A driving playback method in an intelligent cockpit

Abstract

The application provides a driving playback method in an intelligent cockpit, comprising: playback recording system construction, specifically comprising: recorded playback acquisition; construction of a driving route identification system, specifically comprising: identification of a playback route, correct driving acquisition of the same route driving; identification of a playback scene and time, specifically comprising: establishment of a scene identification model, judgment of playback duration, judgment of key part interception; automatic analysis and explanation of driving content; intelligent recommendation of video playing time; protection of driver privacy; and operation video recommendation based on predicted road conditions.

Description

Technical Field

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

[0002] In smart cockpits, newly licensed drivers often encounter unfamiliar operations or make incorrect decisions. For these novice drivers, an instructor cannot always be present. Therefore, if a car could learn to operate based on the cockpit experience of others, it would significantly improve driving safety. They could then learn better driving techniques when encountering similar road conditions and scenarios in the future. Summary of the Invention

[0003] This invention provides a driving playback method in a smart cockpit, mainly including:

[0004] The playback recording system construction specifically includes: acquiring recorded playback data; constructing a driving segment recognition system, which specifically includes: identifying the driving segment to be played back and acquiring correct driving data for the same driving segment; identifying the playback scene and time, which specifically includes: establishing a scene recognition model, determining the playback duration, and determining the extraction of key parts; automatically analyzing and explaining driving content; intelligently recommending video playback time; protecting driver privacy; and recommending operation videos based on predicted road conditions.

[0005] Further, optionally, the playback recording system is constructed by including:

[0006] A playback recording system is constructed to record driving route data and driving operation data through in-vehicle cameras and navigation systems after obtaining user authorization and consent. Encryption methods are used to protect the driver's personal identity information and driver behavior data recorded by various in-vehicle devices during the drive. When the driver drives the car to a predetermined route, the playback recording system installed in the vehicle will begin recording, and recording will end when the driver leaves the route. The recorded video will then be uploaded to the system's storage space. This includes: access to recorded playback.

[0007] The acquisition of the recorded playback specifically includes:

[0008] When a video playback request is received, the video to be played back is acquired; the video to be played back is divided into multiple video segments according to a preset time interval, and the initial state of the video segments is a pending state; it is sequentially determined whether there is a moving target in the multiple video segments; if there is a moving target in the current video segment, the segment is marked as the initial moving segment, and the state of the current video segment is the moving target state; if the moving target is detected to have paused, the segment is marked as the terminated moving segment, and the state of the current video segment is the no moving target state; all video segments between the current video segment and the terminated moving segment are marked as the predicted target state, and the video segments in the target state and the predicted target state are added to the analysis list; the video segments in the analysis list that contain moving targets are acquired, and the video segments that contain moving targets are added to the playlist; when the number of video segments in the playlist reaches a first preset threshold, the video segments in the playlist are played back to provide input for subsequent video analysis.

[0009] Further optionally, the construction of the driving segment identification system includes:

[0010] Based on the driver's historical driving trajectory points and historical roads and their information, a historical trajectory database is constructed to store several historical driving trajectory points and their corresponding historical roads and their information. Each historical driving trajectory point includes a set of historical vectors corresponding to each road segment. GPS trajectory points uploaded by the vehicle are obtained. These GPS trajectory points are matched with the historical driving trajectory points in the historical trajectory database to obtain a set of GPS trajectory points uploaded by the vehicle that do not match the pre-stored historical roads. GPS trajectory points in the set that are useless for newly added historical road information are filtered out. The remaining GPS trajectory points in the set are saved as newly added historical driving trajectory points. The RoadMapReverseUpdate method is used to output newly added historical roads based on the newly added historical driving trajectory points. The vehicle's driving data during the driver's journey on the newly added historical roads is obtained. The driving data is processed to form standard samples of vehicle driving data suitable for learning algorithms. The standard samples of vehicle driving data are input into a pre-constructed learning model to output the road information of the newly added historical roads. This includes: identifying road segments that need to be replayed; and obtaining correct driving information for the same road segments.

[0011] The identification of road sections requiring playback specifically includes:

[0012] While recording driving video, the system also records the corresponding geographical location information and time; combines the geographical location information into a driving trajectory and displays it on an electronic map; based on national traffic accident records, if the number of accidents exceeds a preset threshold, the road segment is determined to be a road segment that needs to be replayed; the geographical location of the road segment that needs to be replayed is determined based on the geographical location information corresponding to the current time of the driving video.

