An automated road cold spray marking machine

Abstract

The utility model discloses an automatic road cold -spraying line marking machine, and the automatic road cold -spraying line marking machine is composed of eight parts of rack, autonomous cruise device, power device, cleaning device, storage device, feeding device, line marking device and energy device. The utility model has the advantages that: the positioning drift and blind area obstacle missing detection problem caused by traditional line marking machine depending on manual marking line or single sensor are solved, especially suitable for satellite signal missing scenes such as tunnels and viaducts, can identify mobile obstacles in real time and trigger path re-planning, and avoid construction interruption caused by fixed track planning of traditional equipment.

Description

Technical Field

[0001] This utility model relates to the field of road marking equipment, and in particular to an automated road cold spray marking machine. Background Technology

[0002] Road marking construction is a crucial aspect of traffic engineering, and its quality directly impacts traffic safety and efficiency. Currently, existing fully automated cold spray marking equipment suffers from severe environmental interference affecting spraying accuracy. When constructing on complex road alignments (such as curved left-turn waiting lanes), the spray nozzles are prone to marking deviation due to road surface bumps or mechanical vibrations, with errors exceeding ±2cm. The main reason for this problem is that existing equipment relies solely on GPS positioning and lacks a dynamic compensation mechanism. Summary of the Invention

[0003] This invention mainly solves the above-mentioned problems and provides an automated road cold spray marking machine. By fusing LiDAR and depth camera, and combining DBSCAN and FICP algorithms to dynamically match point cloud data, it solves the problems of positioning drift and blind spot obstacle omission caused by traditional marking machines relying on manual marking or single sensors.

[0004] The technical solution adopted by this utility model to solve its technical problem is an automated road cold spray marking machine, including a frame composed of a chassis, an outer frame and an upper cover, and further including: The autonomous navigation module, mounted on the rack, includes a binocular depth camera, a lidar, and a control motherboard, and is used to acquire environmental information and plan a route. The power module, mounted on the chassis, includes an AGV steering wheel and omnidirectional wheels, which are diagonally distributed on both sides of the chassis for traveling according to a planned route. The cleaning module, mounted on the chassis, includes a rotating brush and a vacuuming device for removing dust from the road surface; The material storage module, mounted on the chassis, is used to store paint. The feeding module, mounted on the chassis and connected to the storage module, is used to transport the coating. The marking module, set on the outer frame, includes a marking gun, a nozzle, and an atomizing device. The marking gun is connected to the feeding module and is used to spray paint for marking lines. The power supply module provides the energy required for the automated road cold spray marking machine.

[0005] As a preferred embodiment of the above solution, a fixed frame is provided on each of the two front sides of the chassis, and the lidar is installed in the fixed frame; a hollow is provided in the middle of the front of the outer shell, and the binocular depth camera is installed in the hollow; the control motherboard is installed inside the outer frame.

[0006] As a preferred embodiment of the above solution, the cleaning module further includes a brush motor, a connecting rod, driven wheels, and a suction pipe. The rotating brushes are symmetrically arranged on both sides of the center of the bottom surface of the base plate, and the brush motor drives the rotating brushes to rotate through the connecting rod. The driven wheels are symmetrically arranged between the two rotating brushes, and a rotating plate is provided between the two driven wheels. The suction pipe is arranged on the rotating plate and connected to the suction device, and the suction device is arranged on the chassis.

[0007] As a preferred embodiment of the above solution, the feeding module includes a plunger pump, a pulley, a belt, a feeding motor, and a conveying hose. The drive motor drives the plunger pump through the pulley and the belt. The conveying hose connects the plunger pump to the storage module and the plunger pump to the marking module.

[0008] As a preferred embodiment of the above solution, the upper end of the marking gun is connected to the feeding module via a conveying hose, the atomizing device and the nozzle are sequentially arranged at the lower end of the marking gun, and the marking gun is equipped with a filter screen, a flow regulating valve and a solenoid valve.

[0009] As a preferred embodiment of the above solution, a filter screen is provided between the storage module and the feeding module.

