Method and device for detecting state of traffic signboard and electronic equipment

Abstract

The invention relates to the technical field of traffic signboard detection, in particular to a method and device for detecting the state of a traffic signboard and electronic equipment. The method for detecting the state of the traffic signboard comprises the following steps: acquiring a detection image corresponding to the traffic signboard acquired by acquisition equipment; the detection image is input into the optimized cascade type double-model architecture to obtain a detection result, the detection result comprises the state of the traffic signboard, and the state of the traffic signboard is at least one of normal, inclined and damaged; wherein the cascaded double-model architecture is constructed and formed through a detection model and a classification model. According to the technical scheme of the invention, the state of the traffic signboard can be accurately obtained, so that maintenance personnel can timely obtain the traffic signboard in an inclined state or a damaged state, the processing efficiency can be improved, the traffic accident risk caused by the abnormal state (inclined state or damaged state) of the traffic signboard can be reduced, and the safety of the maintenance personnel can be improved. The road safety level is improved.

Description

Technical Field

[0001] This invention relates to the field of traffic sign detection technology, and in particular to a method, apparatus, and electronic device for detecting the status of traffic signs. Background Technology

[0002] Real-time detection and status recognition of traffic signs are crucial for intelligent transportation systems, autonomous driving, and road infrastructure maintenance. Accurate traffic signs improve driving safety, while status analysis (such as tilting or damage) helps in timely repairs and avoids potential risks.

[0003] However, existing traffic sign detection and classification systems rely on shared backbone networks, requiring feature extractors to consider both localization and semantic information, leading to decreased detection accuracy in complex scenarios. Furthermore, current technologies are largely limited to detecting traffic signs in normal states, lacking the ability to recognize traffic signs in various physical states. Therefore, there is an urgent need for a technical solution that can both locate traffic signs in real time and accurately identify their physical states to meet the high standards required by intelligent transportation systems. Summary of the Invention

[0004] To address, or at least partially address, the aforementioned technical problems, this invention provides a method, apparatus, and electronic device for detecting the status of traffic signs. This allows for accurate acquisition of the traffic sign's status, enabling maintenance personnel to promptly identify signs that are tilted or damaged, and to perform timely maintenance. This ensures the traffic signs are returned to a normal state, reduces manual inspection costs, shortens response time, improves processing efficiency, lowers the risk of traffic accidents caused by abnormal traffic sign conditions (tilted or damaged), and enhances road safety.

[0005] In a first aspect, the present invention provides a method for detecting the state of a traffic sign, comprising:

[0006] Acquire the detection images captured by the acquisition device;

[0007] The detected image is input into the optimized cascaded dual-model architecture to obtain the confidence that the detected image contains a traffic sign;

[0008] The target detection image is determined based on the confidence level, and the target detection image is detected to obtain the detection result. The detection result includes the status of the traffic sign, and the status of the traffic sign is at least one of normal, tilted, and damaged.

[0009] The cascaded dual-model architecture is formed by constructing a detection model and a classification model.

[0010] In some embodiments, the optimized cascaded dual-model architecture is obtained through the following method:

[0011] The detection model and the classification model are designed into a cascaded dual-model architecture through task decoupling, and the cascaded dual-model architecture is imported into the model training platform;

[0012] Collect a dataset of detection image samples and preprocess the dataset.

[0013] The preprocessed detection image sample dataset is input into the model training platform;

[0014] The model training platform outputs the state of the traffic sign in the detected image to train and obtain an optimized cascaded dual-model architecture.

[0015] In some embodiments, a single detected image sample includes multiple traffic signs; the cascaded dual-model architecture employs the CloU-Wise algorithm, which dynamically adjusts the matching threshold of positive samples based on the density of the multiple traffic signs;

[0016] Among them, the positive sample is defined as the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box of the traffic sign that is greater than the matching threshold.

[0017] In some embodiments, the density of traffic signs in a single detected image sample is inversely proportional to the matching threshold.

