A conference monitoring method in an autonomous driving intelligent cockpit

Abstract

The application provides a conference monitoring method in an automatic driving intelligent cockpit, comprising: based on driver information, completing driver information system construction, specifically comprising: driver information collection, driver personal information, and conference information encryption protection; based on road condition historical information, constructing a road condition prior information analysis model; based on road condition risk and whether the driver wants to start a conference, constructing a historical road condition risk information and conference start model; constructing a road condition information and conference association model, specifically comprising: based on driver conference information, completing conference level evaluation, based on road condition information, constructing a road condition recognition model, based on road condition risk, constructing a road condition risk evaluation model, based on road condition risk and conference importance level, adjusting a conference warning model; based on conference information, constructing a conference duration prediction model; constructing a conference duration and trip evaluation model.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for monitoring meetings in an autonomous driving smart cockpit. Background Technology

[0002] With the continuous development of the autonomous driving field, L3 and L4 level automated driving vehicles have begun field testing in China. This demonstrates that in specific driving modes, the system is responsible for executing all dynamic driving control of the vehicle, and the driver only needs to respond to the system's intervention in real time in special situations. Meetings are one of the most important management methods, but when team members' schedules are too tight, holding meetings mid-trip is a necessary choice to improve time utilization and advance project progress. When meetings are held, how can the intelligent cockpit, based on road condition information, achieve a balance between road risk warnings and meeting time when meetings of different importance levels are in progress? During autonomous driving, when encountering accident-prone road sections, what methods and judgment standards should the intelligent cockpit use to remind drivers who are in meetings? In certain specific situations, should the intelligent cockpit interrupt the meeting to warn the driver of potential dangers? Furthermore, when holding meetings during a trip, is the required trip duration sufficient to support the successful completion of the meeting, and will the driver's subsequent travel plans be affected by the meeting? Summary of the Invention

[0003] This invention provides a method for monitoring meetings in an autonomous driving smart cockpit, mainly including:

[0004] Based on driver information, a driver information system is constructed. This construction includes: driver information collection; encryption and protection of driver personal information and meeting information; construction of a priori road condition information analysis model based on historical road condition information; construction of a historical road condition risk information and meeting initiation model based on road condition risk and whether the driver intends to start a meeting; and construction of a road condition information and meeting association model. This model includes: completing meeting level assessment based on driver meeting information; constructing a road condition identification model based on road condition information; constructing a road condition risk assessment model based on road condition risk; adjusting meeting warning models based on road condition risk and meeting importance level; constructing a meeting duration prediction model based on meeting information; and constructing a meeting duration and trip evaluation model.

[0005] Further optionally, the construction of the driver information system based on driver information includes:

[0006] Building a driver information system includes information collection and information encryption protection. After obtaining driver authorization, the system records the driver's identity characteristics and driving information, and then encrypts and stores the information using a homomorphic encryption algorithm. This includes: driver information collection; encryption protection of driver personal information and meeting information.

[0007] The driver information collection specifically includes:

[0008] The information collection process involves inputting the driver's basic personal information, ID card photo, driver's license photo, and facial data into the information system. The smart cockpit extracts feature information through intelligent recognition to complete the construction of the smart cockpit information system. The driver's entered identity information and driving information are compared and authenticated with the public security department's authentication interface. If the verification is successful, the information is stored in the system; if the verification fails, the driver is required to re-enter the information for identity authentication. Once the driver's information is successfully verified, the smart cockpit will grant the driver access to start the driving system conference.

[0009] The encryption protection of the driver's personal information and meeting information specifically includes:

[0010] The information encryption protection operation uses homomorphic encryption algorithm to encrypt the driver's basic personal information, ID card information, driving information and facial information. After the original data is encrypted, the ciphertext is obtained. When extracting data, the ciphertext is decrypted to obtain the original data.

[0011] Further, optionally, the construction of the prior information analysis model for road conditions based on historical road condition information includes:

[0012] Construct a historical traffic database and extract road and vehicle information. Road information extraction: Export the city's road network data from the GIS database. This data includes the road's start and end points, the latitude and longitude of the central axis, road grade, and number of lanes. Extend the coordinate range of each road outwards from the central axis according to the number of lanes. Then, discretize the city's Akm*Akm area using Bm*Bm grids and label the grids corresponding to the road coordinates. Vehicle information extraction: Extract daily GPS data and information (latitude and longitude) of taxis in the city. The system determines the vehicle's location based on several factors: whether it is carrying passengers; the vehicle's coordinates are obtained from GPS information and mapped to the existing road grid to identify the road it belongs to; based on the traffic department's experience and data characteristics, the road is segmented using traffic lights, bridges, intersections, urban areas, and road signs; finally, the speed is calculated over a time interval of T; the speed data of all vehicles at the same time and location over the past ten days are weighted and averaged to obtain the predicted speed for the driver on that day; the Euclidean distance from the driver's origin to the destination is divided by the predicted speed to obtain the time required for the driver to reach the destination.

