Hidden driver emergency call method in vehicle

Abstract

The invention discloses an in-vehicle hidden driver emergency call method, and relates to the technical field of vehicle-mounted safety, and the method comprises the steps: installing a pressure sensor group, and generating a trigger signal when a combined trigger mode of a hand pressure value and a pedal pressure value is detected; analyzing the daily driving behavior data of the driver, and comparing the trigger signal with a pre-stored normal driving behavior mode; when the trigger signal deviates from the normal driving behavior mode and meets a preset emergency trigger threshold value, acquiring passenger information in the vehicle and vehicle driving data, and packaging the passenger information and the vehicle driving data into an emergency information packet; and transmitting the emergency information packet to a background rescue system, and sending an emergency call instruction with the in-vehicle condition after determining whether there is a dangerous person. According to the invention, the agility and safety of emergency response can be improved, the overall emergency response and decision-making capability can be enhanced, and the response timeliness and safety prevention and control level of the in-vehicle emergency call system can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle safety, and particularly to a method for a driver's emergency call hidden in a vehicle. Background Art

[0002] In recent years, with the rapid development of vehicle informatization and intelligent technologies, in-vehicle safety monitoring and emergency rescue systems have gradually shifted from traditional mechanical alarms and manual distress calls to automation and intelligence. In the field of in-vehicle safety monitoring of the prior art, emergency calls mostly rely on means such as in-vehicle cameras, voice recognition, or manual operations. Some solutions introduce multiple sensor information such as biometric recognition and pressure sensing in order to achieve real-time monitoring of the states of drivers and passengers.

[0003] However, most of the existing in-vehicle safety systems focus on triggering alarms after monitoring abnormal states, and have deficiencies such as incomplete data collection, a single triggering mechanism, and obvious response delays. At the same time, deep learning and big data analysis have made certain progress in the field of driving behavior determination, but mostly focus on aspects such as assisted driving and collision prevention, and pay insufficient attention to the hidden distress needs of drivers in abnormal situations. Although multi-sensor fusion technology has been applied in some advanced driver assistance systems (ADAS), it has not yet formed a closed-loop, efficient, and dynamically adaptive solution for in-vehicle hidden emergency calls. Due to the limitations of information collection and processing, it is difficult to meet the need for comprehensive judgment of abnormal driving behaviors and in-vehicle passenger states in emergency situations, thereby affecting the timeliness and accuracy of rescue responses. Summary of the Invention

[0004] In view of the problems existing in the existing method for a driver's emergency call hidden in a vehicle, the present invention is proposed. Therefore, the problem to be solved by the present invention is how to provide a method for a driver's emergency call hidden in a vehicle.

[0005] To solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a method for a driver's emergency call hidden in a vehicle, which includes installing a pressure sensor group at the bottom of the vehicle's steering wheel and the clutch pedal, and generating a trigger signal when a combination trigger mode of a hand pressure value and a pedal pressure value is detected;

[0007] Analyzing the daily driving behavior data of the driver based on a deep learning model, and comparing the trigger signal with a pre-stored normal driving behavior pattern;

[0008] When the trigger signal deviates from the normal driving behavior pattern and meets a preset emergency trigger threshold, obtaining in-vehicle passenger information and vehicle driving data, and packaging the passenger information and the vehicle driving data into an emergency information packet;

[0009] Transmit the emergency information packet to the background rescue system, and send an emergency call instruction with the vehicle interior condition after determining whether there are dangerous persons.

[0010] As a preferred solution of the in-vehicle concealed driver emergency call method described in the present invention, wherein: the pressure sensor group includes a first pressure sensor and a second pressure sensor, wherein the first pressure sensor is arranged at the bottom of the steering wheel for collecting the hand pressure value, and the second pressure sensor is arranged on the clutch pedal for collecting the pedal pressure value.