[0018] In some embodiments, determining a target detection image based on the confidence level and detecting the target detection image to obtain a detection result includes:

[0019] The confidence level of the detected image containing traffic signs is compared with a preset confidence level;

[0020] Retain detection images with a confidence level greater than the preset confidence level;

[0021] The classification model is used to perform state recognition on the traffic signs in the retained detection images to obtain the detection results.

[0022] In some embodiments, the detection model is the YOLOv8n model, and the classification model is the YOLOv8n-cls model.

[0023] In some embodiments, the detection model includes a Backbone module and a PECA attention module.

[0024] Secondly, the present invention also provides a detection device for the status of a traffic sign, comprising:

[0025] The first acquisition module is used to acquire the detection images collected by the acquisition device;

[0026] The second acquisition module is used to input the detected image into the optimized cascaded dual-model architecture to obtain the confidence that the detected image contains traffic signs;

[0027] The third acquisition module is used to determine the target detection image based on the confidence level, and to detect the target detection image to obtain the detection result. The detection result includes the status of the traffic sign, and the status of the traffic sign is at least one of normal, tilted, and damaged.

[0028] The cascaded dual-model architecture is formed by constructing a detection model and a classification model.

[0029] Thirdly, the present invention also provides an electronic device, including a processor and a memory, wherein the processor executes the steps of the method for detecting the state of a traffic sign as described in the first aspect by calling a program or instruction stored in the memory.

[0030] The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art:

[0031] The method for detecting the state of traffic signs provided in this invention includes: acquiring a detection image corresponding to the traffic sign collected by a data acquisition device; inputting the detection image into an optimized cascaded dual-model architecture to obtain a detection result, wherein the detection result includes the state of the traffic sign, and the state of the traffic sign is at least one of normal, tilted, and damaged; wherein the cascaded dual-model architecture is constructed by a detection model and a classification model. By using a detection model and a classification model to construct a cascaded dual-model architecture, and through task decoupling design, the detection model focuses on spatial localization, and the classification module focuses on optimizing state recognition, thereby avoiding the problem in related technologies where the detection and classification share a backbone network, causing the feature extractor to need to consider both localization and semantic information, resulting in a decrease in the detection accuracy of traffic signs in complex scenarios. This allows for accurate identification of the status of traffic signs, enabling maintenance personnel to promptly identify and address signs that are tilted or damaged. This ensures that traffic signs are returned to normal, reducing manual inspection costs, shortening response times, improving efficiency, and lowering the risk of traffic accidents caused by signs in abnormal conditions (tilted or damaged), thus enhancing road safety. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 A flowchart illustrating a method for detecting the status of a traffic sign according to an embodiment of the present invention;

[0035] Figure 2 A system architecture diagram provided for an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of a device for detecting the status of a traffic sign, provided in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0038] To better understand the above-mentioned objectives, features, and advantages of the present invention, the solutions of the present invention will be further described below. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0039] Many specific details are set forth in the following description in order to provide a full understanding of the invention, but the invention may also be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the invention, and not all embodiments.

[0040] The traffic sign status detection method provided in this invention employs a cascaded dual-model architecture, consisting of a detection model and a classification model. Through task decoupling, the detection model focuses on spatial localization, while the classification module focuses on optimizing status recognition. This avoids the problem in related technologies where the detection and classification share a backbone network, leading to a decrease in traffic sign detection accuracy in complex scenarios due to the feature extractor needing to consider both localization and semantic information. Therefore, the method can accurately obtain the status of traffic signs, enabling maintenance personnel to promptly identify signs that are tilted or damaged and perform timely maintenance, ensuring that traffic signs are returned to normal. This reduces manual inspection costs, shortens response time, improves processing efficiency, lowers the risk of traffic accidents caused by abnormal traffic sign conditions (tilted or damaged), and enhances road safety.

[0041] The temperature detection method, temperature detection device, and electronic equipment for sintering equipment provided in the embodiments of the present invention will be described exemplarily below with reference to the accompanying drawings.