[0013] Further, optionally, the construction of a historical road condition risk information and meeting initiation model based on road condition risk and whether the driver wants to start the meeting includes:

[0014] A historical traffic accident database is constructed, and the locations of traffic accidents in the driver's city over the past five years are marked in the road condition prior information analysis model. Traffic accidents are divided into four levels: minor accidents, general accidents, major accidents, and extremely serious accidents, with risk levels defined as A, B, C, and D respectively. A risk level of 0 is assigned for no traffic accidents. The five levels increase arithmetic progression. The traffic risk level score for the driver's journey from origin to destination within three days is calculated and denoted by the letter E. The arrival time predicted by the road condition prior information analysis model is denoted by the letter T. A meeting initiation judgment model is constructed: S = FX * E, where FX is the indicator function, X indicates whether the driver needs to initiate a meeting (FX = 1 if there is a need, FX = 0 if there is no need), and S is the risk index. When the traffic risk level exceeds a preset threshold, the system does not recommend the driver to initiate a meeting.

[0015] Further, optionally, the construction of the traffic information and meeting association model includes:

[0016] A meeting level assessment model is constructed to determine the importance level of a meeting based on meeting indicators. Environmental image information collected by a pre-set camera device during vehicle movement is acquired, and a convolutional neural network (CNN) image classification algorithm is used to train the acquired image information. The trained CNN is then used in a road condition recognition system. The meeting level assessment model evaluates the meeting level, the road condition recognition model identifies road driving information and road facility information, and the road condition risk assessment model obtains a risk coefficient through the road condition recognition model. A risk index and risk level are calculated using the meeting level and the road condition risk coefficient HX=FA*P. When the meeting level is X, level X-1 risk is ignored, and a corresponding broadcast method is adopted. Before reaching the location of a historical accident, the meeting warning model, based on voice recognition, broadcasts a warning for the accident site if no voice activity by the driver during the meeting is detected. This includes: completing a meeting level assessment based on driver meeting information; constructing a road condition recognition model based on road condition information; constructing a road condition risk assessment model based on road condition risk; and adjusting the meeting warning model based on road condition risk and meeting importance level.

[0017] The process of assessing the meeting level based on driver meeting information specifically includes:

[0018] Meeting levels are categorized based on three indicators: meeting organizer, meeting urgency, and meeting size. Meeting organizers are divided into seven levels: ordinary employee, team leader, director, supervisor, general manager, president, and chairman. Their quantitative scores increase progressively, with a difference of A. Meeting urgency is divided into three levels: ordinary meeting, temporary ordinary meeting, and temporary emergency meeting. Their quantitative scores also increase progressively, with a difference of B. Meeting size is divided into small, medium, large, and extra-large meetings. Their quantitative scores increase progressively, with a difference of C. A meeting level assessment model is constructed: P = E + F + G / 3, where P represents the importance of the meeting, E represents the meeting organizer's score, F represents the meeting urgency score, and G represents the meeting size score. All meeting types are permuted and combined, and a weighted average is applied based on the scores to categorize meetings into four levels: Level 1, Level 2, Level 3, and Level 4.

[0019] The construction of the road condition recognition model based on road condition information specifically includes:

[0020] Autonomous driving road condition information is divided into road driving information and road facility information, mainly including highway type, highway curves and slopes, traffic lights, road signs, road surface unevenness, speed bumps, road closures and maintenance, and road snow and water accumulation. Road traffic information includes traffic congestion information, traffic accident information, construction closure information, and hazard information. Using the vehicle's front as the center of symmetry and a pre-defined rectangular pixel area directly in front as the pre-defined region, a large number of driving road condition images are captured and processed, and training, validation, and test sets are selected. Regional images are extracted to establish a road condition state dataset, and blurred images are removed. The multi-scale Retinex image enhancement algorithm is applied to unify the image illumination intensity, and image labels are assigned to the road condition state dataset. The training set is imported into a convolutional neural network model for training and recognition, and the model parameters are trained. The test set is imported into the convolutional neural network model for parameter adjustment. After model training is complete, the validation set is imported again to evaluate the accuracy of image classification and recognition results, and further parameter optimization is performed.