[0011] As a preferred solution of the in-vehicle concealed driver emergency call method described in the present invention, wherein: the combined trigger mode includes that if it is detected that the hand pressure value and the pedal pressure value increase from the initial value to be greater than the first set threshold within the first set time, then record and form a first trigger signal;

[0012] If it is detected that the hand pressure value and the pedal pressure value change up and down within the second set threshold within the second set time, then record and form a second trigger signal;

[0013] If it is detected that the hand pressure value and the pedal pressure value decrease to the pressure initial value within the third set time, then record and form a third trigger signal.

[0014] As a preferred solution of the in-vehicle concealed driver emergency call method described in the present invention, wherein: the analysis of the driver's daily driving behavior data based on the deep learning model includes,

[0015] Obtain the driver's daily driving behavior data, including normal driving and abnormal driving, label the data according to the driving standard to form a training set, a validation set and a test set. The driver's daily driving behavior data includes vehicle acceleration, steering angle, brake pedal pressure, throttle depth, vehicle speed;

[0016] Perform data normalization processing on the driver's daily driving behavior data, and the formula is:

[0017]

[0018] In the formula, x is the original data, x min and x max are the minimum value and the maximum value of the data respectively;

[0019] Use the training set data to train the neural network weight parameters and bias parameters of the bidirectional long short-term memory network model, and minimize the loss function through the Adam optimizer;

[0020] Evaluate the model performance on the validation set, adjust the hyperparameters, and verify the generalization ability of the model on the test set;

[0021] By calculating the probability distribution of normal driving behaviors in the training data, a value outside the 95% confidence interval of the probability distribution is set as the emergency trigger threshold.

[0022] As a preferred solution of the in-vehicle concealed driver emergency call method of the present invention, wherein: the obtaining of in-vehicle passenger information and vehicle driving data includes,

[0023] Immediately start the in-vehicle on-board camera to capture the in-vehicle images covering all seat areas inside the vehicle, and collect the in-vehicle images and real-time driving data;

[0024] Locate the face regions in the in-vehicle images to generate face candidate boxes, and screen the face candidate boxes to obtain the face position information;

[0025] According to the facial key point coordinates in the face position information, count the number of in-vehicle passengers, and at the same time extract the face features to form a facial feature vector;

[0026] Integrate and organize the obtained number of in-vehicle passengers, facial feature information, and driving data into an emergency information packet according to the preset JSON format.

[0027] As a preferred solution of the in-vehicle concealed driver emergency call method of the present invention, wherein: the generating of the face candidate boxes includes,

[0028] Grayscale and binarize the collected in-vehicle images. The grayscale conversion formula is:

[0029] Z = 0.299*R + 0.587*G + 0.114*B

[0030] In the formula, Z represents the grayscale image point, R is the red in the image, G is the green in the image, and B is the blue in the image;

[0031] Use the threshold method to convert the grayscale image into a black-and-white image, set the pixels greater than the threshold to black, and the pixels less than the threshold to white;

[0032] Use the CNN convolutional neural network to extract the local features and global features in the binarized image, and determine the position and size of the face according to the response intensity of the feature map to generate face candidate boxes;

[0033] The screening of the face candidate boxes includes,

[0034] Collect all the detected face candidate boxes, and sort the face candidate boxes in descending order according to the confidence score,

[0035] Create an empty list to save the finally retained candidate boxes. From the sorted candidate box list, take out each candidate box in turn, and calculate the overlap degree between the candidate box and the remaining candidate boxes. The formula is:

[0036]

[0037] Among them, IoU is the intersection over union of the candidate bounding boxes, and A and B represent the areas of two candidate bounding boxes respectively;

[0038] Set the removal threshold IoU i , for each remaining candidate bounding box, if the intersection over union value with the current highest-confidence bounding box ≥ the removal threshold IoU i , then remove the candidate bounding box from the list;

[0039] Repeat the steps until all candidate bounding boxes are processed. Each time, select the candidate bounding box with the highest confidence in the current list as the new reference bounding box until the candidate bounding box list is empty. The finally retained set of candidate bounding boxes is the filtered face candidate bounding boxes.

