Vehicle-machine cooperative traffic early warning system, control method, device and robot

Abstract

The invention provides a vehicle-machine cooperative traffic early warning system, a control method, a device and a robot, belongs to the technical field of intelligent traffic, is executed by a vehicle-mounted platform, and comprises the following steps: obtaining road environment information through a sensing module at a vehicle end; based on the road environment information, visual identification and decision making are carried out through a machine learning model, and a navigation instruction used for controlling a movable early warning execution terminal separated from the vehicle-mounted platform to move to a target early warning position on a road is generated; the navigation instruction is sent to the movable early warning execution terminal through a wireless communication link; by means of the method, complex perception and computing power tasks are concentrated on the vehicle end, the execution terminal is made to be light, and therefore unification of high efficiency and low cost is achieved on the system level, and the bottleneck that traditional equipment is limited by the body size, power consumption and computing power is broken through.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a vehicle-machine cooperative traffic early warning system, control method, device and robot. Background Technology

[0002] In scenarios such as highway construction, accident road closures, and tunnel emergency closures, it is necessary to quickly deploy warning signs such as triangular warning signs and traffic cones to warn oncoming vehicles and ensure the safety of workers and accident sites. Currently, this mainly relies on manual deployment, which is inefficient and risky. Although some automated equipment has been attempted, it often suffers from the following drawbacks: Functional limitations: As described in the prior art document CN115946611A, it achieves signal warning by integrating a location-based communication device on the vehicle. Although this avoids the physical placement of the warning triangle, its warning method is still limited to the propagation and reception of wireless signals. It cannot automatically build a physical isolation zone or buffer zone in physical space. Its function is singular and its application scenarios are limited. Over-reliance on intrinsic intelligence: Some intelligent robots attempt to integrate perception, decision-making, and execution into a single device, resulting in large device size, high power consumption, high cost, and difficulty in large-scale deployment; The control mode is singular: it lacks a collaborative control mechanism that takes into account both global planning efficiency and local adaptive robustness in complex emergency scenarios.

[0003] Therefore, there is an urgent need in this field for a traffic early warning solution that can achieve automated deployment of physical entities, and is low-cost, highly reliable, and highly adaptable.

[0004] The embodiments of the present invention are improvements made to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a vehicle-machine collaborative traffic warning system, control method, device and robot. Implementing this invention can achieve automated physical deployment and provide a low-cost, highly reliable and highly adaptable traffic warning solution.

[0006] To achieve the aforementioned objective, in a first aspect, embodiments of the present invention provide a vehicle-machine cooperative traffic warning and control method, the technical solution of which is: Executed by an onboard platform, the process includes the following steps: acquiring road environment information through a vehicle-mounted perception module; based on the road environment information, performing visual recognition and decision-making through a machine learning model to generate navigation instructions for controlling one or more mobile warning execution terminals, separate from the onboard platform, to move to a target warning location on the road; and sending the navigation instructions to the mobile warning execution terminals via a wireless communication link. These mobile warning execution terminals are, but are not limited to: intelligent traffic cone robots, intelligent triangular warning sign robots, mobile reflective devices, traffic warning robots, and potentially future warning drones, tracked warning devices, etc.

[0007] By concentrating complex sensing and computing tasks on the vehicle end through the above methods, the execution terminal can be made lightweight, thereby achieving a balance between high efficiency and low cost at the system level, breaking through the bottleneck of traditional devices being limited by their size, power consumption and computing power.

[0008] Based on the first aspect, in one possible implementation, the aforementioned perception module includes an image acquisition device with a variable field of view; acquiring road environment information includes acquiring road images with different field of view angles through the image acquisition device; generating navigation instructions includes generating the instructions based on a visual recognition algorithm that fuses the image information from the different field of view angles; this implementation, by fusing image information from different field of view angles, takes into account both near-end positioning accuracy and far-end path predictability, effectively improving the accuracy and environmental adaptability of navigation instructions.

[0009] In another possible implementation of the first aspect, the above method generates navigation instructions for multiple mobile early warning execution terminals by performing centralized global planning to generate a deployment scheme for the group; the above method also includes the step of: adaptively and collaboratively controlling the deployment scheme based on distributed perception feedback from the mobile early warning execution terminals. This implementation, through a hybrid control mechanism combining centralized planning and distributed adaptation, ensures the global optimality of the deployment scheme and gives the system strong robustness under local emergencies.

[0010] In another possible implementation of the first aspect, the aforementioned wireless communication link supports multiple communication modes and can adaptively switch based on the communication environment. This implementation ensures the continuity and reliability of control command transmission in the complex and ever-changing electromagnetic environment at the accident site through a multi-mode adaptive communication mechanism.

[0011] According to a second aspect of the present invention, a vehicle-machine cooperative traffic warning device for implementing the above-described method is provided, which is configured to be applied to an in-vehicle platform, including an information acquisition module, an instruction generation module and a communication control module, for executing the vehicle-machine cooperative traffic warning control method as described in the first aspect and any possible embodiment of the first aspect. Specifically, the information acquisition module is configured to acquire road environment information through the vehicle-mounted perception module; the instruction generation module is configured to generate navigation instructions based on the road environment information, using a machine learning model for visual recognition and decision-making, to control one or more mobile warning execution terminals, which are separate from the vehicle platform, to move to the target warning location on the road; and the communication control module is configured to send the navigation instructions to the mobile warning execution terminals via a wireless communication link.

