millimeter-wave radar. The visual camera collects image data and the millimeter-wave radar collects point cloud data at a frame rate / frequency equal to or greater than 30Hz. The vehicle-mounted edge computing unit constructs a target detection and classification model based on the YOLOv5 algorithm, and trains it using a set dataset to obtain a trained target detection and classification model; the trained target detection and classification model is deployed on the vehicle-mounted edge computing unit of the sprinkler truck; the vehicle-mounted edge computing unit is also deployed with a pre-fusion algorithm; The data processing and decision-making module includes a differentiated obstacle avoidance decision-making mechanism. This mechanism includes a pedestrian avoidance strategy, which, upon detecting a pedestrian, analyzes whether there is an effective obstacle. If such an obstacle exists, it triggers cooperative passage and then determines whether the distance between the water truck and the pedestrian has reached a verification distance. If the target is reached, the system will identify whether the pedestrian shows a valid intention to avoid the obstacle. If the pedestrian shows a valid intention to avoid the obstacle, the system will maintain cooperative passage and the pedestrian will pass through the section of road. Otherwise, the water truck will stop spraying and give way. If there is no effective cover, the water truck will stop spraying. The differentiated avoidance decision-making mechanism also includes a non-motorized vehicle avoidance strategy, which is used to initiate and execute pre-cooperative safety when a second type of target is identified; The data processing and decision-making module also includes a multi-objective arbitration mechanism, which is used to generate corresponding avoidance strategies in parallel and independently when the first type of target and the second type of target are identified at the same time. Then, it is determined whether there is a conflict between the avoidance strategies of different targets. If there is, the strategy is fused according to the arbitration principle; otherwise, the corresponding avoidance strategy is executed. The execution control module includes a water pump solenoid valve, a nozzle angle adjustment motor, an audible and visual alarm, and a drive water pressure adjustment mechanism; The environmental perception module is connected to the vehicle-mounted edge computing unit and the data processing and decision-making module, respectively. The data processing and decision-making module is also connected to the water pump solenoid valve, nozzle angle adjustment motor, audible and visual alarm, and drive water pressure adjustment mechanism of the execution control module.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) By designing pedestrian avoidance strategy, non-motorized vehicle avoidance strategy and multi-objective arbitration mechanism, the present invention enables the sprinkler truck to handle the sprinkler avoidance of several pedestrians and / or non-motorized vehicles in parallel. Moreover, the system of the present invention has a deep understanding of the environment and refined decision-making ability, and can simulate human drivers to make judgments based on the situation. It has a high degree of intelligence and solves the problem of reduced operation efficiency of sprinkler truck due to a single strategy.

[0017] (2) The pedestrian avoidance strategy of the present invention intelligently considers the existence of effective shelters and whether pedestrians have the intention to actively avoid them, so that the sprinkler truck has the judgment of a human driver and maintains the sprinkler operation under the condition of ensuring no risk of splashing, which greatly reduces the number of invalid water stoppages and improves the lifespan and operating efficiency of the sprinkler truck.

[0018] (3) The non-motorized vehicle avoidance strategy of the present invention increases the active protection of non-motorized vehicles and avoids secondary accidents caused by the spraying of electric vehicles out of control. Attached Figure Description

[0019] The accompanying drawings are incorporated in and form part of this specification, and together with the description serve to explain the principles of the invention.

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1This is a schematic diagram illustrating the overall framework and workflow of the differentiated obstacle avoidance control system for sprinkler trucks of the present invention. Figure 2 This is a core conceptual diagram of the data processing and decision-making module of the present invention; Figure 3 This is a schematic diagram illustrating the relationship between the water sprinkler truck and the contour mask of pedestrians and the contour mask projection of the obscuring object in this invention. Detailed Implementation

[0022] Exemplary embodiments will now be described in detail. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples consistent with some aspects of the invention as detailed in the appended claims.