[0040] As a preferred solution of the method for emergency call of a concealed driver in a vehicle according to the present invention, wherein: the sending of an emergency call instruction with the vehicle condition after determining whether there is a dangerous person includes,

[0041] After the background rescue system receives the emergency information packet, analyze the face features of the passengers;

[0042] If the analysis result shows that there is a dangerous person in the vehicle, the background system immediately calculates the optimal rescue route according to the vehicle position information, the background system sends an emergency call instruction with the vehicle condition, adjusts the first set threshold and the set time of the hand pressure value and the pedal pressure value, and recalibrates the weight distribution of the normal driving behavior pattern in the deep learning model;

[0043] If the analysis result shows that there is no dangerous person in the vehicle, the background system suspends sending the emergency call instruction, the background system records and files this emergency information, feedbacks the statistical comparison result of the probability distribution of the current driving behavior data and the normal driving behavior data, calibrates and adjusts the initial values of the hand pressure value and the pedal pressure in the pressure sensor group, and updates the normal driving behavior data at the same time;

[0044] If the background rescue system cannot determine whether there is a dangerous person in the vehicle and there is a situation where the face features cannot be accurately judged, send an instruction to the vehicle-mounted camera to re-collect the vehicle interior image, and at the same time adjust the removal threshold for screening the face candidate bounding boxes, increase the acquisition of vehicle interior environment data for secondary face feature analysis, and the adjustment formula is:

[0045]

[0046] In the formula, IoU n is the adjusted removal threshold, IoU i is the initial removal threshold, IoU ais the average intersection over union between candidate boxes, M is the number of overlapping pairs of candidate boxes, and β is an adjustment factor.

[0047] The beneficial effects of the present invention are that it can improve the agility and safety of emergency response, significantly improve the recognition accuracy of the system for abnormal driving behaviors, provide a scientific basis and data support for emergency rescue decision-making, enhance the overall emergency response and decision-making capabilities of the system, and improve the response timeliness and safety prevention and control level of the in-vehicle emergency call system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0049] Figure 1 It is a flowchart of an in-vehicle concealed driver emergency call method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] In order to make the above objects, features, and advantages of the present invention more comprehensible, the following will describe the specific embodiments of the present invention in detail with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0052] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or selectively exclusive embodiments from other embodiments.

[0053] Embodiment 1

[0054] Refer to Figure 1 , which is the first embodiment of the present invention. This embodiment provides an in-vehicle concealed driver emergency call method, including:

[0055] S1: Install a pressure sensor group at the bottom of the steering wheel and the clutch pedal inside the vehicle. When the combination of the detected hand pressure value and the pedal pressure value triggers a mode, a trigger signal is generated.

[0056] Specifically, install a pressure sensor group at the bottom of the steering wheel and the clutch pedal inside the vehicle. The pressure sensor group includes a first pressure sensor and a second pressure sensor. The first pressure sensor is set at the bottom of the steering wheel to collect the hand pressure value, and the second pressure sensor is set on the clutch pedal to collect the pedal pressure value. When the combination of the detected hand pressure value and the pedal pressure value triggers a mode, a trigger signal is generated.

[0057] If it is detected that the hand pressure value and the pedal pressure value increase from the initial value to be greater than the first set threshold within the first set time, then record and form a first trigger signal.

[0058] If it is detected that the hand pressure value and the pedal pressure value change up and down within the second set time, then record and form a second trigger signal.

[0059] If it is detected that the hand pressure value and the pedal pressure value decrease to 0 within the third set time, then record and form a third trigger signal.

[0060] Emergency braking trigger mode. If the pressure on the bottom of the steering wheel by the hand suddenly increases sharply (possibly because the driver encounters an emergency and subconsciously grips the steering wheel tightly), and at the same time the clutch pedal is quickly depressed to the bottom (indicating that the driver wants to perform an emergency braking operation), the vehicle's braking assist system can respond quickly based on this, enhance the braking force, and help the vehicle decelerate or stop as soon as possible.

