License plate real-time detection method and system based on dynamic temperature adjustment feature fusion
Through lightweight network and multimodal synchronization technology, the real-time and accuracy of license plate detection are solved, and efficient license plate detection is achieved on low-power hardware platforms, which is suitable for a variety of hardware platforms and complex scenarios.
Patent Information
- Application Number
- CN202510394172.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-31
- Publication Date
- 2025-07-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing license plate detection technology has high inference delay, insufficient multimodal data coordination, high computing power requirements, and poor adaptability to occlusion scenes on the on-board hardware platform, which cannot meet the requirements of real-time and high precision.
Using lightweight network architecture, dynamic temperature regulation feature fusion and multimodal synchronization technology, the grouping channel shuffling convolution and improved Kalman filtering algorithm are combined with dynamic temperature regulation mechanism and quantized error compensation to achieve feature fusion and data synchronization to adapt to complex scenarios.
Significantly improve computing efficiency and detection accuracy, reduce latency and power consumption, meet the real-time requirements of low-power hardware platforms, and is suitable for a variety of hardware platforms and complex scenarios.
Smart Images
Figure CN120279538A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a real-time license plate detection method and system based on dynamic temperature regulation feature fusion. Background Art
[0002] The existing license plate detection technologies have the following main problems in practical applications:
[0003] Insufficient real-time performance: The existing technologies (such as the solution based on YOLOv3) have a high inference latency (such as 28 milliseconds) on in-vehicle hardware platforms, and cannot meet the requirement of an end-to-end latency less than 10 milliseconds in in-vehicle scenarios.
[0004] Insufficient multi-modal collaboration: The existing technologies do not fully support the synchronization and fusion of multi-modal data (such as cameras, infrared, millimeter-wave radars), resulting in low detection accuracy (such as the mAP is only 83.2% in the rainstorm scenario) in complex scenarios (such as rainstorms, occlusions).
[0005] High computing power requirement: The existing technologies have a large number of model parameters (such as 2.1M), high computational complexity, and are difficult to run efficiently on low-power hardware.
[0006] Poor adaptability to occlusion scenarios: The existing technologies have a high miss detection rate (≥18%) in scenarios where the occlusion area exceeds 30%, and cannot meet the actual application requirements.
[0007] In view of the above problems, the present invention proposes a real-time license plate detection method and system based on dynamic temperature regulation feature fusion, which solves the deficiencies of the existing technologies through a lightweight network architecture, dynamic feature fusion, and multi-modal synchronization technologies. Summary of the Invention
[0008] Technical Problems to be Solved
[0009] In view of the deficiencies of the existing technologies, the present invention provides a real-time license plate detection method and system based on dynamic temperature regulation feature fusion.
[0010] Technical Solutions
[0011] To achieve the above object, the present invention provides the following technical solutions:
[0012] A real-time license plate detection method and system based on dynamic temperature regulation feature fusion, comprising the following steps:
[0013] (a) Extract features from the input image using grouped channel shuffle convolution (the number of groups is 4, and the shuffle interval is 2 layers) to reduce the computational complexity and improve the feature expression ability;
[0014] (b) Dynamically fuse features at different levels through a dynamic temperature adjustment mechanism, where the temperature parameter τ is dynamically adjusted between 0.4 and 0.6 to optimize the feature fusion weights, thereby improving the detection accuracy;
[0015] (c) Use an improved Kalman filter algorithm (process noise covariance Q = 0.01, measurement noise covariance R = 0.1) to perform time alignment on multi-modal data, ensuring that the synchronization error is within 50 milliseconds to meet the real-time requirement.
[0016] As a further aspect of the present invention, the backbone network of the grouped channel shuffle convolution is based on an improved MobileNetV3 architecture. Through grouped convolution and channel shuffle operations, while maintaining the lightweight of the model, it enhances the cross-channel information exchange of features, thereby improving the detection performance.
[0017] As a further aspect of the present invention, the dynamic temperature adjustment mechanism automatically adjusts the temperature parameter τ by monitoring the changes in environmental conditions and feature responses, enabling the feature fusion weights to adaptively reflect the importance of different features, thereby maintaining high detection accuracy in complex scenarios (such as heavy rain, occlusion, etc.).