[0012] Thirdly, the present invention provides a vehicle-machine cooperative traffic warning system, the system comprising: a vehicle-machine cooperative traffic warning device as described in the second aspect; at least one movable warning execution terminal, which is provided with a communication module and an execution mechanism; The aforementioned communication control module establishes a communication connection with the mobile early warning execution terminal through the communication module; the aforementioned mobile early warning execution terminal is configured to receive navigation instructions through the communication module and move on the road based on the aforementioned navigation instructions through the execution mechanism.

[0013] This system integrates the vehicle-mounted device responsible for decision-making with the execution terminal responsible for implementation, forming a complete operational entity. This configuration, through the aforementioned hardware and software collaborative architecture, achieves a fundamental leap from simple signal warnings to the automated deployment of physical entities, enabling the automatic, rapid, and safe construction of physical isolation zones, thus solving practical pain points that existing technologies have failed to address.

[0014] In one possible implementation of the third aspect, the information acquisition module includes an image acquisition device with a variable field of view; the instruction generation module is configured to run a visual recognition algorithm to generate the navigation instructions based on fused image information from different field of view angles.

[0015] Furthermore, in a fourth aspect of the present invention, a traffic warning robot is provided, comprising a robot chassis, drive wheels, omnidirectional wheels, a control module, and a communication module, for use in the system described in the third aspect.

[0016] Specifically, the drive wheels and casters are mounted on the chassis; The control module and communication module are located within the chassis; The communication module is configured to receive navigation instructions from an external vehicle platform, and the control module is configured to control the drive wheels and omnidirectional wheels according to the navigation instructions, so that the robot can move on the road.

[0017] As the system's execution terminal, this robot offers the advantages of extremely simplified structure, significantly reduced costs, and the ability to achieve large-scale, collaborative cluster deployment by receiving unified commands.

[0018] In one possible implementation of the fourth aspect, the top of the robot chassis is provided with an interface structure for mounting a triangular warning sign or traffic cone. This configuration, through standardized interface design, enables the robot to carry different traffic warning facilities, greatly expanding the versatility and flexibility of its application scenarios.

[0019] Furthermore, the above summary does not enumerate all the features required for embodiments of the present invention, and other combinations of these feature groups may also constitute embodiments of the present invention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the background art, the accompanying drawings used in the embodiments of the present invention or the background art will be described below.

[0021] Figure 1 This is a schematic diagram of the architecture of a vehicle-machine collaborative traffic early warning system in one embodiment of the present invention.

[0022] Figure 2 This is a flowchart of a vehicle-machine cooperative traffic warning and control method in one embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the structure of a traffic warning robot in one embodiment of the present invention.

[0024] Figure 4 This is a typical trajectory comparison of the effects of dynamic field-of-view switching and progressive fusion technology in the embodiments of the present invention. Detailed Implementation

[0025] To make the technical means, creative features, objectives and effects of the embodiments of the present invention easier to understand, the embodiments of the present invention are further described below in conjunction with the figures and specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the embodiments of the present invention. The descriptions such as "first" and "second" in the following embodiments are only for distinction and should not be construed as limiting the scope of protection of the present invention. It should be noted that the "module" or "unit" described in the embodiments can be implemented in part or in whole by software, hardware, firmware or any combination thereof.

[0026] To better understand the embodiments of the present invention, the machine learning and computer vision algorithms involved in the embodiments of this application are explained in detail below: BEVFormer: A visual model based on the Transformer architecture, used to transform 2D image features from multiple cameras around a vehicle into a unified bird's-eye-view (BEV) space. In this embodiment of the invention, this technology enables the system to obtain a top-down view of the road environment, thereby enabling more intuitive lane line recognition, drivable area segmentation, and path planning. YOLO v8 (You Only Look Once version 8): An advanced single-stage target detection algorithm. In this embodiment of the invention, it is configured to quickly and accurately identify and locate specific targets (such as traffic cones, debris, other vehicles, and the traffic warning robot of this system) from images, providing key target information for environmental perception. By adopting the YOLO v8 algorithm, the vehicle-side can achieve rapid identification of distant targets with low computing power consumption. This feature is consistent with the system architecture that concentrates computing power on the vehicle-side, jointly realizing the low cost and lightweight of the robot body.

[0027] DeepSORT (Deep Simple Online and Realtime Tracking): A multi-target tracking algorithm. Based on target detection (e.g., via YOLO v8), it combines appearance feature information and motion models to associate the same target across different frames, achieving continuous target tracking and assigning a unique ID. In this application, it is used to continuously track the motion trajectory of a traffic warning robot.

[0028] SLAM (Simultaneous Localization and Mapping) is a combination of technologies that allows devices to simultaneously build an environmental map and estimate their own position within the map using sensor data (such as LiDAR and vision) in unknown environments. In the scenarios described in this application, where GNSS signals are lacking, such as tunnels, SLAM provides the robot with crucial localization capabilities.

[0029] The above algorithms are integrated and applied through the "vehicle-machine collaboration" system architecture described in this application, jointly achieving high-precision environmental perception, reliable decision-making and planning, and robust collaborative control. It should be understood that the algorithms listed above are merely examples and are not intended to limit the scope of protection of this application. Those skilled in the art will recognize that any existing or future-developed algorithm capable of achieving the same or similar functions can be applied to the system architecture described in this application.

[0030] In one feasible implementation, the control method of the present invention is executed by an onboard platform 200 integrated on a rescue vehicle. See also... Figure 2A flowchart of a control method according to an embodiment of the present invention is shown. The method specifically includes the following steps: Step S101 (Acquiring Road Environment Information): The vehicle platform 200 acquires road environment information through its perception module 201. The perception module 201 on the vehicle side acquires road environment image information. Specifically, the perception module 201 is an image acquisition device with a variable field of view, such as a set of telephoto and / or short-focus switching cameras, LiDAR, GPS, and other sensors, used to collect raw data such as images and point clouds around the vehicle.