[0010] As a preferred embodiment of the above solution, the energy module is a gasoline generator, which is mounted on the chassis via a multi-stage damping mechanism. The multi-stage damping mechanism includes a high-damping rubber block, a variable-diameter helical spring, and a microporous foamed rubber pad arranged sequentially from top to bottom.

[0011] The advantages of this invention are: by fusing lidar and depth camera, and combining DBSCAN / FICP algorithm to dynamically match point cloud data, it solves the problems of positioning drift (>50 mm) and blind spot obstacle omission caused by traditional line marking machines relying on manual marking or single sensors. It is especially suitable for scenarios where satellite signals are missing, such as tunnels and overpasses. It can identify moving obstacles (such as vehicles and pedestrians) in real time and trigger path replanning, avoiding construction interruptions caused by fixed trajectory planning of traditional equipment, and improving the safety of operation in complex traffic environments. Attached Figure Description

[0012] Figure 1 This is a side view of an automated road cold spray marking machine.

[0013] Figure 2 This is a front view structural diagram of an automated road cold spray marking machine.

[0014] Figure 3 This is a schematic diagram of the structure from below for an automated road cold spray marking machine.

[0015] Figure 4 This is a schematic diagram of the internal structure of an automated road cold spray marking machine.

[0016] 1-Chassis; 2-Outer frame; 3-Upper cover; 4-Dual-lens depth camera; 5-Fixed frame; 6-LiDAR; 7-AGV steering wheel; 8-Universal wheel; 9-Rotating brush; 10-Connecting rod; 11-Brush motor; 12-Driven wheel; 13-Dust suction pipe; 14-Dust suction device; 15-Storage bin; 16-Plunger pump; 17-Pulley; 18-Belt; 19-Feeding motor; 20-Conveying hose; 21-Marking gun; 22-Nozzle; 23-Gas generator. Detailed Implementation

[0017] The technical solution of the present invention will be further described below through embodiments and in conjunction with the accompanying drawings.

[0018] Example: This embodiment describes an automated road cold spray marking machine, such as... Figures 1 to 4 As shown, it consists of eight parts: frame, autonomous cruise device, power unit, cleaning device, storage device, feeding device, marking device, and energy unit.

[0019] The frame consists of a chassis 1, an outer frame 2, and an upper cover 3. The outer frame 2 is located on the upper surface of the chassis 1, and the upper cover 3 is located on the upper end of the outer frame.

[0020] The autonomous navigation module consists of a binocular depth camera 4, a LiDAR 6, and a control motherboard. The LiDAR is located on the fixed frames 5 on both sides of the front of the chassis, the binocular depth camera is located in the hollow part of the outer frame, and the control motherboard is located inside the outer frame. The autonomous navigation module uses the LiDAR for high-precision ranging and positioning and 3D environment modeling, providing long-range, high-precision geometric structures for global map construction and obstacle avoidance. Then, the depth camera supplements dense details at close range, improving scene understanding and facilitating subsequent high-precision line drawing operations. Finally, the control motherboard performs complex computational analysis to plan the most suitable route.

[0021] The power module consists of AGV steering wheels 7 and omnidirectional wheels 8, which are diagonally distributed on both sides of the chassis. The power module utilizes real-time environmental scanning with LiDAR combined with DBSCAN clustering and FICP registration algorithms to achieve high-precision positioning of ±3 mm. It eliminates dynamic interference through reflector calibration, matches 3D point cloud data with map coordinates, supports dynamic path correction and obstacle avoidance, and employs a PID algorithm to adjust the steering wheel speed and steering angle, reducing path tracking errors and achieving closed-loop power drive control.

[0022] The cleaning module consists of a rotating brush 9, a connecting rod 10, a brush motor 11, driven wheels 12, a suction pipe 13, and a suction device 14. The rotating brushes 9 are symmetrically arranged on both sides of the center of the bottom surface of the base plate. The brush motor 11 drives the rotating brushes 9 to rotate via the connecting rod 10. The driven wheels 12 are symmetrically arranged between the two rotating brushes 12, and a rotating plate is provided between the two driven wheels. The suction pipe 13 is mounted on the rotating plate and connected to the suction device 14, which is mounted on the chassis. The cleaning module uses the brush motor to drive the rotating brushes to sweep dust, small stones, dirt, and other debris to below the suction pipe. The suction device then sucks the debris into the debris box, reducing the decrease in marking accuracy and quality caused by debris during product marking. In this embodiment, the rotating brush speed is linearly related to the suction negative pressure.