[0061] Fatigue driving trigger mode. When the pressure on the bottom of the steering wheel by the hand shows an unstable state (such as being sometimes light and sometimes heavy, changing frequently), and the operation of the clutch pedal also shows irregular conditions (such as the frequency and strength of depressing and releasing having no obvious pattern), after a period of time, a trigger signal is sent out to remind the driver that they may be in a state of fatigue driving, and the vehicle warns the driver to rest through means such as sound, light, or vibration.

[0062] Vehicle out-of-control trigger mode. The pressure on the bottom of the steering wheel by the hand suddenly increases (the driver tries to control the vehicle direction), and at the same time the clutch pedal is abnormally depressed and released frequently (possibly because the vehicle has a fault, resulting in the driver being unable to normally control the vehicle speed and power), indicating that the vehicle may be out of control, such as a steering system failure or a power system abnormality.

[0063] Vehicle collision trigger mode: the pressure on the bottom of the steering wheel by the hand disappears instantly (possibly because the driver is impacted by the collision and both hands leave the steering wheel), and at the same time, the pressure on the clutch pedal also disappears instantly (the driver can no longer control the pedal), indicating that a vehicle collision may have occurred or is about to occur.

[0064] In summary, by installing a pressure sensor group at the bottom of the in-vehicle steering wheel and the clutch pedal, this method captures the driver's key operation data, realizes real-time monitoring of the combined mode of hand pressure and pedal pressure; can sensitively capture abnormal signals by monitoring key operation areas when the driver is in an emergency state; uses multi-dimensional acquisition of sensor data as the basis for subsequent data processing and judgment. It achieves timely capture of abnormal driving behaviors in a concealed state, constructs a primary warning mechanism for the entire system, and improves the agility and safety of emergency response.

[0065] S2: Analyze the driver's daily driving behavior data based on a deep learning model, and compare the trigger signal with the pre-stored normal driving behavior pattern;

[0066] Specifically, the trigger signal is received by the vehicle control module. The vehicle control module analyzes the driver's daily driving behavior data based on the deep learning model, and compares the trigger signal with the pre-stored normal driving behavior pattern;

[0067] Obtain the driver's daily driving behavior data, including normal driving and abnormal driving, label the data according to driving standards to form a training set, a validation set, and a test set. The driver's daily driving behavior data includes vehicle acceleration, steering angle, brake pedal pressure, throttle depth, and vehicle speed;

[0068] Perform data normalization processing on the driver's daily driving behavior data. The formula is:

[0069]

[0070] In the formula, x is the original data, x min and x max are the minimum and maximum values of the data respectively;

[0071] Use the training set data to train the neural network weight parameters and bias parameters of the bidirectional long short-term memory network model, and minimize the loss function through the Adam optimizer;

[0072] Evaluate the model performance on the validation set, adjust the hyperparameters, and verify the generalization ability of the model on the test set;

[0073] By calculating the probability distribution of normal driving behaviors in the training data, set the outside of the 95% confidence interval of the probability distribution as the emergency trigger threshold.

[0074] In summary, this method analyzes the daily driving behavior data of drivers based on a deep learning model, compares the trigger signal with the pre-stored normal driving mode, optimizes the accuracy of anomaly recognition, improves the recognition accuracy of the system for abnormal driving behaviors, and provides a scientific basis and data support for emergency rescue decisions.

[0075] S3: When the trigger signal deviates from the normal driving behavior mode and meets the preset emergency trigger threshold, obtain the information of the passengers in the vehicle and the driving data of the vehicle, and package the passenger information and the vehicle driving data into an emergency information packet;

[0076] Specifically, the in-vehicle control module controls the in-vehicle camera to start the face detection function, obtains the number of passengers in the vehicle and the facial feature information, simultaneously collects the driving data such as the real-time position, driving direction, and vehicle speed of the vehicle, and packages the facial feature information and the driving data into an emergency information packet. The in-vehicle control module automatically adjusts the in-vehicle audio to the mute state and turns off the display screen;

[0077] When the vehicle system detects a specific combination trigger mode of hand pressure and clutch pedal pressure, generate an emergency event code.

[0078] After generating the emergency event code, the in-vehicle camera is immediately started to capture the in-vehicle images covering all seat areas in the vehicle to collect image information.