[0031] The specific process of dynamically adjusting the field of view is as follows: Based on the relative relationship between the current position of the traffic warning robot 300 and the target warning position in the task planning, the system automatically triggers the perception module 201 to switch the focal length. When it is necessary to accurately locate the initial pose of the robot 300, a short focal length is used for wide-angle shooting; when it is necessary to plan a distant path for the robot, it switches to a long focal length for distant shooting, thereby achieving adaptive perception of different ranges of scenes.

[0032] For example, acquiring road environment information specifically includes controlling the camera array to collect road images from different field-of-view angles. In a specific application scenario, at a highway accident scene, a rescue vehicle (vehicle-mounted platform 200) is parked in the emergency lane. The short-focus camera in the perception module 201 first takes a wide-angle shot of the area within 50 meters behind the vehicle, accurately identifying the initial position of the traffic warning robot 300, which has just automatically driven out of the vehicle's trunk, and its relative position to the lane lines. Simultaneously, the long-focus camera takes a long-range shot of the target deployment area 150-200 meters ahead, identifying the overall road conditions in the area and whether there are any obstacles such as loose objects. In this way, the system simultaneously obtains accurate information about the robot's deployment starting point and the destination area.

[0033] In this embodiment, the perception module 201 is specifically a dual-camera system integrating a short-focal-length camera and a long-focal-length camera. The system is designed to resolve the fundamental contradiction that a single camera cannot simultaneously achieve wide near-end coverage and high-resolution far-end coverage from a fixed viewing angle. The short-focal-length camera (e.g., 4-8mm focal length) provides a wide field of view of 80°-110°, suitable for initial target acquisition and precise positioning within 0-50 meters; the long-focal-length camera (e.g., 16-35mm focal length) provides a narrow field of view of 20°-40°, maintaining an image pixel height of 80-150 pixels for the warning execution terminal (such as a traffic cone robot) at a distance of 100-200 meters, ensuring clear features and stable tracking.

[0034] The dynamic switching and fusion process specifically includes: Close-range dominant phase: When the warning execution terminal 300 has just departed from the vicinity of the vehicle platform, the system uses the short-focal-length camera as the primary sensing source. At this time, the target imaging scale is large (e.g., pixel height > 200), the detection confidence is high (> 0.95), and the system can stably establish a tracking ID and initialize its motion trajectory.

[0035] Switching condition judgment: The system monitors the image quality indicators of the target in the short focal length camera in real time, including: Target bounding box pixel height (h_pixels); Target detection confidence (p_conf); The inter-frame jitter amplitude (σ_center) at the center of the bounding box.

[0036] When one or a combination of the following conditions is met, such as h_pixels<35, p_conf decreases by more than 15%, or σ_center exceeds the empirical threshold, it is determined that the target has entered the mid-to-long range, the imaging quality begins to degrade, and the perception dominance is switched.

[0037] The gradual fusion switching phase: The system does not simply switch the dual sensing sources on and off, but rather initiates a gradual fusion process based on state estimation and smooth transition. This process aims to achieve a seamless transition from short-focus to long-focus dominance, and its core lies in dynamic weight calculation and multi-source state fusion. The specific steps are as follows: Cross-field-of-view target association: First, using an algorithm based on reprojection geometry and appearance feature matching, the warning execution terminals detected in short-focal-length and long-focal-length images are paired to confirm that they are observing the same physical target.

[0038] Dynamic Calculation of Fusion Weights: For the same target that is successfully associated, the system dynamically calculates the confidence weight w_short of the short-focal-length camera observation information based on its real-time imaging quality in the short-focal-length image. The weight calculation uses the following smooth transition function: in: The sigmoid function is a standard sigmoid function. Its function is to map the weights to the (0, 1) interval and ensure continuous and smooth changes, avoiding weight jumps caused by small fluctuations of h_pixels_short near the threshold T, thereby ensuring the stability of subsequent state fusion and control.

[0039] h_pixels_short is the pixel height (in pixels) of the bounding box of the target in the current frame image of the short focal length camera, output by the target detection algorithm (such as YOLOv8), which is a direct measurement value.

[0040] T (switching threshold) is a key empirical parameter representing the limit of pixel height at which a short-focal-length camera can reliably image a specific warning target (such as a standard traffic cone 70cm high) for this system. This threshold (T ≈ 35 pixels) was obtained through analysis and calibration of the target's physical size, camera focal length, image resolution, and a large amount of measured data. It is the quantitative switching trigger benchmark set by this solution for the specific problem of "degradation in long-distance imaging of small road warning targets".

[0041] S (smoothing coefficient) is an adjustable parameter used to control the transition slope of the weight curve. Its value (S) is determined through system debugging and is designed to match the typical moving speed of the early warning execution terminal, so that the weight completes a smooth transition from close to 1 to close to 0 within a reasonable moving distance (such as 10-30 meters) after the target crosses the threshold, in order to adapt to the dynamic response characteristics of the control system.

[0042] The weighting for telephoto cameras is as follows: .

[0043] Multi-source state fusion: Under a unified coordinate system, the target state vector bbox_short from the short-focal-length camera and the target state vector bbox_long from the long-focal-length camera are weighted and fused to obtain the final stable state estimate bbox_fused used for decision-making.