[0023] The material storage module is a storage bin located at the left rear of the chassis, used to store paint.

[0024] The feeding module includes a plunger pump 16, a pulley 17, a belt 18, a feeding motor 19, and a conveying hose 20. The feeding motor 19 drives the plunger pump 16 through the pulley 17 and the belt 18. The conveying hose 20 connects the plunger pump 16 to the storage module and the plunger pump 16 to the marking module. A filter screen is installed between the storage module and the feeding module to perform preliminary filtration of the paint.

[0025] The marking module includes a marking gun 21, a nozzle 22, and an atomizing device. The upper end of the marking gun 21 is connected to the feeding module through a conveying hose 20. The atomizing device and the nozzle 22 are arranged sequentially at the lower end of the marking gun 21. The marking gun is equipped with a filter screen, a flow regulating valve, and a solenoid valve. The solenoid valve and the flow regulating valve control the on / off of the paint and regulate the paint flow rate to control the marking thickness.

[0026] The power supply module is a gasoline generator 23, which is mounted on the chassis via a multi-stage damping mechanism. This mechanism includes, from top to bottom, a high-damping rubber block, a variable-diameter helical spring, and a microporous foamed rubber pad. The high-damping rubber block dissipates energy through frictional heat generated by the rubber molecular chains, absorbing high-frequency vibrations (>50 Hz), such as engine ignition knock. The variable-diameter helical spring stores and slowly releases kinetic energy through spring deformation, buffering low-frequency, large-amplitude impacts (<10 Hz), such as road bumps. The microporous foamed rubber pad dissipates remaining energy. Furthermore, the rubber-spring combination alters the system's natural frequency, avoiding the engine's commonly used speed range (25-50 Hz corresponding to 1500-3000 rpm), effectively suppressing resonance.

[0027] In this embodiment, the automated road cold spray marking machine uses high-precision environmental perception and intelligent control as its core to construct a closed-loop system covering the entire "perception-decision-execution" chain: First, it scans the environment using LiDAR and fuses data from a depth camera to achieve sub-centimeter-level 3D modeling (±3mm positioning accuracy) and dynamic obstacle recognition. Combined with the DBSCAN clustering algorithm and FICP registration technology, it completes real-time point cloud matching, providing a precise spatial reference for path planning. Subsequently, based on the PID closed-loop control model, it dynamically analyzes the steering angle, drive torque, and speed parameters, simultaneously controlling the rotation of the brush and negative pressure... The dust extraction module removes foreign objects from the ground to ensure the flatness of the working surface. During the execution phase, the paint is delivered by a plunger pump, and after two-stage filtration and flow regulation by a solenoid valve, it is uniformly sprayed with a marking thickness of ±0.2mm by an atomizing nozzle (0.5-3mm diameter adaptive adjustment). The entire process is powered by a high-power gasoline engine, and laser point cloud, visual semantics and control command data are synchronously transmitted through the EtherCAT bus. Combined with a multi-stage shock absorption structure and IP65 protection design, it ultimately achieves engineering-grade performance of fully autonomous cruise, sub-centimeter marking accuracy and 8 hours of continuous operation under complex working conditions.

[0028] Correspondingly, this embodiment also provides an automated road cold spray marking method, including: Heterogeneous data fusion is performed on data collected by LiDAR and binocular depth cameras. An improved DBSCAN clustering algorithm is used for dynamic obstacle clustering, and an improved FICP registration algorithm is used to match real-time point clouds with a preset map for real-time localization, 3D modeling, and route planning. Semantic tags (such as lane lines and traffic sign locations) are embedded in the preset map, and semantic consistency constraints are added during registration (such as point cloud features needing to match the corresponding semantic regions on the map) to suppress non-rigid deformation or interference from dynamic objects. The AGV steering wheel is controlled by a PID closed-loop control method to drive the automated road cold spray marking machine to travel along the planned route; the PID closed-loop control method includes path deviation calculation and three-loop control of position loop, speed loop and current loop; The cleaning module removes foreign objects from the ground; The line drawing module is used to draw lines.

