AEB function optimization method based on cloud deep learning
By applying deep learning technology and cloud computing power in the AEB system, optimize the AEB algorithm and model training, solve the problem of high misjudgment rate of the AEB system and difficult to control the braking distance, and improve the system recognition ability and robustness. Reduced the situation of error triggering and leakage trigger.
Patent Information
- Application Number
- CN202411543921.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-01-28
AI Technical Summary
The AEB system has high misjudgment rate, difficult braking distance to control and technical limitations, especially in severe weather conditions, resulting in an increase in misjudgment and misjudgment.
Through deep learning technology, use cloud computing training target detection and classification models, collect diverse driving data for labeling and model training, optimize algorithms to reduce error triggers and leakage. sex.
Improving the recognition and classification of target objects by the AEB system, and reducing the situation of error triggering and leakage triggers, especially in a complex or changeable driving environment, enhance the robustness and real -time stability of the system.
Smart Images
Figure CN119377745A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of automobile intelligent control, and specifically relates to an AEB function optimization method based on cloud-based deep learning. Background Art
[0002] AEB, or Autonomous Emergency Braking, is an active safety technology for cars. Its origins can be traced back to the 1950s. As cars entered more families, traffic accidents also increased. In order to reduce the occurrence of accidents, many innovative products closely related to our lives were born during this period, such as seat belts, bumpers, and AEB that can prevent collisions. However, due to the high manufacturing costs at the time and the fact that it was not a mainstream product, AEB technology ultimately failed to be mass-produced. The AEB system uses advanced sensors such as cameras, radars, or lidars to continuously monitor the road. When AEB detects potential danger ahead, the system assesses the risk of collision by analyzing factors such as distance and speed. If the driver fails to respond effectively to the first warning, AEB will automatically activate the vehicle's braking system to completely avoid the collision or reduce the degree of the collision, thereby minimizing accidents and losses. However, the AEB system still faces some technical difficulties and development directions. For example, the perception ability of the AEB system directly affects its judgment and execution capabilities. Therefore, how to make the AEB system more accurately perceive the dangerous situation ahead and make correct decisions and braking actions is one of the current research focuses. In addition, with the development and popularization of autonomous driving technology, the AEB system also needs to be combined with other autonomous driving technologies to improve overall driving safety.
[0003] The main problems of AEB system are as follows: (1) High misjudgment rate. The radar of AEB system may not meet high standards in detection distance and accuracy, especially under normal weather conditions. In addition, there are some problems with information collection cameras, such as short-focus cameras limit the test distance, and long-focus cameras blur short-distance objects, which may cause AEB system to misjudge. In severe weather conditions such as rain, snow, fog and haze, the ranging radar may basically lose its effectiveness, and the camera may also fail if it is blocked by dirt, further increasing the possibility of misjudgment. (2) Braking distance is difficult to control: AEB system may react slowly to static objects appearing in front of the car during high-speed driving, which makes braking distance difficult to control. In an emergency, if the AEB system is activated slightly slower, a collision may occur. (3) Technical limitations: The technical implementation methods of AEB system, such as radar and camera-based perception systems, have their limitations. For example, radar may have a short test distance, and the camera may not be able to effectively identify certain static objects or scenes. Summary of the invention
[0004] In response to the above problems, the present invention proposes a method of using deep learning technology to train a model with a large amount of labeled data to improve the AEB system's ability to identify and classify target objects. The AEB algorithm is continuously iterated and optimized to reduce false triggering and missed triggering, especially in complex or changing driving environments. The specific technical solution is as follows: A method for optimizing AEB functions based on cloud-based deep learning, comprising the following steps: Step (1) Collect a large amount of real driving data and annotate the data in detail; Step (2) Model selection and training: Select a deep learning model suitable for target detection and classification tasks, use labeled data to train the model, and monitor the model training process. Ensure that the model has good performance on the validation set.
[0005] Step (3) Model evaluation and optimization: Use an independent test set to evaluate the model, analyze the false triggering and missed triggering of the model, identify the problems and perform targeted optimization; Step (4) Algorithm iteration and optimization: Based on the results of model evaluation and actual problems, the AEB algorithm is continuously iterated and optimized; the target detection algorithm is optimized to improve the recognition and classification capabilities of different target objects; Step (5) Model update and maintenance: optimize the model’s reasoning speed, introduce a safety monitoring mechanism to ensure the model’s stability and safety in real-time operation, and update and maintain the model to cope with new road conditions and vehicle types.
[0006] Furthermore, the driving data in step (1) includes various driving environments, weather conditions, road types, and vehicle types; the data annotation includes the location, size, category, and speed information of the target object. Ensure the diversity and balance of the data to avoid overfitting or bias of the model in specific scenarios.
[0007] Furthermore, the deep learning model in step (2) is YOLO, SSD or Faster R-CNN; the training strategy adopts data enhancement, learning rate adjustment or batch size optimization.