[0044] bbox_short and bbox_long: These are normalized state vectors of the same target, confirmed by the aforementioned association algorithm and smoothed by a filtering algorithm (such as Kalman filtering). These vectors contain at least information about the target's position and velocity in the bird's-eye view (BEV) coordinate system or the world coordinate system, rather than simply image pixel coordinate boxes.

[0045] This weighted fusion is not a simple image stitching, but a synthesis performed at the state estimation level. During the transition phase, bbox_fused combines the short-range positioning stability of bbox_short with the long-range observation accuracy of bbox_long, thus providing the onboard decision center with continuous, smooth, and high-confidence target pose information throughout the entire movement process, fundamentally avoiding target loss or trajectory jumps during the switching process.

[0046] Long-range dominant phase: When the early warning execution terminal moves to a distance of more than 100 meters (at which point w_short approaches 0), the long-focal-length camera becomes the primary sensing source (w_long ≈ 1), providing high-resolution images to support accurate pose determination and status monitoring. The short-focal-length camera then serves as an auxiliary for environmental monitoring, or provides redundant sensing when the long-focal-length view is temporarily obstructed.

[0047] Step S102 (Generate Navigation Instructions): Based on the road environment information obtained in S101, the vehicle platform 200 performs visual recognition and decision-making through a machine learning model to generate navigation instructions for controlling the traffic warning robot 300, which is separated from the vehicle platform 200, to move to the target warning position on the road. Specifically, this includes generating the instructions based on a visual recognition algorithm that fuses the image information from different field of view angles. This algorithm can specifically adopt a lane line recognition model based on BEVFormer and an obstacle and robot body detection and tracking model based on YOLO series target detection models (such as YOLO v8) (such as DeepSORT). By fusing the precise pose of the near-focus image and the path predictability of the far-focus image, a smooth and safe navigation instruction is finally generated.

[0048] In this embodiment, the visual recognition and decision-making algorithm run by the instruction generation module 202 can be built and trained based on existing advanced machine learning architectures such as BEVFormer and YOLO series. These algorithms are integrated through the "vehicle-machine collaboration" system of this application, specifically adapted to process fused information from the variable field-of-view image acquisition device, thereby generating accurate and reliable navigation instructions in scenarios such as "variable field of view" and "multi-robot collaboration". Those skilled in the art can select suitable existing algorithms or optimize them to achieve the above functions based on the teachings of this application.

[0049] In a specific decision-making scenario, the instruction generation module 202 receives the fused visual information. It converts the multi-view images into a bird's-eye view using the BEVFormer model, clearly "seeing" a feasible path from the robot's current position (emergency lane) to the target position (the boundary between the driving lane and the emergency lane). Simultaneously, the YOLO v8 model identifies a small, scattered tire debris 80 meters ahead on the path. The decision-making algorithm (such as reinforcement learning) then calculates the navigation instructions: "First, proceed straight for 60 meters along the emergency lane, then turn 15 degrees to the left, cutting into the driving lane approximately 0.5 meters to avoid the obstacle, and finally continue straight to the target point." This sequence of instructions ensures the robot arrives safely with a smooth trajectory.

[0050] In this embodiment, by working together with the variable field of view device and combining it with an advanced machine learning model, the industry problem of being unable to balance positioning accuracy and path prediction from a single viewpoint is effectively solved, and the accuracy and environmental adaptability of navigation are significantly improved.

[0051] Step S103 (Sending Navigation Commands): The vehicle platform 200 sends the generated navigation commands to the traffic warning robot 300 via a wireless communication link. For example... Figure 2The process block "Issue instructions to the robot" is described above. The aforementioned wireless communication link supports multiple communication modes and can adaptively switch based on the communication environment. Specifically, this communication link can integrate Wi-Fi, 5G, and UWB communication chips. The communication modules of the vehicle and the robot continuously monitor communication quality (such as signal strength and latency). In open areas, the system prioritizes the use of low-latency, high-speed Wi-Fi for instruction transmission. When the robot moves out of Wi-Fi coverage, it automatically switches to the wider-coverage 5G network. When the vehicle and robot enter areas with severe signal obstruction, such as tunnels or bridges, the system automatically activates the highly penetrating UWB technology for communication.

[0052] In a specific communication scenario, robot 300 initially receives commands via Wi-Fi within 50 meters of the vehicle's rear, with a latency of less than 50ms. When it moves 200 meters away, reaching approximately 120 meters, the vehicle-side communication control module 203 detects that the Wi-Fi signal strength has dropped below the threshold. Upon sending the next command packet, it automatically switches to a 5G network. Although the latency increases slightly to 80ms, the continuity and reliability of the commands are guaranteed. If this incident occurs inside a tunnel, the system will prioritize using UWB base stations deployed within the tunnel for communication, avoiding loss of connection due to signal obstruction.

[0053] In another implementation scenario, such as an emergency tunnel closure, the communication environment is complex and visibility is obstructed. The rescue vehicle (vehicle-mounted platform 200) stops in a safe area before the tunnel entrance. The perception module 201 is activated, with its lidar and variable field-of-view image acquisition device working together. The short-focus camera captures a wide-angle image of the area near the tunnel entrance, accurately identifying the road shoulder, maintenance strip, and the initial position of the newly deployed traffic warning robot 300. The long-focus camera and lidar attempt to probe inside the tunnel, but due to light and signal limitations, the ability to identify detailed road conditions inside the tunnel is limited. The instruction generation module 202 generates navigation instructions based on the fused perception data. Due to the weak GPS signal inside the tunnel, the system prioritizes a combination of lidar SLAM (Simultaneous Localization and Mapping) technology and visual odometry for positioning. The vehicle-side decision algorithm plans a path that closely follows the maintenance strip on the right side of the tunnel to minimize the risk of main traffic flow, and sets the robot's final target warning position 150 meters inside the tunnel.