[0029] By fusing heterogeneous data, the spatial complementarity of long-range sparse point clouds acquired by LiDAR and close-range dense depth maps and RGB semantics acquired by binocular depth cameras is achieved. The time difference and coordinate system difference between LiDAR and binocular depth camera data are eliminated to achieve temporal synchronization, providing a foundation for building a unified environment model.

[0030] Heterogeneous data fusion comprises five steps: data acquisition, spatiotemporal alignment, data-level fusion, feature-level fusion, and environmental model output. Spatiotemporal alignment includes time synchronization and spatial calibration. Time synchronization is achieved at the hardware level using PPS pulse signals to achieve μs-level synchronization, while interpolation is used at the software level to compensate for residuals. Spatial calibration is implemented using the ROS-based lidar_camera_calibration package. Data-level fusion includes point cloud densification and coordinate unification. Point cloud densification involves converting the RGB-D depth map into a pseudo-point cloud (using the PCL library depth_image_to_laser_scan).

[0031] KD-Tree is used to accelerate the matching of LiDAR point clouds and pseudo point clouds, filling the near-field blind zone (point cloud density is increased by 5 times in areas <5m); coordinates are unified by transforming RGB-D point clouds to the LiDAR coordinate system. Feature-level fusion is used to achieve semantic and geometric association, converting the geometric features (curvature, normal vector) of the point cloud acquired by LiDAR and the YOLOv7 semantic segmentation results acquired by the binocular depth camera into dense point clouds with semantic labels (such as "vehicle-dynamic" and "curb-static").

[0032] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm suitable for obstacle segmentation and dynamic assessment in point cloud data. Its core logic is to group high-density connected points into clusters and label low-density regions as noise. This embodiment improves the DBSCAN clustering algorithm to address the problems of traditional DBSCAN clustering algorithms, which require manual setting of density thresholds when used for static point cloud clustering, are insufficiently adaptable to density changes in dynamic environments (such as different distances and occlusion), and are prone to mis-clustering (conflating dynamic obstacles with static environments) or missed clustering (failing to identify obstacles when the point cloud is sparse). These problems include: Acquire dynamic environmental point cloud data collected by lidar and extract the distance information from each point in the point cloud to the lidar.

[0033] Based on the distance information, the density threshold parameter is dynamically adjusted to increase the density of the near-field point cloud and decrease the density of the far-field point cloud, thereby achieving dynamic parameter adaptation.

[0034] Remove noise points from the lidar.

[0035] A point cloud spatial index is constructed to reduce the neighborhood search complexity to O(nlogn), and the neighborhood density of each point cloud is calculated in parallel using GPUs; this improves computational efficiency and enhances the real-time performance of high frame rate point clouds for LiDAR.

[0036] Determine whether the number of points within the preset neighborhood radius ε is greater than or equal to the preset minimum clustering point number MinPts. If so, it is determined as a core point; otherwise, it is marked as a noise point. Expand the clusters corresponding to the core points to generate clusters and calculate the cluster centroids; Calculate the centroid displacement between adjacent frames. If the centroid displacement is greater than the preset dynamic threshold, the cluster corresponding to the centroid is marked as a dynamic obstacle. Otherwise, it is marked as a static obstacle. This eliminates dynamic interference (such as mistakenly identifying the shadow of a stationary vehicle or a temporary occlusion as an obstacle) and improves the robustness of clustering. The improved DBSCAN clustering algorithm also uses semantic information to assist clustering based on data acquired by binocular depth cameras (such as recognizing semantic labels such as "vehicles" and "pedestrians"), filters out non-obstacle point clouds (such as vegetation and guardrails), and reduces the false alarm rate.