[0013] The acquisition of correct driving information for the same road segment specifically includes:

[0014] The system acquires the current vehicle's driving data and the road segment data it travels on. The driving data includes vehicle driving data and driver status data. It then quantifies the current vehicle's driving data and road segment data to obtain driving data in a preset format. This preset format driving data and road segment data are input into a pre-created regression classification model. Based on the output of the regression classification model, it determines whether the current road segment and the road segment to be replayed are the same. If the current road segment and the road segment to be replayed are the same, the system begins recording driving data and the driver's corresponding driving habit data, and establishes a corresponding driving database based on the driving data and driving habit data.

[0015] Further, optionally, the identification of the playback scene and time includes:

[0016] The system uses an in-vehicle camera to capture real-time images of the vehicle's surroundings. These images are then input into a scene recognition model to detect various road facilities and traffic conditions, allowing for real-time acquisition of the current driving route data. Next, the system acquires the output data from a driver behavior recognition model, statistically analyzing the number of different driving behaviors under varying road conditions. Based on these statistical results, the system analyzes the correlation between various driving and non-driving behaviors and road condition information. Finally, it calculates the probability of a driver performing a specific behavior under a given road condition. This process includes: establishing a scene recognition model; determining playback duration; and identifying and extracting key sections.

[0017] The establishment of the scene recognition model specifically includes:

[0018] A road condition recognition neural network model is established based on object detection methods. Road condition information for vehicle driving includes road infrastructure information and road traffic information. The first step involves collecting a large number of road images from the vehicle's perspective using an in-vehicle camera as both a training and testing set. These images are then preprocessed and feature extracted to obtain candidate bounding boxes for various road infrastructure objects in the images. The second step involves manually labeling the candidate bounding boxes extracted from the training set, assigning each object category as a label. The third step involves inputting the candidate bounding boxes into a classifier for training. The classifier learns the sample features of each label and establishes classification decision rules. Finally, the test set is input into the classifier, and classification tests are conducted according to the established classification decision rules to evaluate the accuracy of the classification results. The model parameters are then adjusted based on the test results.

[0019] The determination of playback duration specifically includes:

[0020] The process involves compiling a record of the duration of actions performed on the video based on the video behavior records that meet the requirements. This includes: determining the video actions performed on the video based on the video behavior records that meet the requirements, as well as the start and end times of each video action; determining the video duration corresponding to each video action based on the start and end times of each video action; and summing the video durations corresponding to multiple video actions performed on the video to obtain the total duration of actions performed on the video.

[0021] The determination of extracting key information specifically includes:

[0022] This method, applied to an in-vehicle computer (ECU), a playback recording system, and a host computer, involves the following steps for acquiring key information: the ECU's sensors collect vehicle parameter data in real time and send this data to a microcontroller within the ECU; the microcontroller acquires vehicle position data and driving data in real time, including acceleration and / or angular velocity data; the microcontroller sends all or part of the vehicle parameter data, driving data, and position data to the host computer; an event analysis module analyzes the vehicle parameter data to determine events during vehicle operation; if the event is a pre-set abnormal event, the playback recording system is activated; after video recording is completed, an event video positioning module and an event video cutting module determine key parts of the video; and, based on existing national standards, a focus is placed on road sections prone to traffic accidents.

[0023] Further, optionally, the automatic analysis and explanation of driving content includes:

[0024] Using acquired playback and vehicle driving data, including vehicle driving data and driver status data, the driving content explanation model is input to analyze the driving operations of different drivers on the same road segment, and to count the number of various driving behaviors of drivers under the same road conditions. A Support Vector Machine (SVM) classifier is trained based on known driving and non-driving behavior samples to obtain an SVM classifier that can distinguish between driving and non-driving behaviors. The SVM classifier distinguishes driver behaviors as driving or non-driving behaviors. Based on the statistical results and the classification results of the SVM classifier, the correlation between various driving and non-driving behaviors of drivers and road condition information is analyzed, and the probability of a driver performing a certain behavior under a certain road condition is calculated. When the probability of different drivers performing a certain behavior under a certain road condition exceeds a preset threshold, the behavior is determined to be a correct operation under that road condition, and the playback and driving data of that behavior are sent to the video recommendation system.

[0025] Further, optionally, the intelligent recommended video playback time includes:

[0026] When the driver is planning the navigation route, the system will prompt the driver about special road sections that may be encountered during the trip based on the big data already connected. The system will also prompt the driver whether they want to learn the correct operation and input the available time for the navigation route. The system will automatically determine the available time for learning and determine the video playback time based on the length of the time. If the vehicle speed is detected to be 0 based on the vehicle driving data obtained by the ECU, it means that the vehicle is stationary. At this time, the video will be pushed to the driver.