[0021] The road condition risk assessment model, based on road condition risk, specifically includes:

[0022] Road condition risk warnings are assessed using traffic warning signs. These signs are categorized into six types: intersection warning signs, turning signs, lane change signs, uphill / downhill signs, caution signs, and other signs. Historical traffic databases are accessed to obtain the locations, causes, and times of accidents. Accidents are then classified into six types based on the six warning sign categories, and the frequency of each type is counted. A risk road condition model is constructed. For each route from the driver's origin to their destination, the number of accidents occurring on that route on that day is divided by the total number of accidents occurring in that city / district on that day, denoted by the letter A. When A exceeds a preset threshold, the system uses the road condition recognition model to identify warning signs and, based on the historical database, alerts the driver to the type of accident that may occur on the upcoming route.

[0023] The adjustment of the meeting alert model based on road condition risk and meeting importance level specifically includes:

[0024] The meeting level assessment model is imported into the road condition risk assessment model. The meeting assessment model is adjusted as HX=FA*P, where P is the meeting assessment level (P=1, 2, 3, 4), and FA is an indicator function. When A exceeds a preset threshold, FA=1; when A is below the preset threshold, FA=0. Road condition risk warnings are imported into the C-V2X scenario library warning indicators. The C-V2X scenario library warnings are divided into four levels: Level 1: speed limit warning, red light violation warning; Level 2: forward collision warning, intersection collision warning, blind spot warning, emergency braking warning; Level 3: wrong-way overtaking warning, vulnerable road user collision warning; Level 4: abnormal vehicle alert, vehicle loss of control warning. The C-V2X scenario library warnings are determined by V2X application algorithms to identify each dangerous scenario. Based on the meeting assessment model, when the meeting level P exceeds level 2, the meeting warning model, based on the C-V2X road condition scenario library warnings, ignores the level 1 risk warning information. When a higher level risk warning occurs, the warning system determines the driver's meeting status, and the system does not provide road sign warnings. Meanwhile, before reaching the site of a historical accident, the meeting warning model, based on the recognition of the voice system, will issue a warning broadcast for the accident site if no voice activity of the driver is detected during the meeting.

[0025] Further optionally, the step of constructing a meeting duration prediction model based on meeting information includes:

[0026] Using a speech recognition model and a meeting level assessment model, speech recognition is performed on meeting content to extract keywords and classify meeting types. A smart cockpit meeting recognition corpus is built to record the meeting duration of each driver meeting and correlate it with the corresponding meeting assessment indicators and meeting levels. Using a BP neural network algorithm, the meeting assessment indicators are divided into training and test sets. The training set is used to train the BP neural network, and the test set is used to calculate the error between the predicted and actual values, while optimizing parameters to reduce the error.

[0027] A method for monitoring meetings in an autonomous driving smart cockpit is characterized in that the system comprises:

[0028] By using a road condition prior information analysis model, the predicted travel time is obtained based on the driver's origin and destination information. By using a meeting duration prediction model, the meeting type is classified according to meeting evaluation indicators, and the predicted meeting duration is obtained. A meeting duration and trip evaluation model is constructed to determine whether the driver has an upcoming trip. When the predicted meeting duration is less than the trip duration and the next trip has been arranged, the driver is reminded to speed up the meeting.

[0029] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0030] To address the aforementioned issues, this invention provides a method for monitoring meetings in an autonomous driving smart cockpit. After verifying the driver's identity information, the smart cockpit system determines, based on road risk information, whether to recommend starting a meeting between the departure point and the destination. It also assesses which information should be alerted and which should be ignored based on the importance level of the meeting, and how the system should alert the driver. Simultaneously, the system helps the driver maximize the use of travel time for meetings by predicting the meeting duration and the required travel time. Attached Figure Description

[0031] Figure 1 This is a flowchart of a meeting monitoring method in an autonomous driving smart cockpit according to the present invention.

[0032] Figure 2 This is a schematic diagram of a meeting detection method in an autonomous driving smart cockpit according to the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] Figure 1 This is a flowchart of a meeting monitoring method in an autonomous driving intelligent cockpit according to the present invention. Figure 1 As shown, a meeting monitoring method in an autonomous driving smart cockpit, as described in this embodiment, may specifically include:

[0035] Step 101: Based on driver information, complete the construction of the driver information system.

[0036] Building a driver information system includes information collection and information encryption protection. After obtaining the driver's authorization, the driver's identity characteristics and driving information are entered, and the information is encrypted and stored using a homomorphic encryption algorithm.

