Accelerator mistaken stepping prevention method and system based on driving behavior detection

Abstract

The invention discloses a method and a system for preventing stepping on an accelerator by mistake based on driving behavior detection. The face and upper body images of the driver are obtained through the vehicle-mounted camera, and meanwhile the images of the road and surrounding vehicles, pedestrians and obstacles are obtained according to the installed front camera and the installed rear camera. On the basis, a laser radar sensor is combined, and the distance between front and back vehicles, pedestrians and obstacles is accurately measured. A pressure sensor is installed on an accelerator pedal, and the treading force is monitored in real time. And analyzing the driver image through a deep learning model, and identifying whether the driver is fatigue, distraction or has other abnormal states. And if an abnormal state is detected, giving an alarm to prompt a driver. Front and back vehicles, pedestrians and obstacles are recognized through a target detection algorithm, and the relative distance and speed are calculated. Pressure data of an accelerator pedal are monitored in real time, and whether current operation is reasonable or not is judged by combining the current vehicle speed and road information. If the system detects abnormity, early warning is given out in time, and measures are taken.

Description

Technical Field

[0001] The present application relates to the field of road traffic safety technology, and in particular to a method and system for preventing accidental accelerator pedaling based on driving behavior detection. Background Art

[0002] With the rapid increase in the popularity of cars, road traffic accidents have become a major problem threatening social security. Among them, traffic accidents caused by drivers mistakenly taking the accelerator for the brake and stepping on the accelerator to reverse due to factors such as driver tension, fatigue, lack of concentration, poor night vision, and unfamiliarity with vehicle operation accounted for as high as 12.6%, greatly increasing the risk of road traffic accidents. Especially in congested or dangerous roads, if the driver mistakenly steps on the accelerator, it may cause a rear-end collision; when waiting for a red light or pedestrians at an intersection, mistakenly stepping on the accelerator may cause running a red light or colliding with pedestrians; and when reversing in a complex environment such as a small parking lot, a steep slope, and a large number of vehicles and pedestrians, mistakenly stepping on the accelerator may cause the vehicle to hit objects or people behind.

[0003] Although there are many driving assistance systems on the market, most of them focus on how to improve the comfort and convenience of the driver under normal driving conditions, rather than directly preventing and handling misoperation. The existing technology mainly relies on vehicle sensor data to detect the accidental accelerator pedaling, but fails to fully consider the driver's behavior state, resulting in frequent misjudgments and missed judgments. In addition, the existing protection system also has certain deficiencies in real-time performance and response speed, and cannot timely and effectively avoid the dangers caused by the accidental accelerator pedaling at critical moments. With the development of artificial intelligence and intelligent sensor technology, the accidental accelerator pedaling protection system based on driving behavior detection can more comprehensively analyze the driver's state and his operation of the vehicle, and provide more accurate and reliable accidental accelerator pedaling protection. This new type of protection system can not only quickly identify the behavior of accidental accelerator pedaling, but also predict the possible risk of misoperation in advance, so as to take corresponding preventive measures. Therefore, the development of a method and system for preventing accidental accelerator pedaling based on driving behavior detection has important practical significance and application prospects. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a method and system for preventing accidental accelerator pedal pressing based on driving behavior detection.

[0005] The technical solution provided by the present invention is a method for preventing accidental accelerator pedal pressing based on driving behavior detection, which specifically includes the following steps:

[0006] Step 1: Obtain driver image information. Obtain driver image information through the in-car camera, build a driver image database for comparing and identifying different drivers, and use the driver image to judge the driver's status, such as fatigue, distraction, etc., so as to take corresponding safety measures in time.

[0007] Step 2: Obtain road information. Use the on-board camera and sensors to obtain road environment information, determine the distance between the vehicle and the vehicles in front and behind and obstacles, and build a real-time road environment database.

[0008] Step 3: Obtain the accelerator pedal pressure. The pressure sensor installed on the accelerator pedal monitors the pressure data applied by the driver in real time.

[0009] Step 4: Fusion of multiple information to determine the accidental accelerator pedaling and issue personalized warnings. Comprehensive analysis of driver status information, road environment information, and accelerator pedal pressure data is performed to determine whether the accelerator pedal has been accidentally pedaled using a preset algorithm. Comprehensive analysis and judgment of the driver's status is performed to issue personalized warnings to the driver.