[0079] Apply a face detection algorithm to the obtained images to quickly locate the face regions in the images and generate face candidate boxes. Screen and optimize the obtained face candidate boxes, remove the misdetected non-face regions, obtain the accurate face position information, count the number of passengers in the vehicle according to the facial key point coordinates, and simultaneously extract the texture features, contour features, etc. of each face to form a facial feature vector.

[0080] Grayscale and binarize the collected pictures, and the conversion formula is as follows:

[0081] Z = 0.299*R + 0.587*G + 0.114*B

[0082] In the formula, Z represents the grayscale image point, R is the red in the image, G is the green in the image, and B is the blue in the image.

[0083] Convert the grayscale image into a black-and-white image using the threshold method to further simplify the image. Calculate the grayscale histogram, cumulative probability and cumulative mean, total mean and between-class variance of the image. Assume there is a grayscale image with gray levels in the range of [0, L], and the number of pixels for each gray level i is n i and the total number of pixels in the image is N.

[0084] Calculate the grayscale histogram by counting the number of pixels for each gray level in the image.

[0085] Calculate the cumulative probability and cumulative mean: First, it is necessary to calculate the probability p of each gray level i , which is defined as the number of pixels n of this gray level i divided by the total number of pixels N, and the formula is as follows:

[0086] p i =n i / N

[0087] Then calculate the cumulative probability P(k) and cumulative mean m(k). For gray levels from 0 to k, the calculation methods of the cumulative probability and cumulative mean are as follows:

[0088]

[0089] In the formula, P k is the cumulative probability, and m k is the cumulative mean. The total mean m G is the average gray level of all pixels of all gray levels, and is calculated by the following formula:

[0090]

[0091] For each gray level k, the between-class variance δ 2 B(k) can be calculated by the following formula:

[0092]

[0093] Select the gray level corresponding to the maximum between-class variance as the threshold. The threshold T to be found is the k value that makes δ 2 B(k) maximum:

[0094] T = argmax{k∈[0,L]}δ 2 B(k)

[0095] Select the gray level corresponding to the maximum between-class variance as the threshold; set the pixels greater than the threshold to black and the pixels less than the threshold to white.

[0096] During the search process, extract the feature map of the image, use the multi-layer convolution and pooling operations of the convolutional neural network to capture the local features and global features in the image, and determine the possible face positions and sizes according to the response intensity of the feature map to generate a series of face candidate boxes.

[0097] Since the face candidate boxes may contain misdetected non-face regions, use the non-maximum suppression NMS algorithm to remove redundant candidate boxes. According to the confidence scores and overlap degrees of the candidate boxes, retain the candidate box with the highest confidence and remove the candidate boxes that overlap with it and have lower confidence.

[0098] Collect all the detected candidate bounding boxes, each box containing location information and a confidence score;

[0099] Sort the candidate bounding boxes according to the confidence scores, with the candidate bounding box having the highest confidence score ranked at the front;

[0100] Create an empty list to save the finally retained candidate bounding boxes. From the sorted list of candidate bounding boxes, take out each candidate bounding box in turn, and take out the currently highest-confidence candidate bounding box (i.e., the first box in the list);

[0101] Calculate the degree of overlap between this candidate bounding box and the remaining candidate bounding boxes, using the intersection over union (IoU) as the measurement criterion. The formula is:

[0102]

[0103] where IoU is the intersection over union of the candidate bounding boxes, and A and B represent the areas of the two candidate bounding boxes respectively.

[0104] Set the removal threshold IoU i . For each remaining candidate bounding box, if its IoU value with the current highest-confidence bounding box is greater than or equal to the threshold, it is considered that the two boxes overlap too much, and this candidate bounding box is removed from the list.

[0105] Add the current highest-confidence bounding box to the list of retained bounding boxes. Repeat the steps until all candidate bounding boxes have been processed. Each time, select the candidate bounding box with the highest confidence in the current list as the new reference bounding box, and remove the candidate bounding boxes that overlap with it too much until the list of candidate bounding boxes is empty. The finally retained set of candidate bounding boxes is the detection result after non-maximum suppression.