[0054] In tunnel scenarios, the communication control module 203 automatically activates the pre-configured UWB communication mode. The vehicle and robot 300 communicate through a pre-deployed UWB base station in the tunnel, ensuring low-latency and high-reliability transmission of control commands and status feedback in the heavily shielded tunnel environment. After receiving the command, the traffic warning robot 300 moves towards the target location along the right-side inspection strip of the tunnel. Its built-in simple ultrasonic sensor continuously detects the distance to the tunnel wall during movement, makes local fine adjustments to prevent collisions, and feeds back this positioning assistance information to the vehicle to help update the global map.

[0055] In this embodiment, the multi-mode adaptive switching mechanism based on environmental perception ensures that control commands can be transmitted stably and uninterruptedly throughout the deployment process, thereby guaranteeing the successful execution of the task. The method executes the above steps S101 to S103, realizing a collaborative architecture of "vehicle-side decision-making and robot execution", concentrating complex perception and computing tasks on the vehicle side, making the execution terminal lightweight, thereby achieving a balance between high efficiency and low cost at the system level.

[0056] In one feasible implementation, refer to Figure 2 When the system contains multiple traffic warning robots, the above step S102 (generating navigation instructions) includes performing centralized global planning, whereby the vehicle generates an optimal deployment scheme for the entire robot group. Furthermore, after the robot begins its deployment, the above method enters a continuous monitoring and adjustment cycle, specifically including: Step S104 (Adaptive Cooperative Control): The system dynamically adjusts the deployment scheme based on the movement status feedback (such as position, speed, and attitude) from the robot group. For example, during movement, each robot uses its onboard simple sensors to perceive nearby robots and sudden obstacles, and feeds this information back to the vehicle via a wireless communication link (this constitutes "distributed perception feedback"). The vehicle dynamically adjusts the global scheme based on all feedback information; In a specific collaborative scenario, the vehicle-mounted system planned to deploy five robots (A1-A5) in a straight line between the lane and the emergency lane. During execution, the vehicle-mounted system, through its perception module 201, detected a lost suitcase on the planned path of robot A3 (this perception was also performed by the vehicle-mounted system). Simultaneously, robot A3 reported its real-time position and speed status (this is motion status feedback) to the vehicle-mounted system. After receiving the status information from all robots, the vehicle-mounted system's instruction generation module 202 quickly recalculated the global path and issued new adjustment instructions: "A2, A3 pause; A4 detour 2 meters to the right; A5 follow A4's path." Meanwhile, A1 and A2 also maintained their distance from each other using their infrared sensors, performing self-organized fine-tuning to prevent collisions. This hybrid mechanism ensures that even in the event of unexpected obstacles, the entire deployment task can be flexibly completed under overall control.

[0057] In another implementation scenario, during a highway accident requiring the rapid deployment of a long-distance isolation zone, the system activates a multi-robot collaborative deployment mode. The perception module 201 of the vehicle-mounted platform 200 performs a panoramic scan of the road environment within a 200-meter radius behind the accident site. The instruction generation module 202 performs centralized global planning, generating an optimal deployment scheme for the robot group (e.g., A1, A2, A3, A4, A5): deploying one robot every 50 meters behind the accident site to form a 200-meter-long arc-shaped warning isolation zone. The communication control module 203 uses 5G C-V2X technology to concurrently issue different path point instructions to each robot. The robot group then begins to synchronously execute the deployment task.

[0058] During execution, robot A3 detected a slowly changing lane vehicle suddenly appearing on its planned path using its millimeter-wave radar (this is distributed perception feedback). It immediately reported this threat information to the vehicle. After receiving the status and feedback from all robots, the command generation module 202 on the vehicle quickly performed global path replanning and issued adjustment commands: "A3 pauses, A4 accelerates and moves to a position 60 meters ahead of A3 to fill the gap, A5's path remains unchanged." At the same time, robots A2 and A3 reported their speed and position information to the vehicle via the V2X-DSRC module. The command generation module 202 on the vehicle coordinated the adjustments, calculated speed adjustment commands to avoid rear-end collisions, and issued them to A2 and A3 respectively, thus preventing rear-end collisions during the adjustment process.

[0059] In this embodiment, the hybrid control mechanism that combines centralized planning with feedback-based adaptive cooperative control not only ensures the global optimality of the deployment scheme, but also endows the system with strong robustness in dealing with local emergencies.

[0060] In one feasible implementation, see Figure 1A vehicle-machine cooperative traffic warning system 100, the system comprising: The vehicle platform 200 integrates a vehicle-machine collaborative traffic warning device, which includes an information acquisition module 201, such as an image acquisition device with a variable field of view; an instruction generation module 202, such as an on-board computing unit; and a communication control module 203. At least one traffic warning robot 300 is provided with a second communication module 301, a control module 302, an actuator 303, and a battery 304. The communication control module 303 establishes a communication connection with the traffic warning robot 300 through the second communication module 301. The traffic warning robot 300 is configured to receive navigation instructions through the second communication module 301 and move on the road based on the navigation instructions through the actuator 303.