[0010] Furthermore, the implementation method of constructing the driver image database in step 1 is as follows:

[0011] First, a fixed camera is installed in the car and the angle is adjusted to ensure that the driver's face and facial features are clearly captured. The camera captures the driver's image data in real time and transmits it to the processing unit through a wireless connection. The processing unit obtains the driver's image data in video format. First, the video is extracted into a picture at K frames / second, and the driver's image is preprocessed. The preprocessing includes: image resizing, center cropping, and rotation, so that the image input size is uniform and the driver's main body is prominent.

[0012] Furthermore, the image information in the driver image database is input into a preset driver behavior detection module to obtain the driver status information.

[0013] Furthermore, the specific processing process of the preset driver behavior detection model is as follows: the preprocessed driving behavior image is input, and features are extracted and the size of the feature map is reduced through the initial convolution layer and pooling layer; then, deeper features are extracted through the four stages of Resnet, each stage contains multiple residual blocks, and a squeeze and excitation attention mechanism module, i.e., SEblock module, is added to each stage, and convolution operations, squeeze operations, and excitation operations are performed on the feature map to obtain weight coefficients, and the learned weight coefficients are used to scale each channel of the input feature map, and the importance of each channel in the map is adaptively adjusted to obtain an enhanced feature map; finally, the enhanced feature map is visualized using the GradCAM method to realize the important areas that the model pays attention to in the driving behavior recognition task, and the classification results are output through the fully connected layer output softmax layer.

[0014] Furthermore, the squeeze and excitation attention mechanism module obtains the overall information of each channel through a global average pooling operation. In this operation, a global feature description vector is obtained by averaging each channel, which captures the importance of each channel in the entire feature map, as shown in the following formula:

[0015]

[0016] Where: Z c represents the global average value of the cth channel, u c is the cth channel in the input feature map, W is the width of the feature map, H is the height of the feature map, i, j are the feature maps u c The spatial dimension index of ;

[0017] Then, the dependencies between channels are captured, specifically:

[0018] s=σ(W2δ(W1Z))

[0019] Where: s represents the activation of each channel, which is used to weight the original feature map, δ represents the Relu activation function, σ represents the Sigmoid function, W1 is the weight of the dimension reduction fully connected layer, and W2 is the weight of the dimension increase fully connected layer;

[0020] Finally, s is multiplied by the convolutional features of the previous layer spatially. is the scaled channel feature, specifically:

[0021]

[0022] Among them, F scale Represents a scaling operation, s c Represents the weight coefficient after recalibration of the feature map of the c-th channel.

[0023] Furthermore, Resnet adopts the Resnet50 model.

[0024] Furthermore, the specific implementation of step 2 is as follows:

[0025] Install infrared cameras and laser radars with night vision functions on the front and rear bumpers and sides of the vehicle to measure the distance to obstacles in front, behind and on the sides of the vehicle;

[0026] Connect the camera and lidar sensor to the vehicle computing platform through the interface, ensure that the power supply and data transmission lines are connected correctly, and test to confirm that all hardware devices are working properly;

[0027] The camera collects real-time image data from the front and rear, capturing roads, vehicles, pedestrians, traffic lights and obstacles;

[0028] The LiDAR sensor measures the distance and speed of objects in front and behind, and transmits the data to the onboard computing platform;

[0029] Preprocess the acquired data information, including denoising and filtering operations, to eliminate environmental noise and irrelevant echoes;

[0030] Use the target detection algorithm YOLOV5 to detect and mark vehicles, pedestrians and obstacles in the image;

[0031] The distance data provided by the LiDAR sensor is matched with the target positions in the image to accurately calculate the distance and speed of each target;

[0032] The data from the camera and sensor are fused using Kalman filtering. Based on the fused perception data, the real-time road environment database is updated in real time, including the location, motion trajectory and speed information of the currently identified objects and personnel.