[0023] Example Please see Figures 1-2 This invention provides a differentiated avoidance control method for sprinkler trucks based on target classification, including synchronously acquiring image data and point cloud data of the environment surrounding the sprinkler truck at a frame rate / frequency equal to or greater than 30Hz; An object detection and classification model was built based on the YOLOv5 algorithm, and the model was trained by setting a dataset. Image data was then transmitted to the trained object detection and classification model for processing. The dataset includes images of pedestrians in various poses, different types of non-motorized vehicles, and complex street scenes. The target bounding box, category, and confidence scores of each category in the image coordinate system obtained after processing the target detection and classification model are fused with the point cloud data before input. That is, the target category confidence scores are spatiotemporally aligned and correlated with the precise distance, azimuth, and radial velocity information provided by the millimeter-wave radar to obtain the position coordinates and velocity vectors with at least the labels of the first and second categories of targets. Employing pre-fusion algorithms can effectively improve the accuracy of target classification and compensate for the limitations of single sensors in specific scenarios (such as the inadequacy of vision in adverse weather conditions and the ambiguity of radar in target classification).

[0024] The pre-fusion algorithm for the water truck usage scenario is as follows: As a preferred implementation, the pre-fusion algorithm includes the following core steps: 1. Spatiotemporal alignment: By synchronizing and calibrating extrinsic parameters through hardware, visual and radar data are aligned in time and space.

[0025] 2. Data Association: Project radar detection points onto the image plane and perform location- and category-based association matching with visual detection boxes.

[0026] 3. Attribute Fusion: For the successfully associated targets, fuse their visual and radar data, obtain more accurate position and speed through filtering algorithms, and improve the classification confidence through decision-level fusion.

[0027] Specifically for the sprinkler truck scenario, after the spatio-temporal alignment and data association processing are completed, a water mist interference filtering module and a fusion confidence management module are respectively designed to strengthen the verification. Among them, the water mist interference filtering module is, before data association, based on the reflection characteristics (such as the stability of the radar cross-section area) and motion characteristics of the millimeter-wave radar point cloud, distinguish the point cloud that conforms to the characteristics of real targets from the instantaneous interference point cloud generated by sprinkler splashing, and filter out the latter to improve the data purity.

[0028] The fusion confidence management module is, after attribute fusion, the algorithm comprehensively considers factors such as sensor status and data association quality, and outputs a fusion confidence for each fused target. Based on this confidence, the system marks the target as "perception reliable" or "perception uncertain" status to guide the subsequent decision-making module to adopt avoidance strategies with corresponding accuracies. For example, the system presets a confidence threshold Q (value range 0~1). When the confidence < Q (for example, because the camera is partially blocked by water stains), mark the target as "perception uncertain" and adopt a more conservative avoidance strategy (such as stopping sprinkling in advance or giving up collaborative passage); when ≥ Q, mark the target as "perception category reliable", so that the system can decide what kind of avoidance strategy to take.

[0029] Specifically for the sprinkler truck scenario, after the attribute fusion processing, a behavior intention analysis feature extraction module is designed to strengthen the verification. Among them, the behavior intention analysis feature extraction module is, for the associated pedestrian targets, in the attribute fusion stage, specifically fuse the pixel displacement of its visual contour and the accurate radial velocity of the radar, and calculate its two-dimensional plane velocity vector (V x , V y ) in the vehicle coordinate system. This vector is used as the core basis for judging whether it has the intention to move towards the shelter.

[0030] Specifically, the specific process of analyzing whether there is an effective shelter is as follows: Through the pre-fusion algorithm, the position coordinates (X, Y) and velocity vector ( , ) with the third type of target label can also be obtained, where the third type of target is a shelter (such as a tree, a bus stop sign). After identifying pedestrians and shelters, use a semantic segmentation model to perform pixel-level segmentation on the pedestrian area and shelter area in the image, and generate an accurate contour mask; The semantic segmentation model (such as DeepLabV3+ or UNet) or the instance segmentation model (such as Mask R-CNN) can be used. In this embodiment, DeepLabV3+ is selected and trained on a dataset containing a large number of urban street scenes and labeled with pedestrians and various potential occlusions (trees, bus stops, railings, vehicles, etc.). This dataset was collected from the perspective of a sprinkler truck.

