Heavy vehicle sub-type identification method based on tire spatial topology analysis

Abstract

The application discloses a heavy vehicle sub-type identification method based on tire space topology analysis, the most bottom layer of the vehicle is detected, a non-directional topology graph is constructed with the tire as a node and the space connection relationship as an edge, the intersection and union ratio (IOU) and / or the center distance characteristics between the tires are analyzed, the specific structure such as "three shafts" and "double shafts" is inferred based on the connected component scale of the topology graph, so that the axle type and the sub-type of the vehicle are accurately determined, and the problem that the prior art neglects the deep mining of the tire "space topology relationship", only stays on the overall appearance or the simple size level, and leads to the poor accuracy of the heavy vehicle sub-type identification is solved.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and intelligent traffic monitoring technology, specifically relating to a method for identifying the sub-types of heavy vehicles based on tire spatial topology analysis. Background Technology

[0002] In autonomous driving and intelligent transportation systems, accurately identifying the specific categories of large and medium-sized trucks is crucial for traffic flow analysis, traffic restriction enforcement, and safety warnings. Existing detection methods and their shortcomings are as follows:

[0003] Deep learning-based vehicle classifier methods: When using CNN networks (such as ResNet, VGG, etc.) to perform end-to-end classification of the overall vehicle image, since they rely on the recognition of the overall appearance features of the vehicle, heavy trucks and container trucks often have similar rectangular outlines in appearance. Especially when fully loaded and empty, or when the angle of the truck head and trailer is deflected, the classifier is prone to confusion and it is difficult to accurately distinguish the specific vehicle level (such as distinguishing between ordinary heavy trucks and 100-ton trucks / container trucks).

[0004] Radar / LiDAR-based dimensional calibration method: When using radar or lidar to measure the physical dimensions of a vehicle, such as its length, height, and wheelbase, vehicles of the same length may have different axle configurations (e.g., a 12-meter vehicle may be a two-axle or a three-axle vehicle). Relying solely on length cannot accurately determine the vehicle's load-bearing capacity and legal type.

[0005] Radar / laser measurement size calibration methods measure physical dimensions and combine them with deep learning vehicle model classifier results for calibration and correction. In congested or following traffic scenarios, most of the vehicle body may be obscured, causing the classifier to be unable to obtain complete vehicle appearance features, and radar ranging may also be interfered with. Summary of the Invention

[0006] The technical problem that this invention aims to solve is that the number of axles and tire combination forms (such as single axle, double axle, triple axle) of a vehicle are core physical characteristics for determining the vehicle's load-bearing capacity and legal type. Current classifiers and size calibration methods neglect in-depth exploration of the tire's "spatial topological relationship" and only focus on the overall appearance or simple dimensions.

[0007] To address the aforementioned technical problems, the present invention provides a method for identifying the sub-types of heavy vehicles based on tire spatial topology analysis, comprising the following steps: Target detection is performed on the input image to obtain the overall vehicle detection bounding box; The overall vehicle detection frame is enlarged according to a preset ratio, high-resolution tire regions of interest are extracted, a tire-specific detection model is run within the tire ROI area, tire categories are further identified, and bounding boxes of all visible tires are obtained. The bounding frames of all visible tires are expanded outward according to a preset expansion ratio to obtain the expanded bounding frames; Each detected tire is treated as a tire node in the topology graph. If the intersection-union ratio of any two expanded bounding boxes is greater than a preset threshold, and / or the center distance between any two expanded bounding boxes is less than a preset distance threshold, a connecting edge is established between the two tire nodes. An undirected topology graph is constructed based on the tire nodes and the connecting edge. Perform connectivity analysis on the topology graph and infer hierarchical vehicle types based on the maximum connectivity.

[0008] Preferably, after obtaining the bounding boxes of all visible tires, the vertex positions of the bounding boxes of all visible tires are calculated using the vector cross product principle in combination with the vehicle's driving direction. Only the bounding boxes of tires located in the vehicle chassis area are retained, while the spare tire or falsely detected targets on the vehicle body are filtered out.

[0009] Preferably, a tire adjacency matrix is ​​constructed based on the connectivity relationships in the undirected topology graph, the connectivity of each node is calculated, and a hierarchical vehicle type inference is performed based on the maximum connectivity.

[0010] Preferably, if the maximum connectivity display contains at least 3 interconnected tire nodes, the vehicle type is inferred to be a container truck or a heavy-duty truck; if the maximum connectivity display only contains at least 2 interconnected tire nodes, the vehicle type is inferred to be a heavy-duty truck.