[0018] As a further aspect of the present invention, the Kalman filter algorithm is used to perform time alignment on multi-modal data from different sensors (such as cameras, infrared, millimeter-wave radars), ensuring the consistency of the data in the time dimension, thereby improving the effect of multi-modal data fusion.
[0019] As a further aspect of the present invention, the method further includes a quantization error compensation step. By calibrating the weights during the quantization process, it reduces the accuracy loss caused by quantization, thereby still maintaining high detection performance under low-bit quantization (such as INT8 or 4bit).
[0020] As a further aspect of the present invention, the method supports multiple hardware platforms, including but not limited to in-vehicle chips (such as Horizon Journey 5) and embedded AI chips (such as Huawei Ascend 310), and realizes efficient end-to-end inference through hardware acceleration technology to meet the requirements of different application scenarios.
[0021] As a further aspect of the present invention, the application of the method in an in-vehicle advanced driver assistance system (ADAS) includes real-time license plate detection, recognition, and tracking, and can run at a speed of 60 frames per second at a resolution of 1920×1080, while maintaining the power consumption below 5 watts (in an 85°C environment).
[0022] As a further solution of the present invention, the application of this method in the drone inspection system transmits the detection results through the 5G ultra-reliable and low-latency communication (URLLC) mode, ensuring that the end-to-end delay does not exceed 10 milliseconds, and is applicable to license plate detection in high-speed mobile scenarios.
[0023] Beneficial effects
[0024] Compared with the prior art, the present invention provides a real-time license plate detection method and system based on dynamic temperature regulation feature fusion, having the following beneficial effects:
[0025] Significantly improve the computing efficiency: Through the lightweight hybrid convolutional network and quantization technology, the computing efficiency (TOPS / W) is increased by 92%, meeting the real-time requirements of low-power hardware platforms.
[0026] High-precision detection: In complex scenarios such as heavy rain, the detection accuracy (mAP) reaches 96.5%, which is 6.5 percentage points higher than the industry standard.
[0027] Low latency and high energy efficiency: The end-to-end delay is reduced to 8.2 milliseconds, which is 59% lower than the prior art, and the power consumption is controlled within 5 watts (in an 85°C environment), suitable for vehicle-mounted and embedded scenarios.
[0028] Multi-modal data synchronization: Through the improved Kalman filtering algorithm, the multi-modal data synchronization error is ≤50 milliseconds, meeting the requirements of the ISO 14229-1 standard.
[0029] Wide hardware adaptability: Supports a variety of hardware platforms (such as Horizon Journey 5 chip, Huawei Ascend 310), and realizes high-speed data transmission through the 5G URLLC communication mode. Brief description of the drawings
[0030] Figure 1 It is the system architecture diagram of a real-time license plate detection method and system based on dynamic temperature regulation feature fusion proposed by the present invention;
[0031] Figure 2 It is the temperature coefficient optimization curve diagram of a real-time license plate detection method and system based on dynamic temperature regulation feature fusion proposed by the present invention;
[0032] Figure 3 It is the energy efficiency ratio radar diagram of a real-time license plate detection method and system based on dynamic temperature regulation feature fusion proposed by the present invention;
[0033] Figure 4 It is the dynamic weight distribution surface of a real-time license plate detection method and system based on dynamic temperature regulation feature fusion proposed by the present invention. Detailed implementation manners
[0034] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the following further details the present invention through embodiments and in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0035] The serial numbers assigned to components in this document, such as "first", "second", etc., are only used to distinguish the described objects and do not have any sequential or technical meanings. The "connection" and "coupling" mentioned in the present invention, unless otherwise specified, both include direct and indirect connection (coupling). In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation to the present invention.
[0036] In the present invention, unless otherwise clearly specified and limited, the first feature being "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, the first feature being "above", "over" and "on top of" the second feature may be that the first feature is directly above or obliquely above the second feature, or simply means that the first feature has a higher horizontal height than the second feature. The first feature being "under", "beneath" and "underneath" the second feature may be that the first feature is directly below or obliquely below the second feature, or simply means that the first feature has a lower horizontal height than the second feature.