[0031] Subsequently, based on the spatial geometry calculation of the contour mask, the occlusion effect is quantified: Project the pedestrian's silhouette mask onto a virtual plane perpendicular to the line of sight, and calculate the maximum width of the projection's bounding rectangle. Similarly, project the outline mask of the occluder onto the same virtual plane as the pedestrian, and calculate the maximum continuous width that the occluder projection can cover. ; like Figure 3 As shown, the pedestrian position is calculated under projection. Location of the shelter Connect the points and calculate the distance between this line and the pedestrian's line of sight from the water truck. The included angle θ; like > * K and θ are less than the set threshold If the width of the occlusion is deemed to be valid, it can form an effective occlusion in three-dimensional space. Here, K is a redundancy coefficient greater than 1, which can be taken as 1.2 (to cope with estimation errors and provide additional safety margins for pedestrians). It is 30 degrees Celsius.

[0032] The above method for calculating the projection width is as follows: Input image data + point cloud data, perform semantic segmentation on the image to obtain a pedestrian mask, project the 3D point cloud onto the image plane, or backproject the image detection results into 3D space. Find the set of 3D points that best matches the pedestrian and occlusion mask areas in the image in space, and calculate the centroid or center point of this set of 3D points, which are the 3D positions Pp and Po of the pedestrian and occlusion.

[0033] Specifically, the key inputs used for the above calculations—the three-dimensional positions Pp and Po of the pedestrian and the occupants—are provided by the aforementioned pre-fusion algorithm.

[0034] The system calculates the line-of-sight vector β from the sprinkler truck to the pedestrian's centroid based on the three-dimensional spatial positions of the sprinkler truck and the pedestrian.

[0035] Furthermore, the coordination is specifically achieved by: controlling the audible and visual alarm to emit a warning sound, simultaneously driving the water pressure regulator to reduce the water pressure to 50%~70% of the calibrated value, and controlling the nozzle angle adjustment motor to deflect the main water flow beam 10~15 degrees away from the pedestrian.

[0036] Furthermore, the verification distance Calculated dynamically using the following formula: in, This refers to the instantaneous speed of the water truck. The total response time of the water truck differentiated avoidance control system includes at least the perception and decision delay and the mechanical action time of the actuators (where the actuators mainly refer to the water pump solenoid valve and the nozzle angle adjustment motor), with a value of 200~500 milliseconds; The effective water jet radius under the current pump pressure; The initial velocity of the water flow at the nozzle. P is the water spray pressure. The density of water is given, but in practice, due to nozzle shape and frictional losses, a velocity coefficient is introduced. Adjustments are made, with values ​​ranging from 0.85 to 0.98; The elevation angle of the sprinkler head; The fixed safety buffer distance is 2-5 meters; h is the distance from the sprinkler head to the ground.

[0037] Furthermore, identifying whether a pedestrian exhibits an effective avoidance intention includes: focusing on real-time image sequences of the pedestrian area, quantifying the pedestrian's avoidance intention by analyzing the following features; if all of the following features are met, the pedestrian is considered to have exhibited an effective avoidance intention; otherwise, the pedestrian has no avoidance intention, the sprinkler truck will stop spraying and avoid the pedestrian, controlling the water pump solenoid valve to cut off the water flow in the shortest possible time (e.g., <200 milliseconds) to ensure safety; Feature 1: By tracking the point cloud with millimeter-wave radar, it is calculated whether the centroid of a pedestrian has produced a continuous displacement toward the direction of an effective shield, wherein the continuous displacement is equal to or greater than 0.5 meters; Feature 2: The point cloud tracking data is converted to the camera coordinate system. The displacement distance of the pedestrian's centroid in two adjacent frames is obtained through millimeter-wave radar, thereby obtaining the pedestrian's instantaneous velocity vector. It is then determined whether the angle θ between the instantaneous velocity direction and the line connecting the pedestrian (centroid) and the occluder (centroid) is less than 30 degrees. If it is less than 30 degrees, it is considered that the pedestrian has an effective avoidance tendency.

[0038] Furthermore, the aforementioned pre-emptive collaborative security specifically refers to: when the dynamic distance between the sprinkler truck and the second type of target reaches the warning distance... When the time comes, the sound and light alarm will sound a warning sound, and the current vehicle speed will be reduced by 5% to 10% or decreased by 3 to 5 km / h. The dynamic safety distance was then measured. When the dynamic distance between the water truck and the second target reaches At that time, the water truck stopped spraying and gave way.