[0033] Furthermore, the specific implementation method in step 4 is as follows;

[0034] When the vehicle is started, the system detects that the driver is distracted and starts voice reminders. At the same time, it collects information about the accelerator pedal pressure and the current vehicle speed. If the current vehicle is within a safe distance ahead without other obstacles and the speed remains stable compared to the previous moment, only voice reminders are given to the driver to make him concentrate on driving. If the distance between the current vehicle and the vehicle ahead exceeds a certain threshold, the signal is directly transmitted to the ECU, which controls the engine and reduces the throttle output to reduce the vehicle speed.

[0035] When the driver is detected to be driving normally, the system detects the distance between the current vehicle and surrounding obstacles. If the vehicle is within a safe distance and is accelerating, the system transmits a signal to the ECU to decelerate the vehicle.

[0036] When the vehicle is at an intersection and the traffic light ahead is red, if it detects that the accelerator pedal of the vehicle has applied 1 / 3 of the set threshold, it is determined that the current driver has stepped on the accelerator pedal by mistake. The system transmits the mistaken stepping judgment information to the ECU. After receiving the signal, the ECU immediately activates the braking system and performs emergency braking to prevent the vehicle from accidentally running a red light;

[0037] When the vehicle is in the reversing state, if it is detected that the accelerator pedal of the current vehicle has been pressed with 1 / 3 of the set threshold, it is determined that the current driver has stepped on the accelerator pedal by mistake. The system transmits the mistaken stepping judgment information to the ECU. After receiving the signal, the ECU immediately activates the braking system and performs emergency braking to prevent accidents caused by accidentally stepping on the accelerator pedal when the vehicle is reversing.

[0038] The present invention also provides an anti-accelerator accidental pressing system based on driving behavior detection, comprising a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute an anti-accelerator accidental pressing method based on driving behavior detection as described in the above technical solution.

[0039] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0040] The present invention obtains the driver's image information and road image, first detects the on-board camera to obtain the driver's face and upper body image, uses computer vision technology to detect the driver's state (such as fatigue, distraction, etc.), and simultaneously obtains the road and surrounding vehicles, pedestrians and obstacles according to the installation of front and rear cameras. On this basis, combined with the laser radar sensor, the distance between the front and rear vehicles, pedestrians and obstacles is accurately measured. A pressure sensor is installed on the accelerator pedal to monitor the pedaling force in real time. The driver's image is analyzed by a deep learning model to identify whether the driver is fatigued, distracted or in other abnormal states. If an abnormal state is detected, an alarm is issued to prompt the driver. The surrounding environment model is constructed by combining the front and rear camera images and sensor data. The target detection algorithm is used to identify the front and rear vehicles, pedestrians and obstacles, and calculate the relative distance and speed. The pressure data of the accelerator pedal is monitored in real time, and the current operation is judged to be reasonable in combination with the current speed of the vehicle and the road information. If the system detects an abnormality (such as excessive pressure on the accelerator pedal within a safe distance), an early warning is issued in time and measures are taken (such as automatic deceleration or braking). The method of the present invention can comprehensively monitor the situation around the driver and the vehicle, make timely and accurate judgments and interventions on potential dangers, thereby avoiding accidents to the greatest extent possible, effectively improving driving safety, reducing traffic accidents caused by misoperation, and ensuring driver safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 It is a schematic diagram of the process of an embodiment of the present invention;

[0043] Figure 2 A structural diagram of the driving behavior detection model provided by the present invention;

[0044] Figure 3 Schematic diagram of the specific position of inserting the SE module into the residual network in an embodiment of the present invention;

[0045] Figure 4 The result diagram of some embodiments of the present invention for detecting the driver's state, wherein (a) is the original driving behavior image; (b) is the heat map of the driver behavior detection model including the channel attention mechanism;

[0046] Figure 5 The following are examples of driving behaviors according to an embodiment of the present invention: (a) texting with the right hand; (b) making a phone call with the right hand; (c) texting with the left hand; (d) making a phone call with the left hand;

[0047] Figure 6 A schematic diagram of a process for providing personalized warning of driving behavior provided by the present invention.

