Recognition reading method for lightweight instrument of automobile

Abstract

The invention discloses an automobile lightweight instrument identification reading method, which comprises the following steps of S1, acquiring an instrument image through a vehicle-mounted camera, and performing deblurring optimization on the image by adopting self-adaptive gamma correction and motion based on a gyroscope; s2, outputting an instrument type identifier and four datum point coordinates based on the lightweight neural network model; s3, calling an identification branch according to the instrument type, and if identification fails, calling historical data interpolation module complement; s4, logic verification is carried out on the recognition result and a vehicle speed signal, a gear signal and a cooling fan state obtained from the CAN bus; and S5, outputting a reading through the vehicle-mounted HMI. According to the method, the gearbox speed ratio matrix is dynamically issued through the cloud, and the problem of rotating speed misjudgment caused by mismatching of speed ratios of illegally modified vehicles is thoroughly solved. Compared with a traditional scheme, the method has the advantages that the stable recognition rate is realized under the high-incidence scene of refitting of trucks, engineering machinery and the like, and the refitted vehicle recognition accuracy is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of automotive instrument technology, specifically to a method for recognizing and reading automotive lightweight instrument panels. Background Technology

[0002] The automotive instrument panel is the core device that centrally displays the vehicle's operating status. It is mainly used to monitor and provide feedback on the working status of various systems and is located in front of the driver. Its core functions include real-time display of basic data such as vehicle speed, engine speed, fuel level, and coolant temperature, and it uses indicator lights and warning lights to indicate faults or abnormal conditions.

[0003] The existing solution relies on fixed speed ratio parameters, resulting in a speed misjudgment rate of >35% for vehicles with modified transmissions (speed ratio matrix mismatch). Traditional image processing suffers a sharp drop in recognition accuracy under strong light / low temperature conditions (midday strong light: 76.3%, -40℃: 68.5%) due to the lack of a fault-tolerant mechanism for infrared supplementary light coupled with CAN verification. A full OTA update (such as CN114265840A) requires downloading a model of ≥300MB, and takes ≥72 hours on a 4G network, which cannot meet the needs of rapid adaptation of fragmented instrument panels in commercial vehicles. Summary of the Invention

[0004] The purpose of this invention is to provide a method for recognizing readings of lightweight automotive instrument panels, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for recognizing readings of lightweight automotive instrument panels, comprising the following steps: S1. Acquire instrument images through vehicle-mounted cameras, and optimize the images using adaptive gamma correction (γ∈[0.8,1.5]) and motion deblurring based on gyroscopes (compensating for angular velocity ≤15° / s); S2. Based on the lightweight neural network model MobileNetV3-Small (input resolution 320×240), output the instrument type identifier and the coordinates of 4 reference points (positioning error ±3 pixels). S3. Call the recognition branch according to the instrument type. If recognition fails, call the historical data interpolation module to supplement the data. Pointer-type instruments: Execute Hough transform + angle regression algorithm (error ≤ 1.5°); Digital instrument: Performs character segmentation + CRNN lightweight OCR engine (word error rate <2%); S4. Perform logical verification between the identification result and the vehicle speed signal (0x0CFE000), gear signal (0x0C190000), and cooling fan status (0x18FF00D0) obtained from the CAN bus. Trigger the update mechanism when the confidence level is <85% for 3 consecutive times. S5: Outputs readings via vehicle HMI (refresh rate 10Hz).

[0006] The lightweight neural network model employs a three-stage knowledge distillation process from ResNet50 to MobileNetV3-Small. The training data includes 350,000 synthetic instrument images. The model compression uses 30% channel pruning + INT8 quantization, with a file size ≤ 2.0MB.

[0007] The instrument area positioning in step S2 specifically includes: Construct a perspective transformation matrix based on the four reference points; Retrieve the pre-stored instrument template based on the VIN code and match areas with a similarity of 95% or higher.

[0008] The logical verification includes: Speed ​​coupling verification: When the speed is ≥4000rpm, the theoretical speed is calculated based on the gearbox ratio matrix sent from the cloud. Where V is the real-time vehicle speed in km / h. This refers to the gear ratio corresponding to the current gear in the gearbox gear ratio matrix sent from the cloud. The radius of the wheel's rolling radius; like Then re-identify; Water temperature continuity verification: Calculate the rate of change ΔT / Δt between adjacent sampling periods. When the engine is started and the cooling fan is off, if |ΔT / Δt|>25℃ / s, it is considered abnormal. Cross-verification of oil circuit faults: Check the existence of oil circuit fault code 0x18F00500.