[0039] Furthermore, the warning distance The calculation formula is: in, For the speed of the water truck, The buffer time is set to 2-3 seconds; The dynamic safety distance The calculation formula is: in, This refers to the speed of the water truck. The total response time of the water truck's differentiated avoidance control system includes at least the perception and decision delay and the mechanical action time of the water pump solenoid valve / spray head servo mechanism, and is taken as 200~500 milliseconds; The effective water jet radius under the current pump pressure. The fixed safety buffer distance is a safety margin of 2 to 3 meters to account for uncertainties in water flow, such as wind and swaying.

[0040] Furthermore, the strategy integration in accordance with the arbitration principle includes: when coordinated passage, cessation of spraying and avoidance, and prior coordinated safety exist simultaneously, then cessation of spraying and avoidance are executed; when coordinated passage and prior coordinated safety exist simultaneously, then they are executed in parallel.

[0041] It should be noted that, given limited computing resources, the system prioritizes protecting users who have already entered or are about to enter their respective dynamic safety distances. / For targets that are far away, the update frequency is reduced appropriately to optimize the allocation of computing resources.

[0042] On the other hand, embodiments of the present invention also provide a differentiated avoidance control system for sprinkler trucks based on target classification, employing the differentiated avoidance control method for sprinkler trucks as described above, including: The environmental perception module consists of multiple sensor combinations evenly installed on the sprinkler truck. Each sensor combination includes at least a visual camera and a millimeter-wave radar. The visual camera collects image data and the millimeter-wave radar collects point cloud data at a frame rate / frequency equal to or greater than 30Hz. The vehicle-mounted edge computing unit constructs a target detection and classification model based on the YOLOv5 algorithm, and trains it using a set dataset to obtain a trained target detection and classification model; the trained target detection and classification model is deployed on the vehicle-mounted edge computing unit of the sprinkler truck; the vehicle-mounted edge computing unit is also deployed with a pre-fusion algorithm; The data processing and decision-making module includes a differentiated obstacle avoidance decision-making mechanism. This mechanism includes a pedestrian avoidance strategy, which, upon detecting a pedestrian, analyzes whether there is an effective obstacle. If such an obstacle exists, it triggers cooperative passage and then determines whether the distance between the water truck and the pedestrian has reached a verification distance. If the target is reached, the system will identify whether the pedestrian shows a valid intention to avoid the obstacle. If the pedestrian shows a valid intention to avoid the obstacle, the system will maintain cooperative passage and the pedestrian will pass through the section of road. Otherwise, the water truck will stop spraying and give way. If there is no effective cover, the water truck will stop spraying. The differentiated avoidance decision-making mechanism also includes a non-motorized vehicle avoidance strategy, which is used to initiate and execute pre-cooperative safety when a second type of target is identified; The data processing and decision-making module also includes a multi-objective arbitration mechanism, which is used to generate corresponding avoidance strategies in parallel and independently when the first type of target and the second type of target are identified at the same time. Then, it is determined whether there is a conflict between the avoidance strategies of different targets. If there is, the strategy is fused according to the arbitration principle; otherwise, the corresponding avoidance strategy is executed. The execution control module includes a water pump solenoid valve, a sprinkler head angle adjustment motor, an audible and visual alarm, a chassis control system, and a drive water pressure regulating mechanism (water pressure regulator - optional fully automatic type). After receiving instructions from the data processing and decision-making module, the execution control module controls the water pump solenoid valve, sprinkler head angle adjustment motor, audible and visual alarm, chassis control system, and drive water pressure regulating mechanism via CAN bus or direct hardwired connection. The system records key data for all triggered events, including timestamps, target types, decision results, and sensor data snapshots, for subsequent system optimization and performance analysis.

[0043] The environmental perception module is connected to the vehicle-mounted edge computing unit and the data processing and decision-making module, respectively, and the data processing and decision-making module is connected to the execution control module.

[0044] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention.