[0061] In a specific system scenario, after the rescue vehicle arrives at the accident scene, the operator selects the task of "deploying a 200-meter isolation barrier" on a tablet computer inside the vehicle. The information acquisition module 201 (long and short focal length cameras) of the vehicle-mounted platform 200 begins scanning the environment. The instruction generation module 202 (onboard computer) completes environmental modeling and path planning within seconds, generating a sequence of movement instructions for six robots. The communication control module 203 (onboard communication host) sends these instructions sequentially to the six robots via Wi-Fi. After receiving the instructions, the second communication module 301 of the robot 300 passes them to the control module 302 (the robot's main control MCU) for parsing, and drives the actuators 303 (motors and wheels) to perform movement. The entire system, from receiving instructions to completing deployment, can be automated within 5 minutes without personnel leaving the vehicle, greatly ensuring personnel safety and improving response efficiency.

[0062] In one feasible implementation, see Figure 3 This embodiment provides a traffic warning robot 300, which includes a lightweight robot chassis 305. The robot chassis 305 adopts a lightweight modular design, such as a frame structure. The lower part of the chassis 305 is provided with a drive mechanism, such as drive wheels 306 and omnidirectional wheels 307. The chassis 305 is provided with a control module 302 and a communication module 301. The communication module 301 is configured to receive navigation instructions from an external vehicle platform 200, and does not perform environmental perception and path decision-making itself. The control module 302 is configured to control the drive wheels 306 and omnidirectional wheels 307 according to the navigation instructions, so that the robot moves on the road. To more clearly understand why the technical solution of this application is superior to the prior art, the following embodiments are used to illustrate this: [Example 1] Real-vehicle verification of the effects of dynamic field of view switching and progressive fusion technology 1. Experimental Objective This study verifies whether the dynamic field-of-view switching strategy based on dual cameras and progressive weight fusion described in this application can significantly reduce the lateral error at the endpoint and maintain continuous and stable tracking in long-distance (150 meters) traffic cone deployment tasks, compared to a single fixed focal length scheme.

[0063] 2. Experimental Environment Location: Closed test section of a city's ring expressway (two-way four lanes, dry asphalt, unobstructed).

[0064] Time: 09:00–11:00 on December 1, 2025, with 35,000–45,000 lux of sunlight and wind speed <2 m / s.

[0065] Reference coordinate system: with the center of the rear bumper of the rescue vehicle as the origin, the longitudinal axis of the vehicle is +X, the lateral axis is +Y, and the vertical upward axis is +Z.

[0066] Truth system: Huace i90 RTK base station + mobile station, planar accuracy ±1 cm, data rate 10 Hz, already connected to the vehicle platform for synchronization triggering.

[0067] 3. Experimental setup

[0068] 4. Experimental Grouping Group A: Fixed short focal length (4 mm) for full range operation.

[0069] Group B: Fixed telephoto lens (35 mm) working throughout the entire range Group C: This application's "dynamic switching + gradual integration" scheme.

[0070] Each group consists of 3 robots, each performing 5 straight-line deployment tasks from 0 to 150 m, resulting in 45 trajectories.

[0071] 5. Testing Process (1) The robot starts from 0 m behind the vehicle and moves forward at a constant speed of 1 m / s along the center line of the lane; (2) The vehicle-mounted platform records synchronously at 30 Hz: a. Target pixel height per frame, h_pixels; b. YOLOv8 detection confidence level p_conf; c. Bounding box center jitter σ_center; d.RTK truth value(x, y); (3) After reaching the preset coordinate of 150m, calculate the lateral error at the endpoint Δy = |y_true–y_plan|; (4) The entire original video, RTK log, and CAN bus data are packaged into a ROS2 bag, verified by SHA256, and then uploaded to OSF (DOI 10.17605 / OSF.IO / XXXXX).

[0072] 6. Evaluation Indicators Main: Endpoint lateral error Δy (cm).

[0073] secondary: Average detection confidence p̄_conf; Track ID switching count N_switch; The maximum inter-frame jump during the switching process is Δbbox_max (pixels).

[0074] 7. Experimental Results

[0075] Statistical tests: One-way ANOVA (α = 0.01) + Tukey HSD post-hoc comparison, the differences between group C and groups A and B were all p < 0.001.

[0076] Figure 4 Typical trajectory comparisons are given: Group C maintained subpixel-level tracking stability in the 100–150m range, while Group A experienced three ID loss events at 120m, leading to a sharp increase in endpoint error.

[0077] [Example of a return-to-home scenario] When the early warning mission is completed, and the traffic warning robot needs to return from the distant target location to the vicinity of the vehicle platform, the system executes a dynamic field of view adjustment strategy that is the opposite of the near-to-far process described in Example 1, moving from far to near. The specific process is as follows: Long-distance return phase: The robot begins its return journey from 150 meters away. The system uses a telephoto camera as the primary sensing source, leveraging its high resolution to stably track the robot and ensure that the target is not lost during long-distance movement.

[0078] Switching Condition Judgment: The system monitors the imaging quality of the robot in long-focal-length images in real time. As the robot gradually approaches the vehicle, its pixel height gradually increases, the detection confidence improves, and the bounding box jitter decreases. When preset conditions are met (e.g., pixel height > 200 pixels, confidence > 0.95), the system triggers a gradual switch from long-focal-length to short-focal-length images. Specifically, the preset switching conditions include at least one of the following: The pixel height of the target in the telephoto image exceeds a preset threshold; The target detection confidence level has been raised to the preset level. The inter-frame jitter amplitude at the center of the target bounding box is reduced to below the preset threshold.

[0079] Progressive fusion switching phase: The system performs a cross-field target association and dynamic weight fusion process similar to that in Example 1, but the weight calculation basis is changed from short focal length pixel height to long focal length pixel height to ensure that when the robot approaches the vehicle, the perception information smoothly transitions from long focal length to short focal length, avoiding positioning jumps or target loss.