[0027] Further, optionally, the protection of driver privacy includes:

[0028] The driver's privacy information is encrypted using homomorphic encryption. Data operations are performed on the encrypted privacy information without decryption. The operation result is then processed by a decryption algorithm to obtain the same result as the plaintext operation.

[0029] A method for driving playback in a smart cockpit is characterized in that the system includes:

[0030] The driver inputs the navigation destination, and the vehicle's ECU logs into the in-vehicle navigation app via wireless network to check traffic conditions. It collects traffic control notices from the traffic department's official website, personnel notifications from the 122 accident hotline, and information broadcast by the local traffic radio station. Based on the query request sent by the prompting terminal, it sends information about special road sections and traffic conditions along the current route to the prompting terminal of the connected vehicle. The system determines whether there are any pre-recorded playback sections along the current route. If so, it recommends videos and driving behavior data of other drivers on that route stored in the driving database. After the driver inputs the available driving time, the system estimates the time available for learning operation videos and recommends videos to the driver based on the estimated time.

[0031] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0032] This invention can record the correct driving operations of other drivers on a specific road segment, intelligently identify the specific traffic segment the current driver is in, and push driving videos matching the current specific traffic segment to the current driver based on the recorded correct operations. By analyzing the time the current driver can watch the video, it extracts the correct driving video of the corresponding time length, provides driving suggestions and practical operation methods for the same specific road segment, and analyzes the correct operations, automatically explaining the correct methods through video analysis technology, and comparing incorrect driving postures and methods. It can efficiently identify video content that has an educational effect on driving behavior and does not involve driver privacy, effectively protecting the privacy of other drivers, ensuring that the playback content can explain relevant driving operation issues to the greatest extent possible while avoiding recording content involving driver privacy. Attached Figure Description

[0033] Figure 1 This is a flowchart of a driving playback method in an intelligent cockpit according to the present invention. Detailed Implementation

[0034] 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.

[0035] Figure 1 This is a flowchart of a driving playback method in an intelligent cockpit according to the present invention. Figure 1 As shown, a driving playback method in a smart cockpit according to this embodiment may specifically include:

[0036] Step 101, Construction of the playback recording system.

[0037] A playback recording system is constructed to record driving route data and driving operation data through in-vehicle cameras and navigation systems after obtaining user authorization and consent. Encryption methods are used to protect the driver's personal identification information and driver behavior data recorded by various in-vehicle devices during the drive. When the driver drives the car to the designated route, the playback recording system installed in the vehicle will begin recording. Recording will end when the driver leaves the designated route, and the recorded video will then be uploaded to the system's storage space.

[0038] Recorded playback available.

[0039] When a video playback request is received, the video to be played back is acquired; the video to be played back is divided into multiple video segments according to a preset time interval, and the initial state of each video segment is a pending state; it is sequentially determined whether there is a moving target in the multiple video segments; if there is a moving target in the current video segment, the segment is marked as the initial moving segment, and the current video segment is in the state of having a moving target; if the moving target is detected to have stopped moving, the segment is marked as the terminated moving segment, and the current video segment is in the state of having no moving target; all video segments between the current video segment and the terminated moving segment are marked as the predicted target state, and the video segments in the state of having a target and the predicted target state are added to the list to be analyzed; the video segments in the list to be analyzed that contain moving targets are acquired, and the video segments containing moving targets are added to the playlist; when the number of video segments in the playlist reaches a first preset threshold, the video segments in the playlist are played back to provide input for subsequent video analysis. Determining whether there is a moving target in the video segments helps to filter valid videos recorded while the vehicle is moving and filter out invalid videos when the vehicle is stationary.

[0040] Step 102: Construct a driving route identification system.

[0041] Based on the driver's historical driving trajectory points and historical roads and their information, a historical trajectory database is constructed to store several historical driving trajectory points and their corresponding historical roads and information. Each historical driving trajectory point includes a set of historical vectors corresponding to each road segment. GPS trajectory points uploaded by the vehicle are obtained. These GPS trajectory points are matched with the historical driving trajectory points in the historical trajectory database to obtain a set of GPS trajectory points uploaded by the vehicle that do not match the pre-stored historical roads. GPS trajectory points in the set that are useless for newly added historical road information are filtered out. The remaining GPS trajectory points in the set are saved as newly added historical driving trajectory points. The RoadMapReverseUpdate method is used to output newly added historical roads based on the newly added historical driving trajectory points. The vehicle's driving data during the driver's journey on the newly added historical roads is obtained. The driving data is processed to form standard samples of vehicle driving data suitable for learning algorithms. The standard samples of vehicle driving data are input into a pre-constructed learning model to output the road information of the newly added historical roads. The RoadMapReverseUpdate method is a method for drawing road maps based on GPS trajectory point data, which improves the accuracy of road map drawing. The method addresses the symmetrical distribution of GPS trajectory points on roads by directly fitting these points using the least squares method, and treating the fitted straight line as the road centerline. Road information includes road length, number of motor vehicle lanes, number of non-motor vehicle lanes, width of motor vehicle lanes, width of non-motor vehicle lanes, single / double lane information, and width of the green belts on both sides.