[0045] It should be understood that the present invention is not limited to the content already described above, and various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A differentiated obstacle avoidance control method for sprinkler trucks based on target classification, characterized in that, include: Image and point cloud data of the surrounding environment of the sprinkler truck are collected synchronously at a set frequency. An object detection and classification model was built based on the YOLOv5 algorithm, and the model was trained by setting a dataset. Image data was then transmitted to the trained object detection and classification model for processing. The target bounding box, category, and confidence score of each category in the image coordinate system obtained after processing by the target detection and classification model are fused with the point cloud data before inputting the point cloud data to obtain the position coordinates and velocity vectors with at least the labels of the first and second types of targets. The first type of target is pedestrians. When a pedestrian is detected, the system analyzes whether there are any effective obstructions. If so, it triggers cooperative passage and then determines whether the distance between the water truck and the pedestrian reaches the verification distance. If the target is reached, the system will identify whether the pedestrian shows a valid intention to avoid the obstacle. If the pedestrian shows a valid intention to avoid the obstacle, the system will maintain cooperative passage and the pedestrian will pass through the section of road. Otherwise, the water truck will stop spraying and give way. If there is no effective cover, the water truck will stop spraying and give way. The second type of target is non-motorized vehicles. When the second type of target is identified, the pre-collaborative safety mechanism is activated. When both the first and second types of targets are identified simultaneously, the corresponding avoidance strategies are generated in parallel and independently. Then, it is determined whether there is a conflict between the avoidance strategies of different targets. If so, the strategy fusion is performed in accordance with the arbitration principle. Conversely, execute the corresponding avoidance strategy.

2. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, The set frequency is a frame rate / frequency greater than or equal to 30Hz.

3. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, The specific process for analyzing whether there are effective shields is as follows: The pre-fusion algorithm can also obtain the position coordinates and velocity vectors with the third type of target label, where the third type of target is the occluder. After identifying pedestrians and occluders, a semantic segmentation model or an instance segmentation model is used to perform pixel-level segmentation of the pedestrian region and the occluder region in the image to generate an accurate contour mask. The pedestrian contour mask is then projected onto a virtual plane perpendicular to the line vector connecting the vehicle's centroid and the pedestrian's centroid, and the width of the maximum bounding rectangle of the projection is calculated. Similarly, project the outline mask of the occluder onto the same virtual plane as the pedestrian, and calculate the maximum continuous width that the occluder projection can cover. ; Calculate pedestrian position under projection Location of the shelter Connect the points and calculate the distance between this line and the pedestrian's line of sight from the water truck. The included angle θ; like > * K and θ are less than the set threshold If the width of the obstruction is determined to be valid, then K is considered a redundancy coefficient greater than 1, which can be taken as 1.

2. It is 30 degrees Celsius.

4. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, The collaboration is specifically achieved through: The system controls the audible and visual alarm to sound an alert, while simultaneously controlling the water pressure regulating mechanism to reduce the water pressure to 50%~70% of the calibrated value, and controlling the nozzle angle adjusting motor to deflect the main water stream 10~15 degrees away from pedestrians.

5. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, The verification distance Calculated dynamically using the following formula: in, This refers to the instantaneous speed of the water truck. The total response time of the water truck's differentiated obstacle avoidance control system is set to 200-500 milliseconds. The effective water jet radius under the current pump pressure; The initial velocity of the water flow at the nozzle. P is the water spray pressure. The density of water is given, but in practice, due to nozzle shape and frictional losses, a velocity coefficient is introduced. Adjustments are made, with values ​​ranging from 0.85 to 0.98; The elevation angle of the sprinkler head; The fixed safety buffer distance is 2-5 meters; h is the distance from the sprinkler head to the ground.

6. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, Identifying whether a pedestrian exhibits an effective avoidance intention involves: focusing on a real-time image sequence of the pedestrian area, and quantifying the pedestrian's avoidance intention by analyzing the following features. If all of the following features are met, the pedestrian is considered to have exhibited an effective avoidance intention; otherwise, the pedestrian has no avoidance intention, and the sprinkler truck will stop spraying and give way. Feature 1: By tracking the point cloud with millimeter-wave radar, it is calculated whether the centroid of a pedestrian has produced a continuous displacement toward the direction of an effective shield, wherein the continuous displacement is equal to or greater than 0.5 meters; Feature 2: The point cloud tracking data is converted to the camera coordinate system, and the displacement distance of the pedestrian's centroid in two adjacent frames is obtained through millimeter-wave radar. This gives the instantaneous velocity vector of the pedestrian, and it is determined whether the angle θ between the instantaneous velocity direction and the line connecting the pedestrian's centroid and the occlusion's centroid is less than 30 degrees. If it is less than 30 degrees, it is considered that the pedestrian has an effective avoidance tendency.

7. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, The aforementioned pre-collaborative security specifically refers to: When the dynamic distance between the sprinkler truck and the second type of target reaches the warning distance When the time comes, the sound and light alarm will sound a warning, and the current vehicle speed will be reduced by 5% to 10% or decreased by 3 to 5 km / h. The dynamic safety distance was then measured. When the dynamic distance between the water truck and the second target reaches At that time, the water truck stopped spraying and gave way.

8. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 7, characterized in that, The warning distance The calculation formula is: in, For the speed of the water truck, The buffer time is set to 2-3 seconds; The dynamic safety distance The calculation formula is: in, This refers to the speed of the water truck. The total response time of the water truck's differentiated obstacle avoidance control system is set to 200-500 milliseconds. The effective water jet radius under the current pump pressure. A fixed safety buffer distance of 2 to 3 meters is recommended.

9. The differentiated obstacle avoidance control method for sprinkler trucks according to claim 1, characterized in that, Following the arbitration principle for strategy integration includes: when coordinated passage, cessation of spraying and avoidance, and prior coordinated safety exist simultaneously, then cessation of spraying and avoidance shall be executed; when coordinated passage and prior coordinated safety exist simultaneously, then they shall be executed in parallel.

10. A differentiated obstacle avoidance control system for sprinkler trucks based on target classification, characterized in that, The differentiated collision avoidance control method for sprinkler trucks according to any one of claims 1 to 9 includes: The environmental perception module consists of multiple sensor combinations evenly installed on the sprinkler truck. Each sensor combination includes at least a visual camera and a millimeter-wave radar. The visual camera collects image data and the millimeter-wave radar collects point cloud data at a frame rate / frequency equal to or greater than 30Hz. The vehicle-mounted edge computing unit constructs a target detection and classification model based on the YOLOv5 algorithm, and trains it using a set dataset to obtain a trained target detection and classification model; the trained target detection and classification model is deployed on the vehicle-mounted edge computing unit of the sprinkler truck; the vehicle-mounted edge computing unit is also deployed with a pre-fusion algorithm; The data processing and decision-making module includes a differentiated obstacle avoidance decision-making mechanism. This mechanism includes a pedestrian avoidance strategy, which, upon detecting a pedestrian, analyzes whether there is an effective obstacle. If such an obstacle exists, it triggers cooperative passage and then determines whether the distance between the water truck and the pedestrian has reached a verification distance. If the target is reached, the system will identify whether the pedestrian shows a valid intention to avoid the obstacle. If the pedestrian shows a valid intention to avoid the obstacle, the system will maintain cooperative passage and the pedestrian will pass through the section of road. Otherwise, the water truck will stop spraying and give way. If there is no effective cover, the water truck will stop spraying. The differentiated avoidance decision-making mechanism also includes a non-motorized vehicle avoidance strategy, which is used to initiate and execute pre-cooperative safety when a second type of target is identified; The data processing and decision-making module also includes a multi-objective arbitration mechanism, which is used to generate corresponding avoidance strategies in parallel and independently when the first type of target and the second type of target are identified at the same time. Then, it is determined whether there is a conflict between the avoidance strategies of different targets. If there is, the strategy is fused according to the arbitration principle; otherwise, the corresponding avoidance strategy is executed. The execution control module includes a water pump solenoid valve, a nozzle angle adjustment motor, an audible and visual alarm, and a drive water pressure adjustment mechanism; The environmental perception module is connected to the vehicle-mounted edge computing unit and the data processing and decision-making module, respectively, and the data processing and decision-making module is connected to the execution control module.

Images