[0106] Further optimize the remaining candidate bounding boxes, adjust the position and size of the candidate bounding boxes, and use the key point information of the face (such as the positions of eyes, nose, mouth, etc.) to fine-tune the candidate bounding boxes to improve the accuracy of face detection.

[0107] According to the number of face candidate bounding boxes, count the number of passengers in the vehicle, and at the same time record the position information of each passenger for facial feature extraction and analysis.

[0108] S3.5: Collect real-time driving data, including the real-time position of the vehicle (obtain longitude and latitude coordinates through the GPS positioning module), driving direction (obtain the vehicle heading angle through the gyroscope and magnetometer), vehicle speed (measured through the vehicle speed sensor), etc.

[0109] S3.6: Integrate the obtained number of in-vehicle passengers, facial feature information, and driving data into an emergency information package. The emergency information package is organized in a preset JSON format and includes basic information (such as emergency event code, timestamp), passenger information (face feature vector, number of passengers), and driving data (location, direction, vehicle speed).

[0110] In summary, this method collects in-vehicle passenger information and vehicle driving data after the trigger signal is confirmed to be abnormal, and packages them into an emergency information package to build a comprehensive on-site data foundation; integrates multi-dimensional data such as in-vehicle passenger facial features, vehicle speed, driving direction, and GPS location information into an emergency data carrier, enhancing the overall emergency response and decision-making capabilities of the system.

[0111] S4: Transmit the emergency information package to the background rescue system, determine whether there are dangerous personnel, and then send an emergency call instruction with the in-vehicle situation.

[0112] Specifically, the in-vehicle communication module receives the emergency information package and encrypts and transmits the emergency information package to the background rescue system through the communication channel with the highest priority. The background rescue system analyzes the face features in the emergency information package based on a deep learning algorithm to determine whether there are dangerous personnel, calculates the optimal rescue route according to the vehicle location information, and sends a rescue instruction with the in-vehicle situation to the nearest police station.

[0113] Case 1: It is determined that there are dangerous personnel

[0114] If the analysis result shows that there are dangerous personnel in the vehicle, the background system immediately calculates the optimal rescue route according to the vehicle location information. The background system sends an emergency call instruction with the in-vehicle situation, which details the facial feature information of the dangerous personnel, the real-time location, driving direction, vehicle speed, and the number of in-vehicle passengers of the vehicle, adjusts the dynamic range and sampling frequency of the hand and pedal pressure signals, and at the same time uses the feedback data as a correction basis to recalibrate the weight distribution of the normal driving behavior pattern in the deep learning model;

[0115] Case 2: It is determined that there are no dangerous personnel

[0116] After the background rescue system analyzes the passenger facial features and determines that there are no dangerous personnel in the vehicle. At this time, the background system will not immediately send an emergency call instruction, but further analyze the driving data of the vehicle. Record and file this emergency information for subsequent tracking and analysis of the vehicle's driving situation. At the same time, the background system will regularly monitor the vehicle's driving data to ensure the driving safety of the vehicle.

[0117] The feedback information includes the statistical comparison results of driving behavior data such as vehicle acceleration, steering angle, and brake pedal pressure with the normal driving probability distribution. After internal data cross-comparison, the feedback information is transmitted to data collection and deep learning model training to adjust the initial set values of the hand and pedal pressure signals in the pressure sensor group, such as correcting the sampling time window and signal amplitude range, and at the same time updating normal driving behavior. Case 3: Unable to determine whether there are dangerous persons

[0118] The background rescue system may not be able to determine whether there are dangerous persons in the vehicle. For example, due to poor image quality, unclear facial features, etc., the facial recognition algorithm cannot accurately judge.