[0011] Preferably, if the vehicle type cannot be inferred from the maximum connectivity, the determination is made based on the total number of tires: if the total number of tires is less than a preset first threshold, it is a small truck; if the total number of tires is greater than or equal to the first threshold and less than a second threshold, it is a medium truck; and if the total number of tires is greater than or equal to the second threshold, it is a heavy truck.

[0012] This invention provides a method for identifying the sub-types of heavy vehicles based on tire spatial topology analysis. By enlarging the overall vehicle detection frame, the bounding boxes of all visible tires are obtained and expanded using a tire-specific detection model. Each detected tire is treated as a tire node in the topology graph. Connecting edges are established under preset connectivity conditions. An undirected topology graph is constructed based on the tire nodes and the connecting edges. Connectivity analysis is performed on the topology graph, and hierarchical vehicle type inference is performed based on the maximum connectivity. This method solves the problem that existing technologies neglect in-depth exploration of the "spatial topology relationship" of tires, only focusing on the overall appearance or simple size, resulting in poor accuracy in identifying the sub-types of heavy vehicles. Attached Figure Description

[0013] Figure 1 A decision flow graph based on tire topology analysis; Figure 2 This is a schematic diagram of vehicle ROI extraction and tire inspection. Figure 3 This is a schematic diagram of tire frame expansion and IOU overlap analysis; Figure 4 A comparison of the topological differences between three-wheel (container truck) and two-wheel (heavy truck) trucks; Figure 5 This is a side view of the installation diagram; Figure 6 This is a top-down view of the installation. Detailed Implementation

[0014] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0015] like Figure 1 As shown, the heavy vehicle sub-type identification method based on tire spatial topology analysis provided in this embodiment of the invention includes the following steps: Step 1: Perform object detection on the input image to obtain the overall vehicle detection bounding box, such as... Figure 2 As shown in 'a'.

[0016] Step 2: To improve tire detection rate, the overall vehicle detection frame is enlarged by a preset ratio (1.2 times), and a high-resolution tire region of interest (ROI) is extracted to preserve the surrounding context information of the tire and improve the resolution of local areas, such as... Figure 2 As shown in b, a detection model to identify tire categories is run within the tire ROI region, and the bounding boxes of all visible tires are obtained, as shown in Figure b. Figure 2 As shown in c in the figure.

[0017] After obtaining the bounding boxes of all visible tires, the vertex positions of the bounding boxes of all visible tires are calculated using the vector cross product principle, taking into account the vehicle's driving direction. Only the bounding boxes of tires located in the vehicle chassis area are retained, while the spare tire or falsely detected targets on the vehicle body are filtered out.

[0018] Since the detected tire surround typically only contains the rubber portion, the tire surround needs to be geometrically expanded in order to analyze the physical connection between tires.

[0019] Step 3: For all visible tires, define their bounding boxes (e.g., ...). Figure 3 As shown in a) the bounding box is expanded outward according to a preset expansion ratio (30%) to obtain the expanded bounding box, as shown in the figure. Figure 3 As shown in b in the figure.

[0020] Step 4: Using the camera's relative position information, construct the influence domain of the tires. Within the influence domain, treat each detected tire as a tire node in the topology graph. Establish a connecting edge between two tire nodes if the Intersection over Union (IOU) of any two expanded bounding boxes is greater than a preset threshold, and / or the center distance between any two expanded bounding boxes is less than a preset distance threshold. Figure 3 As shown in c), an undirected topology graph is constructed based on the tire nodes and connected edges.

[0021] The formula for determining whether the Intersection over Union (IOU) ratio of any two expanded bounding boxes is greater than a preset threshold is as follows:

[0022] Intersection-to-Union Ratio = Overlap Area / Union Area

[0023] The formula for determining if the center distance between any two expanded bounding boxes is less than a preset distance threshold is as follows:

[0024]

[0025] Where K = 1.5, The width of the two expanded tire surround frames. This is the distance threshold.

[0026] Step 5: Perform connectivity analysis on the topology graph and infer the hierarchical vehicle type based on the maximum connectivity.

[0027] A tire adjacency matrix is ​​constructed based on the connectivity relationships in the undirected topology graph, and the connectivity of each node is calculated to perform connectivity analysis on the topology graph.

[0028] If the maximum connectivity indicates at least three interconnected tire nodes, it is determined to be a "triple-axle" structure, and the vehicle type is inferred to be a container truck or a heavy-duty truck. Figure 4 As shown in 'a'.

[0029] If the maximum connectivity display only shows at least two interconnected tire nodes, it is determined to be a "twin-axle" structure, and the vehicle type is inferred to be a heavy-duty truck. Figure 4 As shown in b in the figure.