[0037] Driver information collection.

[0038] The information collection process involves inputting the driver's basic personal information, ID card photo, driver's license photo, and facial data into the information system. The smart cockpit extracts feature information through intelligent recognition to complete the construction of the smart cockpit information system. The driver's entered identity information and driving information are compared and authenticated with the public security department's authentication interface. If the verification is successful, the information is stored in the system; if the verification fails, the driver is required to re-enter the information for identity authentication. Once the driver's information is successfully verified, the smart cockpit will grant the driver access to start the driving system conference.

[0039] Driver personal information and meeting information are encrypted and protected.

[0040] The information encryption protection operation uses homomorphic encryption algorithms to encrypt the driver's basic personal information, ID card information, driving information, and facial information. The original data is encrypted to obtain ciphertext, and the ciphertext is decrypted to obtain the original data when data is retrieved. For example, a key generation function generates a key and some public parameters for encrypting the data. This key is held by the user. The user uses the key to encrypt the original data to obtain ciphertext CT. Then, ciphertext CT is stored in the information system using the data processing method f provided by the user. When the user needs to retrieve the data, the system sends the ciphertext to the user, who decrypts the data using their own key to obtain the result. Plaintext M, encrypted using the encryption function F(M)=C, the system processes f(C)=N, and the user-held key decrypts the data K(N)=M to obtain the result.

[0041] Step 102: Based on historical road condition information, construct a priori road condition information analysis model.

[0042] A historical traffic database was constructed, and road and vehicle information was extracted. Road information extraction: Road network data for the driver's city was exported from the GIS database. This data included the road's start and end points, the latitude and longitude of the central axis, road grade, and number of lanes. Extending outwards from the central axis according to the number of lanes, the coordinate range of each road was obtained. Then, the city's Akm*Akm area was discretized using Bm*Bm grids, and the grids were labeled to correspond with the road coordinates. Vehicle information extraction: Daily GPS data and information for taxis in the city, including latitude and longitude and whether the vehicle was carrying passengers, were extracted. Vehicle coordinates were obtained from the GPS information and mapped to the previous road grid to determine the road to which the vehicle belonged. Based on the traffic department's experience and data characteristics, the road was segmented according to traffic lights, bridges, intersections, urban areas, and road signs. Finally, speed was calculated over a time interval of T. The weighted average of all vehicle speed data from the same location at the same time over the past ten days was used as the predicted speed for the driver's journey on that road segment that day. Divide the Euclidean distance from the driver's starting point to the destination by the predicted driving speed to obtain the time required for the driver to reach the destination.

[0043] Step 103: Based on road condition risks and whether the driver wants to start a meeting, construct a historical road condition risk information and meeting initiation model.

[0044] A historical traffic accident database is constructed, and the locations of traffic accidents in the driver's city over the past five years are marked in the road condition prior information analysis model. Traffic accidents are divided into four levels: minor accidents, general accidents, major accidents, and extremely serious accidents, with risk levels defined as A, B, C, and D respectively. A risk level of 0 is assigned for no traffic accidents. The five levels increase arithmetically. The traffic risk level score for the driver's journey from origin to destination within three days is calculated and denoted by the letter E. The arrival time predicted by the road condition prior information analysis model is denoted by the letter T. A meeting initiation judgment model is constructed: S = FX * E, where FX is the indicator function, X indicates whether the driver needs to initiate a meeting (FX = 1 if there is a need, FX = 0 if there is no need), and S is the risk index. When the traffic risk level exceeds a preset threshold, the system does not recommend the driver to initiate a meeting. When Xiao Cai's traffic risk level on his way to Xiao Gong's house is C+D and the predicted time is 25 minutes, and Xiao Cai wants to hold a meeting on the way, the meeting initiation judgment model yields a result S=C+D, indicating that multiple accidents have occurred on this road section within the past three days. The system then suggests that the driver should not initiate a meeting.

[0045] Step 104: Construct a model linking traffic information and meetings.

[0046] A meeting level assessment model is constructed to determine the importance level of a meeting based on meeting indicators. Environmental image information collected by a pre-set camera device during vehicle movement is acquired. A convolutional neural network (CNN) image classification algorithm is used to train the acquired image information, and the trained CNN is then used in a road condition recognition system. The meeting level assessment model evaluates the meeting level, the road condition recognition model identifies road driving information and road facility information, and the road condition risk assessment model obtains a risk coefficient through the road condition recognition model. The risk index and risk level are calculated using the meeting level and the road condition risk coefficient HX=FA*P. When the meeting level is X, level X-1 risk is ignored, and a corresponding broadcast method is adopted. Before reaching the location of a historical accident, the meeting warning model, based on voice recognition, broadcasts a warning for the accident site if no voice activity by the driver during the meeting is detected.