[0037] Refer to Figures 1 - 4 , a real-time license plate detection method and system based on the fusion of dynamic temperature regulation features, comprising the following steps:
[0038] (a) Extract features from the input image using grouped channel shuffle convolution (with a group number of 4 and a shuffle interval of 2 layers) to reduce the computational complexity and improve the feature expression ability;
[0039] (b) Dynamically fuse features at different levels through a dynamic temperature regulation mechanism, where the temperature parameter τ is dynamically adjusted between 0.4 and 0.6 to optimize the feature fusion weight, thereby improving the detection accuracy;
[0040] (c) Use an improved Kalman filter algorithm (process noise covariance Q = 0.01, measurement noise covariance R = 0.1) to perform time alignment on multi-modal data to ensure that the synchronization error is within 50 milliseconds, meeting the real-time requirement.
[0041] Furthermore, by means of grouped channel shuffle convolution and a dynamic temperature adjustment mechanism, the computational complexity is significantly reduced while the feature expression ability is enhanced, ensuring high detection accuracy in complex scenarios.
[0042] The improved Kalman filter algorithm realizes the time alignment of multi-modal data, ensures data consistency, and meets the real-time requirements.
[0043] Specifically, the backbone network of the grouped channel shuffle convolution is based on the improved MobileNetV3 architecture. Through grouped convolution and channel shuffle operations, while maintaining the lightweight of the model, the cross-channel information exchange of features is enhanced, thereby improving the detection performance.
[0044] Furthermore, based on the improved MobileNetV3 architecture, through grouped convolution and channel shuffle operations, while maintaining the lightweight of the model, the cross-channel information exchange of features is enhanced, and the detection performance is improved.
[0045] Specifically, the dynamic temperature adjustment mechanism monitors the changes in environmental conditions and feature responses, and automatically adjusts the temperature parameter τ, enabling the feature fusion weights to adaptively reflect the importance of different features, thereby maintaining high detection accuracy in complex scenarios (such as heavy rain, occlusion, etc.).
[0046] Furthermore, the temperature parameter τ is dynamically adjusted, enabling the feature fusion weights to adaptively reflect the importance of different features, thereby maintaining high detection accuracy in complex scenarios (such as heavy rain, occlusion).
[0047] Specifically, the Kalman filter algorithm is used to perform time alignment on multi-modal data from different sensors (such as cameras, infrared, millimeter-wave radars), ensuring the consistency of data in the time dimension, thereby improving the effect of multi-modal data fusion.
[0048] Furthermore, time alignment is performed on multi-modal data to ensure the consistency of data in the time dimension, improve the effect of multi-modal data fusion, and meet the ISO 14229-1 standard.
[0049] Specifically, this method further includes a quantization error compensation step. By calibrating the weights during the quantization process, the accuracy loss caused by quantization is reduced, so that high detection performance can still be maintained under low-bit quantization (such as INT8 or 4bit).
[0050] Furthermore, the weights are calibrated during the quantization process to reduce the accuracy loss caused by quantization, so that high detection performance can still be maintained under low-bit quantization (such as INT8 or 4bit).
[0051] Specifically, this method supports multiple hardware platforms, including but not limited to in-vehicle chips (such as Horizon Journey 5) and embedded AI chips (such as Huawei Ascend 310), and realizes efficient end-to-end inference through hardware acceleration technology to meet the requirements of different application scenarios.
[0052] Furthermore, efficient end-to-end inference is achieved through hardware acceleration technology to meet the different application scenario requirements of in-vehicle chips (such as Horizon Journey 5) and embedded AI chips (such as Huawei Ascend 310).
[0053] Specifically, the application of this method in in-vehicle advanced driver assistance systems (ADAS) includes real-time license plate detection, recognition, and tracking, which can run at a speed of 60 frames per second at a resolution of 1920×1080 while maintaining a power consumption of less than 5 watts (in an 85°C environment).