[0042] Figure 1 This is a flowchart illustrating a method for detecting the state of a traffic sign according to an embodiment of the present invention. This method is applicable to application scenarios requiring the detection of the state of traffic signs. This method can be executed by a traffic sign state detection device provided in this embodiment of the invention, which can be implemented using software and / or hardware. Figure 1 As shown, the detection of the status of the traffic sign includes the following steps:

[0043] S101. Acquire the detection image corresponding to the traffic sign collected by the acquisition device.

[0044] The data acquisition equipment can be, for example, a vehicle-mounted camera (2 megapixels @ 30FPS) or a roadside monitoring device (H.264 encoding), supporting a wide operating temperature range of -30℃ to 70℃.

[0045] Therefore, vehicle-mounted cameras or roadside monitoring equipment are used to collect images of traffic signs.

[0046] S102. Input the detected image into the optimized cascaded dual-model architecture to obtain the confidence level that the detected image contains traffic signs.

[0047] Specifically, in this step, the optimized cascaded dual-model architecture serves as the target detection model. This architecture can be deployed on an edge computing device, which may be based on, for example, Rockchip RK3588 (6 TOPS computing power) or Jetson Orin NX (100 TOPS), running the Kirin 11 system. In real-time detection using the edge computing device as the detection platform, the RK3588 hardware is used for neural network processing unit (NPU) inference acceleration to meet real-time requirements and achieve higher performance detection.

[0048] After the detection image is acquired in S101, it is directly transmitted to the edge computing device. The detection model in the cascaded dual-model architecture receives the image and uses structures such as convolutional neural networks to locate and initially identify the target in the image, determine whether there is a target box of interest in the image, and give the confidence level of the existence of a real target for each target box.

[0049] S103. The target detection image is determined based on the confidence level, and the target detection image is detected to obtain the detection result. The detection result includes the status of the traffic sign. The status of the traffic sign is at least one of normal, tilted and damaged. The cascaded dual-model architecture is formed by the detection model and the classification model.

[0050] Specifically, in this step, the detection images are filtered based on the confidence level of the detection images obtained in S102 to determine the target detection image, and then the status of the traffic sign is determined according to the classification model in the cascaded dual-model architecture.

[0051] For example, if an edge computing device detects that a traffic sign is tilted or damaged, it can output the tilted or damaged status to the inspection management platform (which supports real-time viewing of inspection results on web and mobile devices (Android / iOS) and provides an API interface to connect to the traffic management big data platform). This allows maintenance personnel to promptly identify and maintain the tilted or damaged traffic signs, ensuring they are returned to normal. This reduces manual inspection costs, shortens problem response time, improves processing efficiency, lowers the risk of traffic accidents caused by abnormal traffic signs (tilted or damaged), and enhances road safety.

[0052] In this embodiment, a cascaded dual-model architecture is constructed by using a detection model and a classification model. Through task decoupling design, the detection model focuses on spatial localization, while the classification module focuses on optimizing state recognition. This avoids the problem in related technologies where the detection and classification share the backbone network, causing the feature extractor to have to take into account both localization and semantic information, resulting in a decrease in the detection accuracy of traffic signs in complex scenarios.

[0053] The method for detecting the state of traffic signs provided in this invention includes: acquiring a detection image corresponding to the traffic sign collected by a data acquisition device; inputting the detection image into an optimized cascaded dual-model architecture to obtain a detection result, wherein the detection result includes the state of the traffic sign, and the state of the traffic sign is at least one of normal, tilted, and damaged; wherein the cascaded dual-model architecture is constructed by a detection model and a classification model. By using a detection model and a classification model to construct a cascaded dual-model architecture, and through task decoupling design, the detection model focuses on spatial localization, and the classification module focuses on optimizing state recognition, thereby avoiding the problem in related technologies where the detection and classification share a backbone network, causing the feature extractor to need to consider both localization and semantic information, resulting in a decrease in the detection accuracy of traffic signs in complex scenarios. This allows for accurate identification of the status of traffic signs, enabling maintenance personnel to promptly identify and address signs that are tilted or damaged. This ensures that traffic signs are returned to normal, reducing manual inspection costs, shortening response times, improving efficiency, and lowering the risk of traffic accidents caused by signs in abnormal conditions (tilted or damaged), thus enhancing road safety.