[0048] Figure 7 This is a graph of the model training loss value and test set accuracy proposed in the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0050] like Figure 1 As shown, the present invention provides a method for preventing accidental accelerator pedal pressing based on driving behavior detection, which specifically includes the following steps:

[0051] Step 1, obtain driver image information and driving status information. Obtain driver image information through the in-vehicle camera, build a driver image database for comparing and identifying different drivers, and obtain driver status information through the driver behavior detection module;

[0052] First, a fixed camera is installed in the car, and the angle is adjusted to ensure that the driver's face and facial features are clearly captured. The camera captures the driver's image data in real time and transmits it to the processing unit through a wireless connection. The processing unit is a computer system, and the processing unit obtains the driver's image data in a video format. First, the video is extracted as a picture at 24 frames per second, and the driver's image is preprocessed. The preprocessing includes: image resizing, center cropping, rotation, etc., so that the image input size is uniform and the driver's main body is prominent.

[0053] Furthermore, the processed image information is input into a preset driver behavior detection model to obtain a detection result. The preset driver behavior detection model combines a channel attention mechanism and a residual network, such as Figure 2 As shown. It is responsible for extracting the image feature data of the current driving behavior and transmitting the feature data to the feature visualization module to judge the driver's state. This model can focus on the key features of driving behavior and has the ability to pay close attention to details and accurately locate. The preset driver behavior detection module is obtained in the following way:

[0054] First, obtain a public data set of driver driving behavior, use image enhancement technology to expand the data set, and randomly divide the data set into a training set and a test set in a ratio of 7:3. Then, input the training set into the network model for training, and obtain the trained network weights. Finally, use the test set to test the network model, and input the test results into the visualization module to judge the driver's state. In this embodiment, the use of a mobile phone by a driver during driving is taken as an example to display the predicted visualization results. Similarly, this model is also applicable to other behaviors.

[0055] The specific processing process of the preset driver behavior detection model is as follows: the pre-processed driving behavior image is input, and features are extracted and the size of the feature map is reduced through the initial convolution and pooling stages; then, deeper features are extracted through the four stages of Resnet, each Stage contains multiple residual blocks, and the SEblock module is added to the Stage, and convolution operations, squeeze operations and excitation operations are performed on the feature map, and the learned weights are used to scale each channel of the input feature map, and the importance of each channel in the map is adaptively adjusted to enhance the feature representation capability; finally, the GradCAM method is used to visualize the feature map of the penultimate layer of the convolution layer to realize the important areas that the model pays attention to in the driving behavior recognition task, and the classification results are output through the fully connected layer output softmax layer. In this embodiment, there are a total of 5 categories, c0 represents normal driving, c1 represents texting with the right hand, c2 represents making a phone call with the right hand, c3 represents texting with the left hand, and c4 represents making a phone call with the left hand. Specifically, Resnet uses Resnet50. The deep residual network ResNet50 is a network with a depth of 50 layers, including convolutional layers, batch normalization layers, activation layers and other convolutional layers. Specifically: the initial layer is a normal convolution structure, which is responsible for preliminary feature extraction of the input image. Layer 1 contains 3 residual blocks (ResidualBlock). Each residual block consists of multiple convolutional layers and jump connections to capture the low-level patterns of the input features. Layer 2 contains 4 residual blocks, which gradually extract higher-level features in a layer-by-layer deepening manner. Layer 3 contains 6 residual blocks to further deepen the network to capture more abstract and complex features. Layer 4 contains 3 residual blocks to further expand the network depth so that the network can learn more complex image features. The Squeeze-and-Excitation (SE) Attention Mechanism module is added to the residual block structure of the residual network. The specific structure diagram is as follows Figure 3 As shown, each residual block can utilize the channel attention mechanism to better capture and utilize the important information in the input features and reduce the attention to irrelevant channels.

[0056] The squeeze and excitation attention mechanism module obtains the overall information of each channel through a global average pooling operation. In this operation, a global feature description vector is obtained by average pooling each channel. This vector captures the importance of each channel in the entire feature map. As shown in the following formula, Z c represents the global average value of the cth channel, specifically:

[0057]

[0058] Where: u c is the cth channel in the input feature map, W is the width of the feature map, H is the height of the feature map, i, j are the feature maps u c The spatial dimension index of .

[0059] Then, the dependencies between channels are captured, specifically:

[0060] s=σ(W2δ(W1Z))

[0061] Where: s represents the activation of each channel, which is used to weight the original feature map, δ represents the Relu activation function, σ represents the Sigmoid function, W1 is the weight of the dimension reduction fully connected layer, and W2 is the weight of the dimension increase fully connected layer.