[0008] Furthermore, in step (3), the evaluation indicators include accuracy, recall, F1 score and mean average precision (mAP); the optimization method includes trying different network structures, loss functions and optimization algorithms.
[0009] Furthermore, in step (4), continuous optimization includes introducing context information, multi-sensor fusion, time series analysis or motion prediction. Introducing technologies such as context information and multi-sensor fusion can improve the robustness of the AEB system in complex environments. Introducing technologies such as time series analysis and motion prediction can predict potential dangerous situations in advance and reduce false triggering and missed triggering.
[0010] The present invention transmits the data of sensors such as radar to the cloud in real time on the AEB controller, uses the cloud's large computing power deep learning model to provide real-time feedback results, and then transmits them to the vehicle controller to adjust the vehicle speed to avoid collisions or reduce the severity of collisions. Deep learning technology is used to improve the AEB (automatic emergency braking) system's ability to identify and classify target objects, and reduce false triggering and missed triggering, especially in complex or changing driving environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Figure 1 Schematic diagram of the AEB system structure. DETAILED DESCRIPTION
[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0013] A method for optimizing AEB functions based on cloud-based deep learning includes the following steps: Step (1) Collect a large amount of real driving data and annotate the data in detail; Step (2) Model selection and training: Select a deep learning model suitable for target detection and classification tasks, use labeled data to train the model, and monitor the model training process. Ensure that the model has good performance on the validation set.
[0014] Step (3) Model evaluation and optimization: Use an independent test set to evaluate the model, analyze the false triggering and missed triggering of the model, identify the problems and perform targeted optimization; Step (4) Algorithm iteration and optimization: Based on the results of model evaluation and actual problems, the AEB algorithm is continuously iterated and optimized; the target detection algorithm is optimized to improve the recognition and classification capabilities of different target objects; Step (5) Model update and maintenance: optimize the model’s reasoning speed, introduce a safety monitoring mechanism to ensure the model’s stability and safety in real-time operation, and update and maintain the model to cope with new road conditions and vehicle types.
[0015] The driving data in step (1) includes various driving environments, weather conditions, road types, and vehicle types; the data annotation includes the location, size, category, and speed information of the target object. Ensure the diversity and balance of the data to avoid overfitting or bias of the model in specific scenarios.
[0016] The deep learning model in step (2) is YOLO, SSD or Faster R-CNN; the training strategy adopts data enhancement, learning rate adjustment or batch size optimization.
[0017] In the step (3), the evaluation indicators include accuracy, recall, F1 score and mean average precision (mAP); the optimization method includes trying different network structures, loss functions and optimization algorithms.
[0018] In step (4), continuous optimization includes introducing context information, multi-sensor fusion, time series analysis or motion prediction. Introducing technologies such as context information and multi-sensor fusion can improve the robustness of the AEB system in complex environments. Introducing technologies such as time series analysis and motion prediction can predict potential dangerous situations in advance and reduce false triggering and missed triggering.
[0019] The above describes in detail the preferred implementation of this patent, but this patent is not limited to the above implementation. Various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of this patent.
Claims
1. A method for optimizing AEB functions based on cloud-based deep learning, characterized in that: The steps include: Step (1) Collect a large amount of real driving data and annotate the data in detail; Step (2) Model selection and training: Select a deep learning model suitable for target detection and classification tasks, use labeled data to train the model, and monitor the model training process. Ensure that the model has good performance on the validation set; Step (3) Model evaluation and optimization: Use an independent test set to evaluate the model, analyze the false triggering and missed triggering of the model, identify the problems and perform targeted optimization; Step (4) Algorithm iteration and optimization: Based on the results of model evaluation and actual problems, the AEB algorithm is continuously iterated and optimized; the target detection algorithm is optimized to improve the recognition and classification capabilities of different target objects; Step (5) Model update and maintenance: optimize the model’s reasoning speed, introduce a safety monitoring mechanism to ensure the model’s stability and safety in real-time operation, and update and maintain the model to cope with new road conditions and vehicle types.
2. The AEB function optimization method based on cloud deep learning according to claim 1, characterized in that: The driving data in step (1) includes various driving environments, weather conditions, road types, and vehicle types; the data annotation includes the location, size, category, and speed information of the target object. Ensure the diversity and balance of the data to avoid overfitting or bias of the model in specific scenarios.
3. The AEB function optimization method based on cloud deep learning according to claim 1, characterized in that: The deep learning model in step (2) is YOLO, SSD or Faster R-CNN; the training strategy adopts data enhancement, learning rate adjustment or batch size optimization.
4. The AEB function optimization method based on cloud deep learning according to claim 1, characterized in that: In the step (3), the evaluation indicators include accuracy, recall, F1 score and mean average precision (mAP); the optimization method includes trying different network structures, loss functions and optimization algorithms.
5. The AEB function optimization method based on cloud deep learning according to claim 1, characterized in that: In step (4), the continuous optimization includes introducing context information, multi-sensor fusion, time series analysis or motion prediction.