[0054] In some embodiments, the optimized cascaded dual-model architecture is obtained through the following methods:

[0055] The detection model and classification model are designed into a cascaded dual-model architecture through task decoupling, and the cascaded dual-model architecture is imported into the model training platform;

[0056] Collect a dataset of detection image samples and preprocess the dataset.

[0057] The preprocessed detection image sample dataset is input into the model training platform;

[0058] The model training platform outputs the state of the corresponding traffic sign in the detected image to train and obtain the optimized cascaded dual-model architecture.

[0059] Specifically, a cascaded dual-model architecture is designed through task decoupling. This architecture is then imported into the model training platform as a pre-trained model. The collected detection image sample dataset is pre-processed and input into the model training platform. The model training platform optimizes the model performance through a cascaded interaction mechanism, thereby training and optimizing the cascaded dual-model architecture.

[0060] The detection image sample dataset consists of detection images of traffic signs under different weather and lighting conditions. The physical states of the traffic signs include normal, tilted, and damaged.

[0061] The preprocessing of the detected image sample dataset includes data annotation and classification labeling. Specifically, the data annotation adopts the COCO format, a commonly used data annotation format. It stores annotation information in JSON file format, containing image-related information (such as filename, size, etc.), target category information, and bounding box coordinates for each target. This format facilitates data sharing and processing in different computer vision tasks such as object detection and segmentation. Many open-source algorithms and tools also support the COCO format, facilitating subsequent model training and development. The bounding boxes are accurate to the pixel level (error < 2px). When annotating target objects in an image, a rectangular box (bounding box) is used to enclose the object. Pixel-level accuracy (error < 2px) means that the error between the bounding box coordinates and the actual object boundary must be controlled within 2 pixels. For example, if the actual top-left corner coordinates of an object are (100, 100), the labeled coordinates should be within the range of (98-102, 98-102) to ensure label accuracy. This allows the model to learn the position and shape information of the target object more accurately during training. Classification labels are used to categorize the state of the target object (e.g., 0 - normal; 1 - tilted; 2 - damaged). Through these classification labels, the model can learn the characteristics of different states and achieve the recognition of the target object's state.

[0062] The model was trained using the PyTorch framework on the Windows system, with a training period of 400 epochs and an initial learning rate of 0.01. Label Smoothing (label_smoothing=0.1) was used to alleviate the imbalance of samples.

[0063] In some embodiments, the cascading interaction mechanism includes dynamic ROI filtering and feedback training.

[0064] Specifically, dynamic ROI (Region of Interest) filtering first selects high-quality candidate boxes from the detection model's output (based on confidence, size, and non-maximum suppression), retaining only valid regions for input into the classification model to reduce computational redundancy. Feedback training then uses misclassified samples from the classification model (such as misclassification of damaged symbols) to enhance the detection model's learning of these difficult samples (e.g., increasing loss weights or data augmentation), forming a closed loop of detection-classification-optimization. The model must simultaneously meet the following criteria to achieve satisfactory performance: detection mAP@0.5≥89.7%, classification accuracy≥91.2%, real-time performance ≥45FPS on the RK3588 chip, and generalization ability verified through independent test sets and extreme scenarios (e.g., performance fluctuation <10% under rain and fog). Here, mAP@0.5≥89.7% indicates that when the IoU between the predicted and ground truth boxes is ≥0.5, the model's average AP across all categories exceeds 89.7%, representing an extremely high accuracy level.