[0047] Based on driver meeting information, complete the meeting level assessment.

[0048] Meeting levels are categorized based on three indicators: meeting organizer, meeting urgency, and meeting size. Meeting organizers are divided into seven levels: ordinary employee, team leader, director, supervisor, general manager, president, and chairman. Their quantitative scores increase progressively, with a difference of A. Meeting urgency is divided into three levels: ordinary meeting, temporary ordinary meeting, and temporary emergency meeting. Their quantitative scores also increase progressively, with a difference of B. Meeting size is divided into small, medium, large, and extra-large meetings. Their quantitative scores increase progressively, with a difference of C. A meeting level assessment model is constructed: P = E + F + G / 3, where P represents the importance of the meeting, E represents the meeting organizer's score, F represents the meeting urgency score, and G represents the meeting size score. All meeting types are permuted and combined, and a weighted average is applied based on the scores to categorize meetings into four levels: Level 1, Level 2, Level 3, and Level 4. For example, if A=0.1, B=0.2, and C=0.1, then a small, ordinary meeting held by a regular employee would score 0, while a small, ordinary meeting held by a team leader would score 0.1. All meeting types are scored and ranked, divided into four groups representing four levels. For instance, the meeting size is differentiated by a preset threshold for the number of participants. If a team leader is preparing to hold a 5-person meeting in the smart cockpit, this meeting would be considered a small, temporary, ordinary meeting. Therefore, the meeting importance level determined by the meeting level assessment model would be one.

[0049] A road condition recognition model is constructed based on road condition information.

[0050] Autonomous driving road condition information is divided into road driving information and road facility information, mainly including highway type, highway curves and slopes, traffic lights, road signs, road surface unevenness, speed bumps, road closures and maintenance, and road snow and water accumulation. Road traffic information includes traffic congestion information, traffic accident information, construction closure information, and hazard information. Using the vehicle's front as the center of symmetry and a pre-defined rectangular pixel area directly in front as the pre-defined region, a large number of driving road condition images are captured and processed, and training, validation, and test sets are selected. Regional images are extracted to establish a road condition state dataset, and blurred images are removed. The multi-scale Retinex image enhancement algorithm is applied to unify the image illumination intensity, and image labels are assigned to the road condition state dataset. The training set is imported into a convolutional neural network model for training and recognition, and the model parameters are trained. The test set is imported into the convolutional neural network model for parameter adjustment. After model training is complete, the validation set is imported again to evaluate the accuracy of image classification and recognition results, and further parameter optimization is performed. For example: Import images of road surface flooding, train a convolutional neural network (CNN) with 11 layers. After three convolutional and pooling processes, the data is stored in the prediction set through fully connected layers 1 and 2. Calculate the error between the prediction and the true value, optimize the parameters using gradient descent, and save the parameters after training. Apply the trained CNN model to a road condition recognition model.

[0051] Based on road condition risks, a road condition risk assessment model is constructed.

[0052] Road condition risk warnings are assessed using traffic warning signs. These signs are categorized into six types: intersection warning signs, turning signs, lane change signs, uphill / downhill signs, caution signs, and other signs. Historical traffic databases are accessed to obtain the locations, causes, and times of accidents. Accidents are then classified into six types based on the six warning sign categories, and the frequency of each type is counted. A risk road condition model is constructed. For each route from the driver's origin to their destination, the number of accidents occurring on that route on that day is divided by the total number of accidents occurring in that city / district on that day, denoted by the letter A. When A exceeds a preset threshold, the system uses a road condition recognition model to identify warning signs and, based on the historical database, alerts the driver to the type of accident that may occur on the upcoming route. For example, in the risk road condition model, if the value of A exceeds 5%, the system considers the risk of an accident on that route to be high. On his way to Chenghua Avenue, Zhang San was warned by the system that the road section was prone to accidents. The system indicated that a rear-end collision had occurred 500 meters ahead on Chenghua Avenue and that there was a slippery road sign 50 meters ahead. He was advised to be careful of the road conditions.

[0053] Adjust the meeting alert model based on road condition risk and meeting importance level.