[0009] The update mechanism includes: Sample upload: When S4 triggers the update condition, the image is uploaded to the cloud via T-Box encryption; Incremental package generation: The cloud-based GAN is used to expand the sample to 5000 images, and after gradient cropping and fine-tuning of the model, the data is compressed to ≤3.2MB; Vehicle-side update: The incremental package is loaded in the MCU security partition through a double buffering mechanism. After the accuracy is verified to be ≥98% by virtual instrument simulation, the switch takes effect.

[0010] A lightweight automotive instrument panel reading system includes: Image acquisition module: OV9281 global shutter camera (with 850nm infrared illumination) to eliminate motion blur caused by high-speed vibration; Edge computing module: Equipped with NXP S32G-VNP chip (2×Cortex-A53@1GHz), real-time lightweight inference; Security update module: integrates the HSM encryption engine and executes the update mechanism.

[0011] Compared with the prior art, the beneficial effects of the present invention are: This invention completely solves the problem of misjudged engine speed caused by mismatched gear ratios in illegally modified vehicles by dynamically distributing the gearbox ratio matrix from the cloud. Compared to traditional solutions, this patent achieves a stable recognition rate in high-modification scenarios such as trucks and construction machinery, significantly improving the accuracy of modified vehicle identification.

[0012] This invention employs an incremental update mechanism, requiring only an update package of ≤3.2MB to adapt to 5000 new instrument types. In a 4G network environment, the time for a complete update is reduced from 72 hours in the traditional solution to less than 10 seconds, improving efficiency and meeting the rapid after-sales adaptation needs of commercial vehicles, achieving a lightweight model update at the second level.

[0013] This invention combines infrared illumination with multimodal CAN verification, maintaining a 95.1% recognition rate at -40℃ low temperature startup and a 98.1% recognition rate under 100,000 lux strong light at noon, eliminating the risk of failure under extreme working conditions.

[0014] This invention introduces a water temperature-fan state coupling criterion: when the water tank fan is off, the maximum allowable temperature change is 25°C / s; when the fan is on, the change threshold is relaxed to 100°C / s. This mechanism reduces the false alarm rate of cold start by 37 percentage points, solves the problem of false fault codes triggered by a sudden rise in water temperature in traditional solutions, and constructs a multimodal safety verification closed loop. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 The present invention provides a technical solution: a method for recognizing readings of lightweight automotive instrument panels, comprising the following steps: S1. Acquire instrument images through vehicle-mounted cameras, and optimize the images using adaptive gamma correction (γ∈[0.8,1.5]) and motion deblurring based on gyroscopes (compensating for angular velocity ≤15° / s); S2. Based on the lightweight neural network model MobileNetV3-Small (input resolution 320×240), output the instrument type identifier and the coordinates of 4 reference points (positioning error ±3 pixels). S3. Call the recognition branch according to the instrument type. If recognition fails, call the historical data interpolation module to supplement the data. Pointer-type instruments: Execute Hough transform + angle regression algorithm (error ≤ 1.5°); Digital instrument: Performs character segmentation + CRNN lightweight OCR engine (word error rate <2%); S4. Perform logical verification between the identification result and the vehicle speed signal (0x0CFE000), gear signal (0x0C190000), and cooling fan status (0x18FF00D0) obtained from the CAN bus. Trigger the update mechanism when the confidence level is <85% for 3 consecutive times. S5: Outputs readings via vehicle HMI (refresh rate 10Hz).

[0018] The lightweight neural network model employs a three-stage knowledge distillation process from ResNet50 to MobileNetV3-Small. The training data includes 350,000 synthetic instrument images. The model compression uses 30% channel pruning + INT8 quantization, with a file size ≤ 2.0MB.

[0019] The instrument area positioning in step S2 specifically includes: Construct a perspective transformation matrix based on the four reference points; Retrieve the pre-stored instrument template based on the VIN code and match areas with a similarity of 95% or higher.