[0065] In some embodiments, a single detected image sample includes multiple traffic signs; the cascaded dual-model architecture employs the CloU-Wise algorithm, which dynamically adjusts the matching threshold of positive samples based on the density of multiple traffic signs; wherein, a positive sample is defined as the intersection-union ratio (IU / W) between the predicted bounding box and the ground truth bounding box of a traffic sign that is greater than the matching threshold.

[0066] Specifically, the CloU-Wise algorithm is a strategy used to optimize positive and negative sample matching in a dual-cascaded dual-model architecture. It is based on an improved IoU (Intersection over Union) calculation method (such as CIoU or CloU) and comprehensively considers the overlapping area, the distance between the center points, and the aspect ratio.

[0067] In this embodiment, the density of traffic signs in a detected image sample is inversely proportional to the matching threshold. For example, an image may contain multiple densely arranged traffic signs, some of which are damaged (e.g., cracked, faded). In related technologies, a fixed matching threshold is set, for example, 0.5. If the IoU between traffic sign A and the ground truth bounding box is 0.45 (below the threshold of 0.5), it is judged as a negative sample, and the model cannot learn its features. However, in this embodiment, the CloU-Wise algorithm is used to dynamically adjust the matching threshold. When there are multiple densely arranged traffic signs in an image, the matching threshold is lowered to 0.4. If the IoU between traffic sign A and the ground truth bounding box is 0.45 (greater than the threshold of 0.4), it is matched as a positive sample, and traffic sign A is given double the loss weight, significantly optimizing the model's prediction results.

[0068] In some embodiments, determining a target detection image based on the confidence level and detecting the target detection image to obtain a detection result includes:

[0069] The confidence level of the detected image containing traffic signs is compared with a preset confidence level;

[0070] Retain detection images with a confidence level greater than the preset confidence level;

[0071] A classification model is used to identify the state of the traffic signs in the retained detection images to obtain the detection results.

[0072] Specifically, the detection model receives the detection image and, using structures such as convolutional neural networks, locates and initially identifies targets in the image, determining whether a target of interest (i.e., a traffic sign) exists. It then provides a predicted bounding box for each target and its corresponding confidence score, indicating the model's certainty that a target exists within that bounding box. The model's output is categorized based on the confidence score. Regions with high confidence (regions of interest) are those where the model considers a real target highly likely; these regions are retained and proceed to the next step. Regions with low confidence, where the model has a lower certainty of the target's presence, are considered false detections or noise and are filtered out without further processing. Finally, a classification model is used to perform state recognition on the traffic signs in the retained detection images to obtain the detection results.

[0073] In some embodiments, the detection model is the YOLOv8n model and the classification model is the YOLOv8n-cls model.

[0074] Among them, the YOLOv8n model is an efficient object detection model; the YOLOv8n-cls model focuses on accurately classifying the specific categories of objects.

[0075] In some embodiments, the detection model includes a Backbone module and a PECA attention module.

[0076] Specifically, the PECA attention module can increase the weight of key target features such as signage areas in the model calculation, making the model pay more attention to the feature information of these areas. This enhances the model's ability to detect targets, enabling it to more accurately identify target objects and reduce false positives and false negatives.

[0077] Based on the above embodiments, Figure 2 This is a system architecture diagram provided for an embodiment of the present invention. (See diagram below.) Figure 2As shown, the system architecture includes an input layer, an auxiliary system, an edge computing layer, and an output layer. In the input layer, vehicle-mounted cameras or roadside monitoring equipment are used to acquire detection images corresponding to traffic signs, and these images undergo preprocessing (including distortion correction and adaptive illumination enhancement). In the edge computing layer, the YOLOv8n detection model receives the preprocessed detection images, outputs high-confidence detection images to the YOLOv8n-cls classification model, and directly filters low-confidence detection images. The YOLOv8n-cls classification model classifies the detection images and outputs the classification to the state quantization module. The state quantization module converts the state information of the traffic signs in the detection images into quantifiable parameters. These quantifiable parameters are then output to the output layer, where the detection results (including the location, category, and state parameters of the traffic signs) are transmitted to the autonomous driving decision-making system and the traffic management platform.