[0054] The meeting level assessment model is imported into the road condition risk assessment model. The meeting assessment model is adjusted as HX=FA*P, where P is the meeting assessment level (P=1, 2, 3, 4), and FA is an indicator function. When A exceeds a preset threshold, FA=1; when A is below the preset threshold, FA=0. Road condition risk warnings are imported into the C-V2X scenario library warning indicators. The C-V2X scenario library warnings are divided into four levels: Level 1: speed limit warning, red light violation warning; Level 2: forward collision warning, intersection collision warning, blind spot warning, emergency braking warning; Level 3: wrong-way overtaking warning, vulnerable road user collision warning; Level 4: abnormal vehicle alert, vehicle loss of control warning. The C-V2X scenario library warnings are determined by V2X application algorithms to identify each dangerous scenario. Based on the meeting assessment model, when the meeting level P exceeds level 2, the meeting warning model, based on the C-V2X road condition scenario library warnings, ignores the level 1 risk warning information. When a higher level risk warning occurs, the warning system determines the driver's meeting status, and the system does not provide road sign warnings. Simultaneously, before reaching the location of a historical accident, the meeting warning model, based on voice recognition, issues warnings for the accident site when no voice activity from the driver is detected during the meeting. For example, if the meeting assessment level is Level 1, it will remind the driver of a speed limit warning and a traffic light sign 50 meters ahead, indicating an accident-prone area. If the meeting assessment level is higher than Level 2, the system will not remind the driver of the speed limit warning and traffic light sign, but will instead remind the driver of a high-risk rear-end collision area 200 meters ahead if the driver is not speaking. If the system detects any overtaking in the wrong direction, it will immediately issue a voice warning to the driver to be aware of the driving situation. When a vulnerable road user collision warning occurs, the system will simultaneously issue voice and text alerts to warn the driver of road safety. When a vehicle loss of control warning occurs, the system will immediately interrupt the meeting and warn the driver to be aware of the road conditions.

[0055] Step 105: Based on the meeting information, construct a meeting duration prediction model.

[0056] This system uses a speech recognition model and a meeting level assessment model to perform speech recognition on meeting content, extract keywords, and classify meeting types. A smart cockpit meeting recognition corpus is built to record the meeting duration of each driver meeting and correlate it with corresponding meeting assessment indicators and meeting levels. A backpropagation (BP) neural network algorithm is used, dividing the meeting assessment indicators into training and testing sets. The training set is used to train the BP neural network, while the testing set is used to calculate the error between predicted and actual values, and to optimize parameters to reduce this error. For example, if the meeting host introduces themselves as "Zhang San, the head of the administrative department, is here to hold an urgent meeting for our group," the speech recognition system will identify the host as a head, the meeting size as a small meeting, and the meeting as an urgent, ordinary meeting. The three meeting assessment indicators—host, urgency, and size—are used as the input layer of the neural network (n=3), the output layer as the meeting duration (m=1), and three hidden layers. After training, the model takes these three indicators as input and outputs the predicted meeting duration.

[0057] Step 106: Construct a model for evaluating meeting duration and itinerary.

[0058] By using a road condition prior information analysis model, the predicted travel time is obtained based on the driver's origin-to-destination information. Similarly, a meeting duration prediction model is used, categorizing meetings by evaluation indicators to determine their predicted duration. A meeting duration and travel time assessment model is then constructed to determine if the driver has an upcoming trip. If the predicted meeting duration is less than the travel time and the next trip is already scheduled, the driver is alerted to expedite the meeting. For example, if Zhang San decides to start a meeting using the smart cockpit, and the system estimates the meeting duration at 25 minutes and the travel time at 20 minutes, the smart cockpit will remind Zhang San to expedite the meeting if he has an upcoming trip.