[0119] The feedback information includes multi-dimensional data comparison results, auxiliary parameters such as ambient light and in-vehicle noise level, and statistical characteristics of historical driving records and in-vehicle environment data. An instruction is sent to the in-vehicle camera to re-collect in-vehicle images, and at the same time, the threshold parameters of grayscale conversion, binarization in the image processing module and face candidate box screening in the convolutional neural network are adjusted. The deep learning model expands the input features during the next data processing, adding in-vehicle environment parameters as auxiliary judgment bases. The adjustment formula for the removal threshold of face candidate box screening is:

[0120]

[0121] In the formula, IoU n is the adjusted removal threshold, IoU i is the initial removal threshold, IoU a is the average intersection over union between candidate boxes, M is the number of overlapping pairs of candidate boxes, and β is the adjustment factor (usually with a value range from 0 to 1).

[0122] In summary, this method forms a closed-loop emergency response link by transmitting the emergency information package to the background rescue system and sending an emergency call instruction with the in-vehicle situation after confirming whether there are dangerous persons in the vehicle, realizing remote intelligent processing and automatic decision-making of emergency information, thus avoiding false alarms or delayed responses, providing a scientific and reliable decision-making basis and data support for rescue operations, and improving the response timeliness and safety prevention and control level of the in-vehicle emergency call system.

[0123] This embodiment also provides a computer device, applicable to the situation of the in-vehicle concealed driver emergency call method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement all or part of the steps of the method described in the embodiments of the present invention as proposed above.

[0124] This embodiment also provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method in any optional implementation manner of the above embodiment. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM for short), electrically erasable programmable read-only memory (EEPROM for short), erasable programmable read-only memory (EPROM for short), programmable read-only memory (PROM for short), read-only memory (ROM for short), magnetic memory, flash memory, magnetic disk or optical disc.

[0125] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. For technical details not described in detail in this embodiment, reference can be made to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0126] In summary, this method can improve the agility and security of emergency response, significantly improve the recognition accuracy of the system for abnormal driving behaviors, provide a scientific basis and data support for emergency rescue decisions, enhance the overall emergency response and decision-making capabilities of the system, and improve the response timeliness and safety prevention and control level of the in-vehicle emergency call system.

Claims

1. A hidden driver emergency call method in a car, characterized by: include, A pressure sensor group is installed at the bottom of the steering wheel and the clutch pedal in the vehicle, and a trigger signal is generated when a combined trigger pattern of hand pressure value and pedal pressure value is detected; Analyze the driver's daily driving behavior data based on the deep learning model, and compare the trigger signal with the pre-stored normal driving behavior pattern; When the trigger signal deviates from the normal driving behavior pattern and meets the preset emergency trigger threshold, the passenger information and vehicle driving data in the vehicle are obtained, and the passenger information and vehicle driving data are packaged into an emergency information package; The emergency information package is transmitted to the background rescue system, and after determining whether there are dangerous people, an emergency call instruction with the situation inside the car is sent.

2. The hidden driver emergency call method in a vehicle as claimed in claim 1, characterized in that: The pressure sensor group includes a first pressure sensor and a second pressure sensor, wherein the first pressure sensor is arranged at the bottom of the steering wheel to collect hand pressure values, and the second pressure sensor is arranged at the clutch pedal to collect pedal pressure values.

3. The hidden driver emergency call method in a vehicle as claimed in claim 2, characterized in that: The combined trigger mode includes recording and forming a first trigger signal if it is detected that the hand pressure value and the pedal pressure value increase from an initial value to a value greater than a first set threshold value within a first set time; If it is detected that the hand pressure value and the pedal pressure value change above and below the second set threshold value within the second set time, then record and generate a second trigger signal; If it is detected that the hand pressure value and the pedal pressure value decrease to the initial pressure value within the third set time, then the third trigger signal is recorded and generated.