[0042] Identify road sections that need to be replayed.

[0043] While recording driving video, the system also records the corresponding geographical location information and time; combines the geographical location information into a driving trajectory and displays it on an electronic map; based on national traffic accident records, if the number of accidents exceeds a preset threshold, the road segment is determined to be a road segment that needs to be replayed; the geographical location of the road segment that needs to be replayed is determined based on the geographical location information corresponding to the current time of the driving video.

[0044] Obtain the correct driving information for driving on the same road section.

[0045] The system acquires the current vehicle's driving data and the road segment data it travels on. The driving data includes vehicle driving data and driver status data. It then quantifies the current vehicle's driving data and road segment data to obtain driving data in a preset format. This preset format driving data and road segment data are input into a pre-created regression classification model. Based on the output of the regression classification model, it determines whether the current road segment and the road segment to be replayed are the same. If the current road segment and the road segment to be replayed are the same, the system begins recording driving data and the driver's corresponding driving habit data, and establishes a corresponding driving database based on the driving data and driving habit data. The vehicle driving data includes vehicle speed, vehicle acceleration, vehicle heading angle, vehicle light status, and vehicle gear status; the driver status data includes the number of times the driver looks left and right, the frequency and force of pressing the accelerator and brake pedals, etc.; driving data recording methods include, for example, acquiring a road image ahead of the vehicle; determining the distance the vehicle travels forward based on the road image; driving habit data recording methods include, for example, recognizing lane markings in the road image; prompting the driver with driving operations based on the traffic markings; the driving database is used to store driving data and driving habit data.

[0046] Step 103: Identify the playback scene and time.

[0047] The vehicle uses an in-vehicle camera to capture real-time images of the vehicle's surroundings. These images are then input into a scene recognition model to detect various road facilities and traffic conditions, allowing for real-time acquisition of current road segment data. The output data from a driver behavior recognition model is then obtained, and the number of different driving behaviors of the driver under varying road conditions is statistically analyzed. Based on these statistical results, the correlation between the driver's various driving and non-driving behaviors and road condition information is analyzed. Finally, the probability of the driver performing a specific behavior under a given road condition is calculated.

[0048] Establish a scene recognition model.

[0049] A road condition recognition neural network model is established based on object detection methods. Road condition information for vehicle driving includes road infrastructure information and road traffic information. The first step involves collecting a large number of road images from the vehicle's perspective using an in-vehicle camera as training and testing sets. These images are then preprocessed and feature extracted to obtain candidate boxes for various road infrastructure objects in the images. The second step involves manually labeling the candidate boxes extracted from the training set, assigning the object category within each box as a label. The third step involves inputting the candidate boxes into a classifier for training. The classifier learns the sample features of various labels and establishes classification decision rules. Finally, the test set is input into the classifier, and classification tests are performed according to the established classification decision rules to evaluate the accuracy of the classification results. Model parameters are then adjusted based on the test results. For example, road condition information for vehicle driving includes road infrastructure information and road traffic information. The in-vehicle camera acquires the current specific road conditions. Road infrastructure information mainly includes highway type information, highway curve and slope information, road surface unevenness information, speed bump information, road snow and water accumulation information, traffic light information, and road sign information. Road traffic information includes traffic accident information, traffic congestion information, construction site information, and hazard information. The front-facing camera captures images of the surrounding environment while the car is driving. The system then inputs these images into a scene recognition model to identify various road facilities and traffic conditions on the current road.

[0050] Determine the playback duration.

[0051] The process involves compiling a record of the duration of actions performed on the video based on the video behavior records that meet the requirements. This includes: determining the video actions performed on the video based on the video behavior records that meet the requirements, as well as the start and end times of each video action; determining the video duration corresponding to each video action based on the start and end times of each video action; and summing the video durations corresponding to multiple video actions performed on the video to obtain the total duration of actions performed on the video.

[0052] Determine and extract the key parts.