[0059] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for monitoring meetings in an autonomous driving smart cockpit, characterized in that, The method includes: Based on driver information, a driver information system was constructed, which specifically includes: driver information collection, encryption and protection of driver personal information and meeting information. Based on historical road condition information, a road condition prior information analysis model is constructed; Based on road condition risks and whether drivers want to start a meeting, a model for combining historical road condition risk information with meeting initiation is constructed. The process of constructing a road condition information and meeting association model includes: constructing a meeting level assessment model based on driver meeting information, whereby the meeting level assessment model evaluates the meeting level, thus completing the meeting level assessment; constructing a road condition identification model based on road condition information; constructing a road condition risk assessment model based on road condition risk; and adjusting the meeting warning model based on road condition risk and meeting importance level. Specifically, adjusting the meeting warning model based on road condition risk and meeting importance level includes: importing the meeting level assessment model into the road condition risk assessment model; adjusting the meeting assessment model, HX=FA*P, where P is the meeting assessment level (P=1,2,3,4), and FA is an indicator function. When A exceeds a preset threshold, FA=1; when A is below a preset threshold, FA=1. At the threshold, FA=0; the road condition risk warning imports warning indicators from the C-V2X scenario library, which is divided into four levels; the C-V2X scenario library warning is determined by the V2X application algorithm for each dangerous scenario; based on the meeting evaluation model, when the meeting level P exceeds level two, the meeting warning model, based on the C-V2X road condition scenario library warning, ignores the first-level risk warning information. When a higher-level risk warning appears, the warning system makes a judgment based on the driver's meeting status, and the system will not issue road sign warnings; at the same time, before reaching the location of a historical accident, the meeting warning model, based on the voice recognition system, will issue a warning broadcast for the accident site if no voice behavior of the driver is detected during the meeting. Based on meeting information, construct a meeting duration prediction model; The model for evaluating meeting duration and itinerary includes: using a priori road condition information analysis model to obtain the predicted trip duration based on the driver's origin and destination information; using a meeting duration prediction model to classify meeting types based on meeting evaluation indicators to obtain the predicted meeting duration; and using a model for evaluating meeting duration and itinerary to determine whether the driver has an upcoming trip. When the predicted meeting duration is less than the trip duration and the next trip has already been arranged, the driver is reminded to expedite the meeting.

2. The method according to claim 1, wherein, The construction of the driver information system based on driver information includes: Building a driver information system includes information collection and information encryption protection. After obtaining driver authorization, the system records the driver's identity characteristics and driving information, and then encrypts and stores the information using a homomorphic encryption algorithm. This includes: driver information collection; encryption protection of driver personal information and meeting information. The driver information collection specifically includes: The information collection process involves inputting the driver's basic personal information, ID card photo, driver's license photo, and facial data into the information system. The smart cockpit extracts feature information through intelligent recognition to complete the construction of the smart cockpit information system. The driver's entered identity information and driving information are compared and authenticated with the public security department's authentication interface. If the verification is successful, the information is stored in the system. If the verification fails, the driver is required to re-enter the information for identity authentication. Once the driver's information is successfully verified, the smart cockpit will grant the driver access to start the driving system conference. The encryption protection of the driver's personal information and meeting information specifically includes: The information encryption protection operation uses homomorphic encryption algorithm to encrypt the driver's basic personal information, ID card information, driving information and facial information. After the original data is encrypted, the ciphertext is obtained. When extracting data, the ciphertext is decrypted to obtain the original data.

3. The method according to claim 1, wherein, The aforementioned construction of a priori road condition information analysis model based on historical road condition information includes: Construct a historical traffic database and extract road and vehicle information. Road information extraction: Export the city's road network data from the GIS database. This data includes the road's start and end points, the latitude and longitude of the central axis, road grade, and number of lanes. Extend the coordinate range of each road outwards from the central axis according to the number of lanes. Then, discretize the city's Akm*Akm area using Bm*Bm grids and map the grids to the road coordinates. Vehicle information extraction: Extract daily GPS data and information from taxis in the city, including latitude and longitude and whether the vehicle is carrying passengers. Obtain the vehicle's coordinates from the GPS information, and then map these coordinates to the previous road grids to determine the road to which the vehicle belongs. Based on the experience and data characteristics of the transportation department, the road is segmented by traffic lights, bridges, intersections, urban areas, and road signs. Finally, the speed is calculated at a time interval of T. The speed data of all vehicles at the same time and place in the past ten days are weighted and averaged to obtain the predicted speed for the driver to travel on that road segment on that day. The Euclidean distance from the driver's origin to the destination is divided by the predicted speed to obtain the time required for the driver to reach the destination.

4. The method according to claim 1, wherein, The aforementioned model, which constructs historical road condition risk information and a meeting initiation model based on road condition risks and whether the driver wants to start a meeting, includes: A historical traffic accident database is constructed, and the locations of traffic accidents in the driver's city over the past five years are marked in the road condition prior information analysis model. Traffic accidents are divided into four levels: minor accidents, general accidents, major accidents, and extremely serious accidents, with risk levels defined as A, B, C, and D respectively. A risk level of 0 is assigned for no traffic accidents. The five levels increase arithmetic progression. The traffic risk level score for the driver's journey from origin to destination within three days is calculated and denoted by the letter E. The arrival time predicted by the road condition prior information analysis model is denoted by the letter T. A meeting initiation judgment model is constructed: S = FX * E, where FX is the indicator function, X indicates whether the driver needs to initiate a meeting (FX = 1 if there is a need, FX = 0 if there is no need), and S is the risk index. When the traffic risk level exceeds a preset threshold, the system does not recommend the driver to initiate a meeting.