[0080] Close-range return phase: When the robot enters within 50 meters of the vehicle platform, the system uses a short-focal-length camera as the main sensing source and utilizes its wide field of view to accurately locate and adjust the robot's posture until the robot safely returns to the designated location on the vehicle platform (such as the vehicle's trunk).

[0081] This return-to-home scheme and the aforementioned deployment scheme from near to far are mirrored processes, together forming a complete "deployment-return-to-home" closed-loop perception and control process, demonstrating the technical advantages of this system in both bidirectional mobile tasks, namely high precision and high robustness.

[0082] 8. Reproducibility statement Code: GitHub.com / YourOrg / DualCam-EXP1 (MIT License).

[0083] One-click reproduction: python run_exp1.py --config configs / exp1.yaml.

[0084] Dependencies: Python 3.10, PyTorch 2.1, OpenCV 4.8, ROS2 Humble.

[0085] Dataset: A 450 GB ROS2 bag has been made public and can be downloaded directly via OSF and its SHA256 hash verified.

[0086] [Example 2] Comparison Experiment of Size, Power Consumption and Cost of Zero-Perception Robot 1. Experimental Objective The quantitative analysis of the improvements in size, power consumption, and cost of the "zero-perception" architecture (no camera, no LiDAR, no edge computing unit) of this application compared to existing technologies (D1 triangular warning sign vehicle, D2 intelligent traffic cone) demonstrates its "unexpected technical effects".

[0087] 2. Comparison Objects D1: Commercially available "Automatic Triangle Warning Sign Cart" (with 1×1080p camera + 2×ultrasonic sensors).

[0088] D2: Intelligent traffic cone (4 ultrasonic sensors at the bottom + 1 720p camera).

[0089] This application: Zero-sensory robot (MCU + motor + communication module + battery only).

[0090] 3. Measurement Method Volume: water displacement method (5000 mL graduated cylinder, accuracy ±10 mL).

[0091] Power consumption: Keysight N6705B DC power analyzer, sampling at 1 kHz, recording for 10 seconds average.

[0092] Cost: BOM publicly quoted price (2025-12-01 LCSC / Digi-Key spot price, including customs duties and VAT).

[0093] Weight: Electronic scale, 0.1 g accuracy.

[0094] 4. Experimental Results

[0095] Experiments show that the proposed solution reduces volume by 50%, power consumption by 75%, and cost by 69%, achieving unexpected technical results.

[0096] 5. Reproducibility BOM (Bill of Materials): GitHub.com / YourOrg / ZeroSense-BOM (including LCSC order number and Digi-Key link).

[0097] Test fixture STL file: can be replicated by 3D printing from the same repository.

[0098] Measurement video: The OSF repository provides full 4K 60 fps recording, allowing you to check the readings frame by frame.

[0099] [Example 3] Stability Verification of Adaptive Switching in Multimode Communication Links 1. Experimental Objective Verify the latency and packet loss performance of the "Wi-Fi → 5G → UWB" adaptive switching mechanism described in this application in complex scenarios such as tunnels.

[0100] 2. Experimental Scenario Location: A very long tunnel (3.8 km in length, with weak 4G / 5G coverage, and 6 UWB base stations have been deployed).

[0101] Robot speed: 1 m / s constant speed.

[0102] Command packets: 256 B, frequency 10 Hz, total 1000 packets.

[0103] 3. Test Indicators Switching delay T_switch (ms): The time from triggering the switching condition to receiving the first ACK for the new link.

[0104] Packet loss rate P_loss (%).

[0105] End-to-end average delay T_e2e (ms).

[0106] 4. Experimental Results

[0107] 5. Reproducibility Script: GitHub.com / YourOrg / CommSwitch-EXP3 (Python + Scapy).

[0108] The original pcap file is publicly available in the OSF repository and can be opened directly with Wireshark to reproduce the statistical results.

[0109] It is important to note that the traffic warning robot 300 itself does not have sensing devices (such as cameras or lidar) or corresponding decision-making units for environmental perception and navigation decisions. Its core functions are limited to receiving instructions, executing movements, and reporting status, thus achieving extreme lightweighting and low cost; all complex perception and computing tasks are completed by the vehicle-mounted platform 200. Specifically, the perception module 201 of the aforementioned vehicle-mounted platform 200 is a set of image acquisition devices with variable field of view (such as long-focus and short-focus switching cameras), which are physically positioned high outside the vehicle, such as on the vehicle roof, at the rearview mirror position, or above the tailgate. This high-positioning is designed to provide the vehicle-mounted computing module 202 with a wide field of view to minimize blind spots, thereby enabling simultaneous and clear observation of the robot 300's near starting position and the far-end environment of the road, in order to generate high-quality navigation instructions.

[0110] Furthermore, the top of the robot chassis 305 is provided with an interface structure 308 for mounting a warning triangle or traffic cone, through which the warning triangle or traffic cone 309 can be quickly installed; in this embodiment, the robot serves as the execution terminal of the system, and its structure is simplified and its cost is controllable.

[0111] It should be understood that the terms "one embodiment," "an embodiment," "a feasible implementation," or "some implementations" used throughout the specification mean that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the present invention. Therefore, "one embodiment," "an embodiment," "a feasible implementation," or "some implementations" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Those skilled in the art should also recognize that the embodiments described in the specification are optional embodiments, and the actions and modules involved are not necessarily essential to the embodiments of the present invention.

[0112] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by instructions. These instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] The above description is merely a specific embodiment of the present invention, but the protection scope of the embodiments of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of the present invention should be included within the protection scope of the embodiments of the present invention. Therefore, the protection scope of the embodiments of the present invention should be determined by the protection scope of the claims.