[0030] If the above inference conditions are not met, the determination is based on the total number of tires. If the total number of tires is less than the preset first threshold, it is a small truck; if the total number of tires is greater than or equal to the first threshold and less than the second threshold, it is a medium truck; and if the total number of tires is greater than or equal to the second threshold, it is a heavy truck.

[0031] Example 1: Identifying Container Trucks 1. Input: A snapshot of a highway containing a container truck.

[0032] 2. Detection: The vehicle body frame was detected, and the Region of Interest (ROI) was captured. Five tires were detected within the ROI.

[0033] 3. Topology analysis: Inflate the five tires by 30%.

[0034] Calculations revealed that after the three rear tires expanded, there was an IOU overlap between each pair (OverlapNo>= 2).

[0035] The existence of a "three-wheel" structure was determined.

[0036] 4. Output: Recognized as a container truck (although the body may be partially obscured by another vehicle).

[0037] Example 2: Identifying Ordinary Heavy-Duty Trucks 1. Input: Image of a national highway checkpoint, including a heavy dump truck.

[0038] 2. Inspection: 4 tires were detected.

[0039] 3. Topology analysis: After expansion and calculation, it was found that there is overlap between every two tires in the rear wheel area, but they do not form 3 connected areas.

[0040] The presence of a "double wheel" structure was determined.

[0041] 4. Output: Identified as a heavy truck.

[0042] The innovative aspects of this invention are as follows: 1. Invariant Feature Identification Based on Tire Spatial Topology: A topological graph analysis method is proposed, using tires as nodes and the spatial connectivity between tires as edges. The number of axles and axle type (e.g., single axle, double axle, triple axle) are used as core classification features. These features are more stable than vehicle appearance and are less affected by modifications, loads, additional equipment on the vehicle body, or changes in vehicle paint.

[0043] 2. Robust inference capability for unknown vehicle models: Traditional classifiers based on overall appearance are prone to misclassification when encountering unfamiliar functional vehicles or modified vehicles; while axle type and tire combination are fixed attributes at the vehicle engineering and regulatory level, and tire topology analysis can reliably determine the vehicle type even with appearance variations.

[0044] 3. Reliable identification under partially visible and occluded conditions: This method relies on tire information at the vehicle chassis level. Tires are usually more visible in the shooting angle (especially at close range or from a top-down camera), so it can still provide a reliable classification basis when the vehicle body is occluded or the view is incomplete.

[0045] 4. Lightweight graph structure reasoning, easy to deploy at the edge: The topology graph construction is based solely on the geometric expansion of the tire frame, IOU and center distance calculation. Compared with full graph depth classification or segmentation models, it has less computational cost and is suitable for real-time operation on embedded or edge devices.

[0046] 5. Can be fused with multimodal features such as size / wheelbase: The method supports parallel or posterior fusion of tire topology results with length, wheelbase or radar / laser measurement results, further improving the accuracy of sub-classification and reducing misjudgments.

[0047] The design principles and implementation rationale of the embodiments of the present invention are as follows: • Vehicle appearance can change significantly with model updates, functional modifications, or operational uses (e.g., adding tarpaulins, modifying cargo boxes, painting vehicle advertisements, etc.). Classifiers that rely on appearance texture or overall outline are not robust enough when encountering unseen samples. In contrast, the number of axles and axle type are basic invariants in vehicle engineering design and rarely change, making them more reliable features for determining vehicle load rating and regulatory category.

[0048] • Tires, as the components that directly contact the vehicle with the ground, are more likely to remain observable when obscured or viewed from a partial angle. By locating and analyzing the spatial distribution of tires, core structural information of the vehicle can be recovered even when complete exterior information of the vehicle body is missing.

[0049] • The reason for using a topology graph instead of a simple count is that different configurations with the same number of tires (e.g., multiple single-axle distributions vs. tightly combined twin-axles) correspond to different axle types and load characteristics. Topology connectivity analysis can distinguish these differences, thus avoiding the ambiguity caused by relying solely on the total number of tires.

[0050] • To balance detection errors with the actual structure, the design incorporates tire frame expansion (expanding the influence domain) and connectivity determination based on IOU / center distance, which can tolerate single tire detection position deviations or partial occlusions. At the same time, the probability of false connections is reduced through driving direction filtering and spare tire rejection rules.

[0051] • Differences in the field camera's viewing angle, resolution, and installation height can affect the detection scale. Therefore, the system adopts an adaptive dilation ratio and distance threshold, combined with an online statistical correction mechanism, so that the method can operate stably for a long time in different checkpoints or road sections.

[0052] The beneficial effects of the embodiments of the present invention are as follows: 1. Extremely high segmentation accuracy: By recognizing the strong feature of "three-wheeled" trucks, it can distinguish between ordinary trucks and heavy-duty trucks with 100% accuracy, solving the problem that traditional trucks cannot be distinguished by vehicle length alone.