[0062] Finally, s is multiplied by the convolutional features of the previous layer spatially. is the scaled channel feature, specifically:

[0063]

[0064] Among them, F scale Represents a scaling operation, s c Represents the weight coefficient after recalibration of the feature map of the cth channel;

[0065] Step 2: Obtain road information. Use the on-board camera and sensors to obtain road environment information, determine the distance between the vehicle and the front and rear vehicles and obstacles, and use the collected data information to build a real-time road environment database to provide key information such as the distance and location of the front and rear vehicles and obstacles.

[0066] In this embodiment, infrared cameras and laser radars with night vision functions are installed on the front and rear bumpers and sides of the vehicle to measure the distance of obstacles in front, behind and on the sides of the vehicle;

[0067] Connect the camera and lidar sensor to the vehicle computing platform through the interface, ensure that the power supply and data transmission lines are connected correctly, and test to confirm that all hardware devices are working properly;

[0068] The camera collects real-time image data from the front and rear, capturing roads, vehicles, pedestrians, traffic lights and obstacles;

[0069] The lidar sensor measures the distance, speed and other information of objects in front and behind, and transmits the data to the on-board computing platform;

[0070] Preprocess the acquired data information, including denoising, filtering and other operations, to eliminate environmental noise and irrelevant echoes;

[0071] Use the object detection algorithm (YOLO V5) to detect and mark vehicles, pedestrians, and obstacles in the image;

[0072] The distance data provided by the LiDAR sensor is matched with the target positions in the image to accurately calculate the distance and speed of each target;

[0073] The data from the camera and sensor are fused using Kalman filtering to improve the accuracy of information such as the position, speed, and movement trajectory of objects and personnel. Based on the fused perception data, the real-time road environment database is updated in real time, including the position, movement trajectory, and speed information of the currently identified objects and personnel.

[0074] Step 3: Obtain the accelerator pedal pressure. The pressure sensor installed on the accelerator pedal monitors the pressure data applied by the driver in real time.

[0075] In this embodiment, a strain gauge sensor is installed under the accelerator pedal to ensure that the force applied by the driver can be accurately sensed, and the accelerator pedal pressure data is collected in real time through the sensor and transmitted to the vehicle's control unit (ECU).

[0076] Step 4: Multi-faceted information fusion to determine the situation of mis-stepping on the accelerator. Comprehensively analyze the driver's status information, road environment information and accelerator pedal pressure data; use the driver's image to determine the driver's status, such as fatigue, distraction, etc., so that corresponding safety measures can be taken in time, and use the preset algorithm to determine whether there is a mis-stepping on the accelerator. Once the system detects that the driver has mis-stepped on the accelerator, it will immediately take measures to brake the vehicle.

[0077] Furthermore, the driver status information, road information and accelerator pedal data are obtained to make a comprehensive judgment on multiple aspects of information, such as Figure 3 As shown;

[0078] When the vehicle is started, the system automatically starts working to detect the driver's distracted state. If the system detects that the driver is distracted, it will remind the driver to concentrate through voice. At the same time, the system will collect accelerator pedal pressure information and current vehicle speed.

[0079] Normal driving: If there are no obstacles in front of the vehicle and the speed is stable, the system will only remind the driver to stay focused through voice, and no further intervention is required.

[0080] There is a vehicle within a safe distance: If the vehicle in front is too close, the system will transmit a signal to the ECU, and the ECU will control the engine to reduce the throttle output, thereby reducing the vehicle speed and avoiding the risk of collision.

[0081] Acceleration control in normal driving scenarios

[0082] When the system detects that the driver is in normal driving state, it will continuously monitor the distance between the vehicle and surrounding obstacles.

[0083] Accelerating and no obstacles: If the vehicle is accelerating and the distance to surrounding obstacles is within a safe range, the system will continue to monitor the driving environment and take no further action.

[0084] Accelerating and there is an obstacle: If the vehicle is close to an obstacle and is still accelerating, the system will transmit a signal to the ECU, which will ensure safety by reducing throttle output or actively intervening in deceleration.