Claims

1. A traffic warning control method in cooperation with a vehicle machine, characterized by, executed by a vehicle-mounted platform, comprising: acquiring road environment information by a perception module at the vehicle end, the perception module being configured to include a dual-camera system of a short-focus camera and a long-focus camera, and dynamically adjusting a main perception source to acquire road images containing different range scenes; based on the road environment information, performing visual recognition and decision-making by a machine learning model to generate navigation instructions for controlling one or more movable early warning execution terminals separated from the vehicle-mounted platform to move to a target early warning position on the road; sending the navigation instructions to the movable early warning execution terminal through a wireless communication link; wherein the perception module is configured to be able to dynamically adjust its field of view angle to acquire road images containing different range scenes, and the dynamic adjustment includes: adaptively selecting the short-focus camera or the long-focus camera as the main perception source according to the relative relationship between the current position of the movable early warning execution terminal and the target early warning position, and the real-time imaging features of the early warning execution terminal in the image, and fusing the dual-path image information; the generation of the navigation instructions includes generating the instructions based on a visual recognition algorithm that fuses the image information of different field angles.

2. The method of claim 1, wherein, The dynamic adjustment of the field of view angle specifically includes the following steps: in the near-distance movement stage of the early warning execution terminal, the short-focus camera is used as the main perception source to acquire large field of view images for initial detection, accurate positioning and motion state estimation of the execution terminal; when the target imaging quality is detected to drop to a preset switching condition, the gradual switching of the perception dominance from the short-focus camera to the long-focus camera is started; in the switching process, the same early warning execution terminal detected by the short-focus and long-focus cameras is cross-field target associated, and the fusion weight of the two-path image information is dynamically calculated based on the pixel height of the target in the short-focus image; in the long-distance movement stage of the early warning execution terminal, the long-focus camera is used as the main perception source to acquire high-resolution images to maintain stable tracking and state monitoring of the execution terminal.

3. The method of claim 1, wherein, When the movable early warning execution terminal needs to return from the long-distance target early warning position to the vicinity of the vehicle-mounted platform, the dynamic adjustment of the field of view angle includes the following steps: in the long-distance return stage of the early warning execution terminal, the long-focus camera is used as the main perception source to acquire high-resolution images to maintain stable tracking and state monitoring of the execution terminal; when the target imaging quality is detected to improve to a preset switching condition, the gradual switching of the perception dominance from the long-focus camera to the short-focus camera is started; in the switching process, the same early warning execution terminal detected by the short-focus and long-focus cameras is cross-field target associated, and the fusion weight of the two-path image information is dynamically calculated based on the pixel height of the target in the long-focus image; in the near-distance return stage of the early warning execution terminal, the short-focus camera is used as the main perception source to acquire large field of view images for accurate positioning and attitude adjustment of the execution terminal until the return is completed.

4. The method according to claim 1 or 2, characterized in that, When there are multiple movable early warning execution terminals: The step of generating the navigation instruction comprises: performing, by the vehicle-mounted platform, central path planning based on global environment perception to generate a collaborative deployment scheme for the robot group. The method further comprises the steps of: based on distributed local state feedback from the early warning execution terminal, dynamically collaboratively controlling and re-planning the execution of the deployment scheme, forming a hybrid control mechanism combining central planning and distributed self-adaptation.

5. The method of claim 1, wherein, The wireless communication link supports multiple communication modes and can adaptively switch based on the communication environment status.

6. A traffic warning device in cooperation with a car machine, characterized by, The device is applied to a vehicle-mounted platform and comprises: An information acquisition module configured to acquire road environment information through a perception module of the vehicle end, wherein the perception module is configured to be able to dynamically adjust its field of view angle; An instruction generation module configured to generate, based on the road environment information, navigation instructions for controlling one or more movable early warning execution terminals separated from the vehicle-mounted platform to move to a target early warning position on the road through visual recognition and decision making by a machine learning model; A communication control module configured to send the navigation instructions to the movable early warning execution terminal through a wireless communication link; Wherein, the dynamic adjustment of the field of view angle of the perception module comprises: adaptively selecting a short focal length camera or a long focal length camera as the main perception source according to the relative relationship between the current position of the movable early warning execution terminal and the target early warning position, and the real-time imaging features of the early warning execution terminal in the image, and fusing the double-path image information.

7. A traffic warning system in cooperation with a car machine, characterized by, Comprise: The traffic early warning device of claim 6; At least one movable early warning execution terminal provided with a communication module and an execution mechanism; Wherein, the communication control module establishes a communication connection with the movable early warning execution terminal through the communication module; The movable early warning execution terminal is configured to receive the navigation instructions through the communication module and move on the road based on the navigation instructions through the execution mechanism.

8. The system of claim 7, wherein, The information acquisition module comprises a short focal length camera and a long focal length camera; the instruction generation module is configured to run a visual recognition algorithm to generate the navigation instructions based on dynamically fused image information of different field of view angles.

9. A traffic warning robot for use in a system as claimed in claim 7 or 8, characterized in that, The robot comprises: A robot chassis; A driving mechanism provided on the chassis; A control module and a communication module provided in the chassis; Wherein, the communication module is configured to receive navigation instructions from an external vehicle-mounted platform, and the control module is configured to control the driving mechanism according to the navigation instructions to move the robot on the road; The robot itself is not provided with a perception device and a corresponding decision unit for environment perception and navigation decision making. 10.The traffic warning robot of claim 9, wherein, The top of the robot chassis is provided with an interface structure for carrying a triangular warning sign or a traffic cone.

Images