4. The hidden driver emergency call method in a vehicle as claimed in claim 3, characterized in that: The analysis of the driver's daily driving behavior data based on the deep learning model includes: Obtain the driver's daily driving behavior data, including normal driving and abnormal driving, and annotate the data according to driving standards to form training sets, validation sets, and test sets. The driver's daily driving behavior data includes vehicle acceleration, steering angle, brake pedal pressure, throttle depth, and vehicle speed; The driver's daily driving behavior data is normalized, and the formula is: In the formula, x is the original data, x min and x max are the minimum and maximum values ​​of the data respectively; Use the training set data to train the neural network weight parameters and bias parameters of the bidirectional long short-term memory network model, and minimize the loss function through the Adam optimizer; Evaluate model performance on the validation set, adjust hyperparameters, and verify the generalization ability of the model on the test set; By calculating the probability distribution of normal driving behavior in the training data, the emergency trigger threshold is set outside the 95% confidence interval of the probability distribution.

5. The hidden driver emergency call method in a vehicle as claimed in claim 4, characterized in that: The acquisition of in-vehicle passenger information and vehicle driving data includes: Immediately activate the vehicle's internal camera to capture the interior images covering all seating areas in the vehicle, and collect vehicle interior images and real-time driving data; Locate the face area in the vehicle interior image to generate a face candidate frame, and filter the face candidate frame to obtain face position information; Count the number of passengers in the car according to the coordinates of facial key points in the face position information, and extract facial features to form a facial feature vector; The number of passengers in the car, facial feature information, and driving data obtained are integrated and organized into an emergency information package according to the preset JSON format.

6. The hidden driver emergency call method in a vehicle as claimed in claim 5, characterized in that: The generating of the face candidate frame comprises: The collected vehicle interior images are grayed and binarized, and the graying conversion formula is: Z=0.299*R+0.587*G+0.114*B Where Z represents the grayscale image point, R represents the red color in the image, G represents the green color in the image, and B represents the blue color in the image; Use the threshold method to convert the grayscale image into a black and white image, setting pixels greater than the threshold to black and pixels less than the threshold to white; The CNN convolutional neural network is used to extract local and global features from the binarized image, and the face position and size are determined according to the response intensity of the feature map to generate a face candidate frame. The screening of the face candidate frames includes: Collect all detected face candidate frames and sort them in descending order according to the confidence scores. Create an empty list to save the final candidate boxes. From the sorted candidate box list, take each candidate box in turn and calculate the degree of overlap between the candidate box and the remaining candidate boxes. The formula is: Among them, IoU is the intersection over union ratio of the candidate boxes, A and B represent the areas of the two candidate boxes respectively; Set the removal threshold IoU i , for each remaining candidate box, if the intersection-over-union ratio with the current highest confidence box is ≥ the removal threshold IoU i , then remove the candidate box from the list; Repeat the steps until all candidate frames have been processed, and each time select the candidate frame with the highest confidence in the current list as the new reference frame until the candidate frame list is empty. The final set of candidate frames retained is the filtered face candidate frame.

7. The hidden driver emergency call method in a vehicle as claimed in claim 6, characterized in that: The sending of an emergency call instruction with the vehicle status after determining whether there is a dangerous person includes: After receiving the emergency information package, the back-end rescue system analyzes the facial features of the passengers; If the analysis results show that there are dangerous people in the car, the background system immediately calculates the optimal rescue route based on the vehicle location information, sends an emergency call instruction with the status of the car, adjusts the first set threshold and set time of the hand pressure value and the pedal pressure value, and recalibrates the weight distribution of the normal driving behavior mode in the deep learning model; If the analysis results show that there are no dangerous persons in the car, the background system will stop sending emergency call instructions, record and file the emergency information, feedback the statistical comparison results of the probability distribution of the current driving behavior data and the normal driving behavior data, calibrate and adjust the hand pressure value and the initial value of the pedal pressure in the pressure sensor group, and update the normal driving behavior data at the same time; If the background rescue system cannot determine whether there are dangerous people in the car, and the facial features cannot be accurately judged, the on-board camera is instructed to re-collect the images in the car, and the removal threshold for generating the face candidate frame screening is adjusted, and the in-car environment data is obtained for secondary facial feature analysis. The adjustment formula is: Where, IoU n is the adjusted removal threshold, IoU i is the initial removal threshold, IoU a is the average intersection-over-union ratio between candidate boxes, M is the number of overlapping pairs of candidate boxes, and β is the adjustment factor.

Images