[0085] Intervention for accidental accelerator pedal pressing at red light at intersection

[0086] When the vehicle approaches an intersection and the traffic light ahead is red, the system monitors accelerator pedal pressure data.

[0087] Mis-operation of the accelerator: If the accelerator pedal pressure exceeds the set pressure threshold, the system will determine it as mis-operation and trigger automatic braking measures to prevent the vehicle from running a red light.

[0088] Mild accidental stepping intervention: If the accelerator pedal pressure is detected to reach 1 / 3 of the set threshold, the system will regard it as the driver's accidental stepping on the accelerator and immediately transmit the accidental stepping judgment information to the ECU. After receiving the signal, the ECU activates the braking system and performs emergency braking to prevent the vehicle from accidentally running a red light.

[0089] Handling the Accelerator by Misoperation in Reverse State

[0090] The system also monitors accelerator pedal pressure when the vehicle is in reverse.

[0091] Mistaken accelerator pedal pressure: If the system detects that the accelerator pedal pressure reaches 1 / 3 of the set threshold, it will determine that the driver has mistakenly stepped on the accelerator. The system will then transmit this information to the ECU. After receiving the signal, the ECU will immediately activate the braking system to prevent accidents caused by mistaken accelerator pedal pressure during reversing.

[0092] Table 1 shows the experimental results of the comparison between the model proposed in the present invention and other models.

[0093] Table 1

[0094]

[0095] Table 2

[0096]

[0097] Through the comparison of different model experiments, this model has a higher accuracy rate by testing the driver behavior data of the test set. In the recognition of the second, third, and fourth types of driving behavior categories, the accuracy rate is significantly improved, and the average accuracy rate of the model is 99.89%. Figure 7 As shown, it can be explained that the method of the present invention can accurately identify the driver's behavior.

[0098] Another invention, an embodiment of the present invention also provides an anti-accelerator accidental pressing system based on driving behavior detection, including a processor and a memory, the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute an anti-accelerator accidental pressing method based on driving behavior detection as described in the above technical solution.

[0099] The specific implementations described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for preventing accidental accelerator pedal pressing based on driving behavior detection, characterized in that: The steps include: Step 1, obtaining driver image information through an in-vehicle camera, building a driver image database for comparing and identifying different drivers, and using the driver image to determine the driver status information; Step 2: Use the vehicle-mounted camera and sensors to obtain road environment information, determine the distance between the vehicle and the front and rear vehicles and obstacles, and build a real-time road environment database; Step 3, monitoring the pressure data applied by the driver in real time through a pressure sensor installed on the accelerator pedal; Step 4: Comprehensively analyze the driver status information, road environment information and accelerator pedal pressure data, determine whether the accelerator is accidentally pressed through a preset algorithm, conduct a comprehensive analysis and judgment of the driver's status, and provide personalized warnings to the driver.

2. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 1, characterized in that: The implementation method of building the driver image database in step 1 is as follows: First, a fixed camera is installed in the car and the angle is adjusted to ensure that the driver's face and facial features are clearly captured. The camera captures the driver's image data in real time and transmits it to the processing unit through a wireless connection. The processing unit obtains the driver's image data in video format. First, the video is extracted into a picture at K frames / second, and the driver's image is preprocessed. The preprocessing includes: image resizing, center cropping, and rotation, so that the image input size is uniform and the driver's main body is prominent.

3. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 1, characterized in that: The image information in the driver image database is input into a preset driver behavior detection module to obtain the driver status information.

4. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 3, characterized in that: The specific processing process of the preset driver behavior detection model is as follows: the preprocessed driving behavior image is input, and features are extracted and the size of the feature map is reduced through the initial convolution layer and pooling layer; then, deeper features are extracted through the four stages of Resnet, each stage contains multiple residual blocks, and a squeeze and excitation attention mechanism module, i.e., SEblock module, is added in each stage to perform convolution operations, squeeze operations, and excitation operations on the feature map to obtain weight coefficients, and the learned weight coefficients are used to scale each channel of the input feature map, and the importance of each channel in the map is adaptively adjusted to obtain an enhanced feature map; finally, the enhanced feature map is visualized using the GradCAM method to realize the important areas that the model pays attention to in the driving behavior recognition task, and the classification results are output through the fully connected layer output softmax layer.

5. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 4, characterized in that: The squeeze and excitation attention mechanism module obtains the overall information of each channel through a global average pooling operation. In this operation, a global feature description vector is obtained by averaging each channel. This vector captures the importance of each channel in the entire feature map, as shown in the following formula: Where: Z c represents the global average value of the cth channel, u c is the cth channel in the input feature map, W is the width of the feature map, H is the height of the feature map, i, j are the feature maps u c The spatial dimension index of ; Then, the dependencies between channels are captured, specifically: s=σ(W2δ(W1Z)) Where: s represents the activation of each channel, which is used to weight the original feature map, δ represents the Relu activation function, σ represents the Sigmoid function, W1 is the weight of the dimension reduction fully connected layer, and W2 is the weight of the dimension increase fully connected layer; Finally, s is multiplied by the convolutional features of the previous layer spatially. is the scaled channel feature, specifically: Among them, F scale Represents a scaling operation, s c Represents the weight coefficient after recalibration of the feature map of the c-th channel.

6. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 4, characterized in that: Resnet uses the Resnet50 model.

7. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 1, characterized in that: The specific implementation of step 2 is as follows: Install infrared cameras and laser radars with night vision functions on the front and rear bumpers and sides of the vehicle to measure the distance to obstacles in front, behind and on the sides of the vehicle; Connect the camera and lidar sensor to the vehicle computing platform through the interface, ensure that the power supply and data transmission lines are connected correctly, and test to confirm that all hardware devices are working properly; The camera collects real-time image data from the front and rear, capturing roads, vehicles, pedestrians, traffic lights and obstacles; The LiDAR sensor measures the distance and speed of objects in front and behind, and transmits the data to the onboard computing platform; Preprocess the acquired data information, including denoising and filtering operations, to eliminate environmental noise and irrelevant echoes; Use the target detection algorithm YOLOV5 to detect and mark vehicles, pedestrians and obstacles in the image; The distance data provided by the LiDAR sensor is matched with the target positions in the image to accurately calculate the distance and speed of each target; The data from the camera and sensor are fused using Kalman filtering. Based on the fused perception data, the real-time road environment database is updated in real time, including the location, motion trajectory and speed information of the currently identified objects and personnel.

8. The method for preventing accidental accelerator pedal pressing based on driving behavior detection as claimed in claim 1, characterized in that: The specific implementation method in step 4 is as follows; When the vehicle is started, the system detects that the driver is distracted and starts voice reminders. At the same time, it collects information about the accelerator pedal pressure and the current vehicle speed. If the current vehicle is within a safe distance ahead without other obstacles and the speed remains stable compared to the previous moment, only voice reminders are given to the driver to make him concentrate on driving. If the distance between the current vehicle and the vehicle ahead exceeds a certain threshold, the signal is directly transmitted to the ECU, which controls the engine and reduces the throttle output to reduce the vehicle speed. When the driver is detected to be driving normally, the system detects the distance between the current vehicle and surrounding obstacles. If the vehicle is within a safe distance and is accelerating, the system transmits a signal to the ECU to decelerate the vehicle. When the vehicle is at an intersection and the traffic light ahead is red, if it detects that the accelerator pedal of the vehicle has applied 1 / 3 of the set threshold, it is determined that the current driver has stepped on the accelerator pedal by mistake. The system transmits the mistaken stepping judgment information to the ECU. After receiving the signal, the ECU immediately activates the braking system and performs emergency braking to prevent the vehicle from accidentally running a red light; When the vehicle is in the reversing state, if it is detected that the accelerator pedal of the current vehicle has been pressed with 1 / 3 of the set threshold, it is determined that the current driver has stepped on the accelerator pedal by mistake. The system transmits the mistaken stepping judgment information to the ECU. After receiving the signal, the ECU immediately activates the braking system and performs emergency braking to prevent accidents caused by accidentally stepping on the accelerator pedal when the vehicle is reversing.

9. A system for preventing accidental accelerator pedal pressing based on driving behavior detection, characterized in that: It includes a processor and a memory, the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute a method for preventing accidental accelerator pedaling based on driving behavior detection as described in any one of claims 1 to 8.

Images