[0078] The auxiliary system includes a model update service and a standard sign database. Traffic scenarios are dynamic, and the model update service continuously monitors these changes, collects new data, and uses this new data to train and optimize the detection and classification models. This enables the models to adapt to the ever-changing traffic environment and maintain high detection and classification accuracy. The standard sign database stores standard information on various traffic signs, including appearance characteristics (shape, color, pattern, etc.), size specifications, and meaning explanations. When the model performs target detection and classification, the standard sign database serves as a reference, helping the model to more accurately identify targets. For example, when the model detects a target that resembles a traffic sign, it compares it with the standard information in the database to determine its specific category and meaning.

[0079] Therefore, this invention achieves a triple breakthrough in traffic sign detection and status recognition technology by setting up a dual-model dynamic cascade architecture: In terms of technical performance, the detection accuracy reaches 89.7% mAP@0.5, the multi-state classification accuracy is 91.2%, and the processing speed is 45 FPS while power consumption is reduced by 42.8%; in terms of economic value, hardware deployment costs are reduced by 83%, road maintenance efficiency is improved by 100%, and annual operation and maintenance costs are reduced by 62%; in terms of social benefits, it can reduce traffic accidents by 32% and support network monitoring of 100,000+ devices. Compared with existing technologies, this solution exceeds industry standards by more than 50% in core indicators such as detection latency (22ms) and robustness to extreme weather (89%). It has passed automotive-grade certification and has been successfully applied in intelligent transportation demonstration zones, demonstrating significant technological advancement and industrialization promotion value.

[0080] Among them, the dual-model dynamic cascade architecture achieves task decoupling through the cascade design of the detection model YOLOv8n and the classification model YOLOv8n-cls, while ensuring 30FPS real-time performance and improving the multi-state recognition accuracy to 89.2%. It innovatively integrates corner regression, feature activation analysis and structural similarity comparison to achieve accurate quantitative output of tilt (±3.2° error) and damage degree, filling the gap of existing technologies that only support binary classification. It adopts the PECA attention mechanism and dynamic ROI filtering to achieve ultra-low power operation of 6.3W on the RK3588 chip, which reduces energy consumption by 42.8% compared with similar solutions.

[0081] Based on the same inventive concept, embodiments of the present invention also provide a detection device for the status of traffic signs. Figure 3 This is a schematic diagram of a traffic sign status detection device provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the first acquisition module 31 is used to acquire the detection image collected by the acquisition device; the second acquisition module 32 is used to input the detection image into the optimized cascaded dual-model architecture to acquire the confidence level of the traffic sign contained in the detection image; the third acquisition module 33 is used to determine the target detection image based on the confidence level and to detect the target detection image to obtain the detection result, which includes the state of the traffic sign, and the state of the traffic sign is at least one of normal, tilted and damaged; wherein, the cascaded dual-model architecture is formed by constructing a detection model and a classification model.

[0082] The traffic sign status detection device provided in the above embodiments can perform the traffic sign status detection methods provided in the above embodiments and has the same or corresponding beneficial effects, which will not be described in detail here.

[0083] This invention also provides a storage medium that stores a program or instructions that cause a computer to execute the steps of the method for detecting the state of a traffic sign provided in the above embodiments.

[0084] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0085] Based on the above embodiments, this invention also provides an electronic device. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 4 As shown, the electronic device includes a processor 401 and a memory 402. The processor 401 executes programs or instructions stored in the memory, such as... Figure 1 The steps of the temperature detection method for the sintering equipment described herein have the beneficial effects of the above embodiments, and will not be repeated here.

[0086] like Figure 4 As shown, an electronic device may include at least one processor 401, at least one memory 402, and at least one communication interface 403. The various components in the electronic device are coupled together via a bus system 404. The communication interface 403 is used for information transmission with external devices. It is understood that the bus system 404 is used to implement communication between these components. In addition to a data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 4 The general designated all buses as Bus System 404.