[0054] Furthermore, in in-vehicle advanced driver assistance systems, real-time license plate detection, recognition, and tracking are achieved, which can run at a speed of 60 frames per second at a resolution of 1920×1080 while maintaining low power consumption (≤5W).
[0055] Specifically, the application of this method in drone inspection systems transmits detection results through the 5G ultra-reliable low-latency communication (URLLC) mode, ensuring that the end-to-end delay does not exceed 10 milliseconds, and is applicable to license plate detection in high-speed mobile scenarios.
[0056] Furthermore, the detection results are transmitted through the 5G ultra-reliable low-latency communication (URLLC) mode, ensuring that the end-to-end delay does not exceed 10 milliseconds, and is applicable to license plate detection in high-speed mobile scenarios.
[0057] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.
[0058] The above-described embodiments merely represent several implementation manners of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the appended claims.
Claims
1. A real-time license plate detection method and system based on dynamic temperature adjustment feature fusion, comprising the following steps: (a) Extract features from the input image using grouped channel shuffle convolution (with 4 groups and a shuffle interval of 2 layers) to reduce computational complexity and improve feature expression ability; (b) Dynamically fuse features at different levels through a dynamic temperature adjustment mechanism, where the temperature parameter τ is dynamically adjusted between 0.4 and 0.6 to optimize the feature fusion weight and thus improve detection accuracy; (c) Use an improved Kalman filter algorithm (process noise covariance Q = 0.01, measurement noise covariance R = 0.1) to perform time alignment on multi-modal data, ensuring that the synchronization error is within 50 milliseconds to meet the real-time requirement.
2. The real-time license plate detection method and system based on dynamic temperature regulation feature fusion according to claim 1, characterized in that The backbone network of the grouped channel shuffle convolution is based on the improved MobileNetV3 architecture. Through grouped convolution and channel shuffle operations, it enhances cross-channel information exchange of features while keeping the model lightweight, thereby improving detection performance.
3. A real-time license plate detection method and system based on dynamic temperature adjustment feature fusion according to claim 1, characterized in that, The dynamic temperature adjustment mechanism automatically adjusts the temperature parameter τ by monitoring changes in environmental conditions and feature responses, enabling the feature fusion weight to adaptively reflect the importance of different features, thus maintaining high detection accuracy in complex scenarios (such as heavy rain, occlusion, etc.).
4. A real-time license plate detection method and system based on dynamic temperature adjustment feature fusion according to claim 1, characterized in that, The Kalman filter algorithm is used to perform time alignment on multi-modal data from different sensors (such as cameras, infrared, millimeter-wave radars), ensuring the consistency of data in the time dimension, thereby improving the effect of multi-modal data fusion.
5. A real-time license plate detection method and system based on dynamic temperature adjustment feature fusion according to claim 1, characterized in that This method also includes a quantization error compensation step. By calibrating the weights during the quantization process, it reduces the accuracy loss caused by quantization, so that high detection performance can still be maintained under low-bit quantization (such as INT8 or 4bit).
6. The real-time license plate detection method and system based on dynamic temperature adjustment feature fusion according to claim 1, characterized in that This method supports multiple hardware platforms, including but not limited to in-vehicle chips (such as Horizon Journey 5) and embedded AI chips (such as Huawei Ascend 310), and realizes efficient end-to-end inference through hardware acceleration technology to meet the requirements of different application scenarios.
7. A real-time license plate detection method and system based on dynamic temperature adjustment feature fusion according to claim 1, characterized in that, The application of this method in an in-vehicle advanced driver assistance system (ADAS) includes real-time license plate detection, recognition, and tracking, and can run at a speed of 60 frames per second at a resolution of 1920×1080, while keeping the power consumption below 5 watts (in an 85°C environment).
8. A real-time license plate detection method and system based on dynamic temperature adjustment feature fusion according to claim 1, characterized in that, The application of this method in an unmanned aerial vehicle inspection system transmits the detection results through the 5G ultra-reliable low-latency communication (URLLC) mode, ensuring that the end-to-end delay does not exceed 10 milliseconds, and is suitable for license plate detection in high-speed mobile scenarios.