[0053] 2. Anti-obstruction and robustness: Even if the top of the vehicle is obscured, as long as the wheels are visible, the vehicle model can be inferred by observing the tire combination on the chassis. This is of great advantage in close-range capture at checkpoints or in tunnel scenarios.

[0054] 3. Computational resource saving: Compared with full image segmentation, only component detection and simple IOU geometric calculation are performed on the vehicle ROI region, resulting in minimal computational load, making it suitable for embedded edge deployment.

[0055] 4. Conforms to physical principles: Direct measurement of the axle structure, which determines the vehicle's load-bearing capacity, is closer to the vehicle's essential properties than "appearance speculation" and has strong interpretability.

[0056] like Figure 5 and Figure 6 As shown, to ensure stable observation of the tire chassis area, a lane-side shooting configuration is recommended: install one side-view monitoring camera on each side of the lane (one pair per lane, one on the left and one on the right). The equipment requirements and camera installation recommendations are as follows: • Camera position: Installed symmetrically on both sides of the lane, with the center line of the lens parallel to the center line of the lane, and the field of view covering the entire width of the lane and including the chassis area; • Installation height and viewing angle: The recommended installation height is 0.8–1.5 meters, and the side view angle ensures good visibility of the tire in the image without causing severe perspective distortion; • Resolution and frame rate: A resolution of at least 1920×1080 and a frame rate of 30fps or higher are recommended to ensure controllable motion blur and improve tire detection rate. • Calibration and Alignment: The camera needs to be calibrated for internal and external parameters and aligned to the lane coordinate system through lane line calibration or calibration board in order to map the tire detection position to a unified reference system; • Redundancy and Degradation: One unit on each side can provide redundant observations when either side is blocked or missing. The system supports single-sided degradation identification, but simultaneous availability will improve the confidence level of the judgment. • Resources and Deployment: Since the algorithm mainly focuses on component detection and simple geometric calculations within the ROI, it is recommended to deploy inference nodes on edge devices or local servers to meet real-time requirements.

Claims

1. A method for identifying the sub-types of heavy vehicles based on tire spatial topology analysis, characterized in that, Includes the following steps: Target detection is performed on the input image to obtain the overall vehicle detection bounding box; The overall vehicle detection frame is enlarged according to a preset ratio, high-resolution tire regions of interest are extracted, a tire-specific detection model is run within the tire ROI area, tire categories are further identified, and bounding boxes of all visible tires are obtained. The bounding frames of all visible tires are expanded outward according to a preset expansion ratio to obtain the expanded bounding frames; Each detected tire is treated as a tire node in the topology graph. If the intersection-union ratio of any two expanded bounding boxes is greater than a preset threshold, and / or the center distance between any two expanded bounding boxes is less than a preset distance threshold, a connecting edge is established between the two tire nodes. An undirected topology graph is constructed based on the tire nodes and the connecting edge. Perform connectivity analysis on the topology graph and infer hierarchical vehicle types based on the maximum connectivity.

2. The method for identifying heavy-duty vehicle sub-types based on tire spatial topology analysis as described in claim 1, characterized in that, After obtaining the bounding boxes of all visible tires, the vertex positions of the bounding boxes of all visible tires are calculated using the vector cross product principle, taking into account the vehicle's driving direction. Only the bounding boxes of tires located in the vehicle chassis area are retained, while the spare tire or falsely detected targets on the vehicle body are filtered out.

3. The method for identifying heavy-duty vehicle sub-types based on tire spatial topology analysis as described in claim 1, characterized in that, A tire adjacency matrix is ​​constructed based on the connectivity relationships in the undirected topological graph. The connectivity of each node is calculated, and the vehicle type is inferred hierarchically based on the maximum connectivity.

4. The method for identifying heavy-duty vehicle sub-types based on tire spatial topology analysis as described in claim 3, characterized in that, If the maximum connectivity display contains at least 3 interconnected tire nodes, the vehicle type is inferred to be a container truck or a heavy-duty truck. If the maximum connectivity display only contains at least 2 interconnected tire nodes, the vehicle type is inferred to be a heavy-duty truck.

5. The method for identifying heavy-duty vehicle sub-types based on tire spatial topology analysis as described in claim 3, characterized in that, If the vehicle type cannot be determined by the maximum connectivity, it is determined based on the total number of tires. If the total number of tires is less than a preset first threshold, it is a small truck; if the total number of tires is greater than or equal to the first threshold but less than the second threshold, it is a medium truck; and if the total number of tires is greater than or equal to the second threshold, it is a heavy truck.

Images