[0087] It is understood that the memory 402 in this embodiment can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. In some embodiments, the memory 402 stores the following elements: executable units or data structures, or subsets thereof, or extended sets thereof, operating systems, and applications. In this embodiment of the invention, the processor 401 executes the steps of the various embodiments of the method provided in this embodiment of the invention by calling the programs or instructions stored in the memory 402.

[0088] The method provided in this embodiment of the invention can be applied to processor 401, or implemented by processor 401. Processor 401 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the hardware of processor 401 or by instructions in software form. The processor 401 can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor.

[0089] The steps of the method provided in this embodiment of the invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software units in the decoding processor. The software units can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 402, and processor 401 reads the information in memory 402 and combines it with its hardware to complete the steps of the method.

[0090] The electronic device may also include one or more physical components to execute instructions generated by the processor 401 when performing the methods provided in this embodiment of the invention. Different physical components may be located within the electronic device or outside the electronic device, such as in a cloud server. Each physical component, together with the processor 401 and the memory 402, works to realize the functions of the electronic device in this embodiment.

[0091] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0092] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions as those in the present invention.

Claims

1. A method for detecting the status of a traffic sign, characterized in that, include: Acquire the detection images captured by the acquisition device; The detected image is input into the optimized cascaded dual-model architecture to obtain the confidence level that the detected image contains traffic signs; The target detection image is determined based on the confidence level, and the target detection image is detected to obtain the detection result. The detection result includes the status of the traffic sign, and the status of the traffic sign is at least one of normal, tilted, and damaged. The cascaded dual-model architecture is formed by constructing a detection model and a classification model.

2. The method for detecting the status of traffic signs according to claim 1, characterized in that, The optimized cascaded dual-model architecture was obtained through the following method: The detection model and the classification model are designed into a cascaded dual-model architecture through task decoupling, and the cascaded dual-model architecture is imported into the model training platform; Collect a dataset of detection image samples and preprocess the dataset. The preprocessed detection image sample dataset is input into the model training platform; The model training platform outputs the state of the traffic sign in the detected image to train and obtain an optimized cascaded dual-model architecture.

3. The method for detecting the status of traffic signs according to claim 2, characterized in that, A single detected image sample includes multiple traffic signs; the cascaded dual-model architecture employs the CloU-Wise algorithm, which dynamically adjusts the matching threshold based on the density of the multiple traffic signs; A positive sample is one where the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box of the traffic sign is greater than the matching threshold.

4. The method for detecting the state of a traffic sign according to claim 3, characterized in that, The density of traffic signs in a single detected image sample is inversely proportional to the matching threshold.

5. The method for detecting the state of a traffic sign according to claim 1, characterized in that, The process includes determining a target detection image based on the confidence level and performing detection on the target detection image to obtain detection results, including: The confidence level of the detected image containing traffic signs is compared with a preset confidence level; Retain detection images with a confidence level greater than the preset confidence level; The classification model is used to perform state recognition on the traffic signs in the retained detection images to obtain the detection results.

6. The method for detecting the state of a traffic sign according to claim 1, characterized in that, The detection model is the YOLOv8n model, and the classification model is the YOLOv8n-cls model.

7. The method for detecting the state of a traffic sign according to claim 1, characterized in that, The detection model includes a Backbone module and a PECA attention module.

8. A device for detecting the status of a traffic sign, characterized in that, include: The first acquisition module is used to acquire the detection images collected by the acquisition device; The second acquisition module is used to input the detected image into the optimized cascaded dual-model architecture to obtain the confidence that the detected image contains traffic signs; The third acquisition module is used to determine the target detection image based on the confidence level, and to detect the target detection image to obtain the detection result. The detection result includes the status of the traffic sign, and the status of the traffic sign is at least one of normal, tilted, and damaged. The cascaded dual-model architecture is formed by constructing a detection model and a classification model.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the processor executes the steps of the method for detecting the state of a traffic sign as described in any one of claims 1 to 7 by calling a program or instruction stored in the memory.

Images