[0037] Traditional FICP (Fast Iterative Closest Point) registration algorithms are feature-point-based (e.g., planes, edges) registration algorithms. If feature extraction is inaccurate or scene features are scarce (e.g., low-texture road surfaces), cumulative errors easily occur, resulting in positioning accuracy typically at the centimeter level (around 10cm), which is insufficient to meet sub-centimeter requirements. Furthermore, registration is easily affected by moving objects (e.g., pedestrians, vehicles), leading to positioning drift. This embodiment improves the FICP registration algorithm to address these problems, including: Using the clustering results of the improved DBSCAN clustering algorithm, the detected dynamic obstacle point clouds are removed, while the static point cloud features are retained, thus avoiding registration interference from moving objects. Multi-scale feature descriptors (such as combining point cloud normal vectors, curvature, height difference and other multi-dimensional information) are used to extract strong geometric feature points (such as corners, curbs, etc.) to reduce registration ambiguity in weak feature regions. Layered registration includes coarse registration and fine registration; the coarse registration includes using IMU pre-integration results or odometry data to provide an initial pose to reduce the pose search space; the fine registration includes introducing LM optimization during the FICP iteration process and combining point-to-surface error to improve convergence accuracy. The improved FICP registration algorithm introduces pose continuity constraints during the iteration process and eliminates accidental mismatches through sliding window optimization.

[0038] The PID closed-loop control method includes path deviation calculation and three-loop control consisting of a position loop, a speed loop, and a current loop. Path deviation calculation obtains the AGV pose (x, y, θ) through point cloud matching and transforms the global coordinates (x, y) to the path coordinate system. (Refer to path functions) Finally, calculate the corresponding heading deviation. lateral deviation curvature compensation K is the path curvature, and v is the vehicle speed.

[0039] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. An automated road cold spray marking machine, comprising a frame consisting of a chassis, an outer frame, and a top cover, characterized in that: include: The autonomous navigation module, mounted on the rack, includes a binocular depth camera, a lidar, and a control motherboard, and is used to acquire environmental information and plan a route. The power module, mounted on the chassis, includes an AGV steering wheel and omnidirectional wheels, which are diagonally distributed on both sides of the chassis for traveling according to a planned route. The cleaning module, mounted on the chassis, includes a rotating brush and a vacuuming device for removing dust from the road surface; The material storage module, mounted on the chassis, is used to store paint. The feeding module, mounted on the chassis and connected to the storage module, is used to transport the coating. The marking module, set on the outer frame, includes a marking gun, a nozzle, and an atomizing device. The marking gun is connected to the feeding module and is used to spray paint for marking lines. The power supply module provides the energy required for the automated road cold spray marking machine.

2. The automated road cold spray marking machine according to claim 1, characterized in that: The chassis has a fixed frame on each side of the front end, and the lidar is installed in the fixed frame; the front center of the outer shell has a hollowed-out section, and the binocular depth camera is installed in the hollowed-out section; the control motherboard is installed inside the outer frame.

3. The automated road cold spray marking machine according to claim 1, characterized in that: The cleaning module also includes a brush motor, a connecting rod, driven wheels, and a suction pipe. The rotating brushes are symmetrically arranged on both sides of the center of the bottom surface of the chassis. The brush motor drives the rotating brushes to rotate through the connecting rod. The driven wheels are symmetrically arranged between the two rotating brushes. A rotating plate is provided between the two driven wheels. The suction pipe is arranged on the rotating plate and connected to the suction device. The suction device is arranged on the chassis.

4. The automated road cold spray marking machine according to claim 1, characterized in that: The feeding module includes a plunger pump, a pulley, a belt, a feeding motor, and a conveying hose. The feeding motor drives the plunger pump through the pulley and belt. The conveying hose connects the plunger pump to the storage module and the plunger pump to the marking module.

5. The automated road cold spray marking machine according to claim 1, characterized in that: The upper end of the marking gun is connected to the feeding module via a conveying hose. The atomizing device and the nozzle are sequentially arranged at the lower end of the marking gun. The marking gun is equipped with a filter screen, a flow regulating valve, and a solenoid valve.

6. The automated road cold spray marking machine according to claim 1, characterized in that: A filter screen is provided between the storage module and the feeding module.

7. The automated road cold spray marking machine according to claim 1, characterized in that: The power supply module is a gasoline generator, which is mounted on the chassis via a multi-stage damping mechanism. The multi-stage damping mechanism includes a high-damping rubber block, a variable-diameter helical spring, and a microporous foamed rubber pad arranged sequentially from top to bottom.

Images