[0020] The logical verification includes: Speed ​​coupling verification: When the speed is ≥4000rpm, the theoretical speed is calculated based on the gearbox ratio matrix sent from the cloud. Where V is the real-time vehicle speed in km / h. This refers to the gear ratio corresponding to the current gear in the gearbox gear ratio matrix sent from the cloud. The radius of the wheel's rolling radius; like Then re-identify; Water temperature continuity verification: Calculate the rate of change ΔT / Δt between adjacent sampling periods. When the engine is started and the cooling fan is off, if |ΔT / Δt|>25℃ / s, it is considered abnormal. Cross-verification of oil circuit faults: Check the existence of oil circuit fault code 0x18F00500.

[0021] The update mechanism includes: Sample upload: When S4 triggers the update condition, the image is uploaded to the cloud via T-Box encryption; Incremental package generation: The cloud-based GAN is used to expand the sample to 5000 images, and after gradient cropping and fine-tuning of the model, the data is compressed to ≤3.2MB; Vehicle-side update: The incremental package is loaded in the MCU security partition through a double buffering mechanism. After the accuracy is verified to be ≥98% by virtual instrument simulation, the switch takes effect.

[0022] A lightweight automotive instrument panel reading system includes: Image acquisition module: OV9281 global shutter camera (with 850nm infrared illumination) to eliminate motion blur caused by high-speed vibration; Edge computing module: Equipped with NXP S32G-VNP chip (2×Cortex-A53@1GHz), real-time lightweight inference; Security update module: integrates the HSM encryption engine and executes the update mechanism.

[0023] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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.

Claims

1. A method for identifying readings of lightweight automotive instrument panels, characterized in that, Includes the following steps: S1. Acquire instrument images through vehicle-mounted cameras, and optimize images using adaptive gamma correction and motion deblurring based on gyroscopes to compensate for angular velocities ≤15° / s; S2. Output the instrument type identifier and the coordinates of four reference points based on a lightweight neural network model; S3. Call the recognition branch according to the instrument type. If recognition fails, call the historical data interpolation module to supplement the data. Pointer-type instruments: perform Hough transform and angle regression algorithms; Digital meter: Performs character segmentation and CRNN lightweight OCR engine; S4. Logically verify the identification results with the vehicle speed signal, gear signal and cooling fan status obtained from the CAN bus. Trigger the update mechanism when the confidence level is <85% for 3 consecutive times. S5: Output readings via the vehicle-mounted HMI.

2. The method for identifying readings of lightweight automotive instrument panels according to claim 1, characterized in that: The lightweight neural network model employs a three-stage knowledge distillation process from ResNet50 to MobileNetV3-Small. The training data includes 350,000 synthetic instrument images. The model compression uses 30% channel pruning + INT8 quantization, with a file size ≤ 2.0MB.

3. The method for identifying readings of lightweight automotive instrument panels according to claim 1, characterized in that: The instrument area positioning in step S2 specifically includes: Construct a perspective transformation matrix based on the four reference points; Retrieve the pre-stored instrument template based on the VIN code and match areas with a similarity of 95% or higher.

4. The method for identifying readings of lightweight automotive instrument panels according to claim 1, characterized in that: The logical verification includes: Speed ​​coupling verification: When the speed is ≥4000rpm, the theoretical speed is calculated based on the gearbox ratio matrix sent from the cloud. Where V is the real-time vehicle speed in km / h. This refers to the gear ratio corresponding to the current gear in the gearbox gear ratio matrix sent from the cloud. The radius of the wheel's rolling radius; like Then re-identify; Water temperature continuity verification: Calculate the rate of change ΔT / Δt between adjacent sampling periods. When the engine is started and the cooling fan is off, if |ΔT / Δt|>25℃ / s, it is considered abnormal. Cross-verification of oil circuit faults: Detecting the presence of oil circuit fault codes.

5. The method for identifying readings of automotive lightweight instrument panels according to claim 1, characterized in that: The update mechanism includes: Sample upload: When S4 triggers the update condition, the image is uploaded to the cloud via T-Box encryption; Incremental package generation: The cloud-based GAN is used to expand the sample to 5000 images, and after gradient cropping and fine-tuning of the model, the data is compressed to ≤3.2MB; Vehicle-side update: The incremental package is loaded in the MCU security partition through a double buffering mechanism. After the accuracy is verified to be ≥98% by virtual instrument simulation, the switch takes effect.

6. A lightweight automotive instrument reading system, characterized in that, include: Image acquisition module: Employs a global shutter camera to eliminate motion blur caused by high-speed vibration; Edge computing module: Real-time lightweight inference; Security update module: integrates the HSM encryption engine and executes the update mechanism.

Images