5. The method according to claim 1, wherein, The construction of the road condition information and meeting association model includes: A meeting level assessment model is constructed to determine the importance level of a meeting based on meeting indicators. Environmental image information collected by a pre-set camera device during vehicle movement is acquired, and a convolutional neural network (CNN) image classification algorithm is used to train the acquired image information. The trained CNN is then used in a road condition recognition system. The meeting level assessment model evaluates the meeting level, the road condition recognition model identifies road driving information and road facility information, and the road condition risk assessment model obtains a risk coefficient through the road condition recognition model. A risk index and risk level are calculated using the meeting level and the road condition risk coefficient HX=FA*P. When the meeting level is X, level X-1 risk is ignored, and a corresponding broadcast method is adopted. Before reaching the location of a historical accident, the meeting warning model, based on voice recognition, broadcasts a warning for the accident site if no voice activity by the driver during the meeting is detected. This includes: completing a meeting level assessment based on driver meeting information; constructing a road condition recognition model based on road condition information; constructing a road condition risk assessment model based on road condition risk; and adjusting the meeting warning model based on road condition risk and meeting importance level. The process of assessing the meeting level based on driver meeting information specifically includes: Meeting levels are categorized based on three indicators: meeting organizer, meeting urgency, and meeting size. Meeting organizers are divided into seven levels: ordinary employee, team leader, director, general manager, president, and chairman, with their quantitative scores increasing arithmetically, a difference of A. Meeting urgency is divided into three levels: ordinary meeting, temporary ordinary meeting, and temporary emergency meeting, with their quantitative scores increasing arithmetically, a difference of B. Meeting size is divided into small, medium, large, and extra-large meetings, with their quantitative scores increasing arithmetically, a difference of C. A meeting level assessment model is constructed: P = E + F + G / 3, where P represents the importance of the meeting, E represents the meeting organizer's score, F represents the meeting urgency score, and G represents the meeting size score. All meeting types are permuted and combined, and a weighted average is applied based on the scores to categorize meetings into four levels: Level 1, Level 2, Level 3, and Level 4. The construction of the road condition recognition model based on road condition information specifically includes: Autonomous driving road condition information is divided into road driving information and road facility information, mainly including highway type, highway curves and slopes, traffic lights, road signs, road surface unevenness, speed bumps, road closures and maintenance, and road snow and water accumulation. Road traffic information includes traffic congestion information, traffic accident information, construction closure information, and hazard information. A large number of driving road condition images are captured and processed using the vehicle's front as the symmetrical center plane and a pre-defined rectangular pixel area directly in front of the vehicle's camera. Training, validation, and test sets are selected. A road condition state dataset is established by extracting images from the selected areas and removing blurry images. The multi-scale Retinex image enhancement algorithm is applied to unify the image illumination intensity, and image labels are assigned to the road condition state dataset. The training set is imported into a convolutional neural network model for training and recognition, and the model parameters are trained. The test set is imported into the convolutional neural network model for parameter adjustment. After model training is complete, the validation set is imported to evaluate the accuracy of image classification and recognition results, and further parameter optimization is performed. The aforementioned road condition risk assessment model, based on road condition risk, specifically includes: Road condition risk warnings are assessed using traffic warning signs. These signs are categorized into six types: intersection warning signs, turning signs, lane change signs, uphill / downhill signs, caution signs, and other signs. The system reads historical traffic databases to obtain the locations, causes, and times of accidents. Accidents are then classified into six types based on these six warning sign categories, and the frequency of each type is recorded. A risk road condition model is constructed by dividing the number of accidents occurring on a given road segment from the driver's origin to their destination by the total number of accidents occurring in that city / district on that day, denoted by the letter A. When A exceeds a preset threshold, the system uses a road condition recognition model to identify warning signs and uses historical databases to alert the driver to the type of accident that may occur on the next section of the road.

6. The method according to claim 1, wherein, The method for constructing a meeting duration prediction model based on meeting information includes: Using a speech recognition model and a meeting level assessment model, speech recognition is performed on meeting content to extract keywords and classify meeting types. A smart cockpit meeting recognition corpus is built to record the meeting duration of each driver meeting and correlate it with the corresponding meeting assessment indicators and meeting levels. Using a BP neural network algorithm, the meeting assessment indicators are divided into training and test sets. The training set is used to train the BP neural network, and the test set is used to calculate the error between the predicted and actual values, while optimizing parameters to reduce the error.

Images