[0053] This method, applied to an in-vehicle computer (ECU), a playback recording system, and a host computer, involves the following steps for acquiring key information: the ECU's sensors collect vehicle parameter data in real time and send this data to a microcontroller within the ECU; the microcontroller acquires vehicle position data and driving data in real time, including acceleration and / or angular velocity data; the microcontroller sends all or part of the vehicle parameter data, driving data, and position data to the host computer; an event analysis module analyzes the vehicle parameter data to determine events during vehicle operation; if the event is a pre-set abnormal event, the playback recording system is activated; after video recording is completed, an event video positioning module and an event video cutting module determine key parts of the video; and, based on existing national standards, a focus is placed on road sections prone to traffic accidents. The host computer is a computer that can accept control commands and is used to process and calculate vehicle parameter data, driving data, and position data sent by the ECU; the playback and recording system is a software system running on the host computer for recording and playing back videos; the event analysis module is used to determine one or more of the events such as rapid acceleration, rapid deceleration, and sharp turning during vehicle operation by analyzing the positioning data; the event video positioning module is used to determine the position of the event video corresponding to the original video by analyzing the start and end times of the original video and the start and end times of the event video to be cut; the event video cutting module is used to cut the original video according to the determined position to obtain the event video.

[0054] Step 104: Automatically analyze and explain the driving content.

[0055] Using acquired playback and vehicle driving data, including vehicle driving data and driver status data, the driving content explanation model is input to analyze the driving operations of different drivers facing the same road segment, and to count the number of various driving behaviors of drivers under the same road conditions. A Support Vector Machine (SVM) classifier is trained based on known driving and non-driving behavior samples to obtain an SVM classifier that can distinguish between driving and non-driving behaviors. The SVM classifier distinguishes driver behaviors as driving or non-driving behaviors. Based on the statistical results and the classification results of the SVM classifier, the correlation between various driving and non-driving behaviors and road condition information is analyzed, and the probability of a driver performing a certain behavior under a certain road condition is calculated. When the probability of different drivers performing a certain behavior under a certain road condition exceeds a preset threshold, the behavior is determined to be a correct operation under that road condition, and the playback and driving data of that behavior are sent to the video recommendation system. For example, on a road section about to enter a tunnel, if the system detects that 80% of passing vehicles choose to slow down and turn on their headlights within a day, exceeding the system's set 70%, then the system determines that slowing down and turning on the headlights is the correct action on that road section and sends this action to the recommendation system.

[0056] Step 105: Intelligent recommendation of video playback time.

[0057] When the driver plans the navigation route, the system uses big data to indicate potential special road sections along the journey and prompts the driver to learn the correct operating procedures. The driver then inputs the available time for the navigation route, and the system automatically determines the available time for learning and adjusts the video playback time accordingly. Based on vehicle driving data obtained from the ECU, if the vehicle's speed is detected to be 0, it indicates the vehicle is stationary, and a video is pushed to the driver. Special road sections include accident-prone areas, consecutive curves, and sharp turns.

[0058] Step 106: Protect driver privacy.

[0059] The driver's privacy information is encrypted using homomorphic encryption. Data computation is performed on the encrypted privacy information without decryption, and the result is then processed by a decryption algorithm to obtain the same result as the plaintext computation. The driver's personal identification information and the driver's behavior data recorded by various in-vehicle devices during driving are considered personal privacy information, and privacy leaks must be prevented during information collection, transmission, verification, and subsequent use. For example, the core of homomorphic encryption is to directly use encrypted data for computation, and then obtain the result through a decryption algorithm, yielding the same result as directly using unencrypted data for computation. If a homomorphic encryption function F and plaintext M exist, encrypting M yields ciphertext F(M) = m. Ciphertext m is transmitted from the sender to the receiver. The receiver does not need to decrypt ciphertext m; instead, it directly performs computation C on ciphertext m to obtain the result C(m) = n. Then, it decrypts n and calculates f to obtain the result N = f(n). By directly performing the operation C on the plaintext M, we can also obtain the result N, i.e., C(M) = N. Therefore, in this operation, f(C(F(M))) = C(M), i.e., F(C(M)) = C(F(M)). In this way, the receiver can calculate the data normally and obtain the correct result without knowing the plaintext content.

[0060] Step 107: Based on predicted road conditions, recommend operation videos.

[0061] The driver inputs the navigation destination, and the vehicle's ECU logs into the in-vehicle navigation app via wireless network to check traffic conditions. It collects traffic control notices from the traffic department's official website, personnel notifications from the 122 accident hotline, and information broadcast by local traffic radio stations. Based on the query request sent by the prompting terminal, it sends information about special road sections and traffic conditions along the current route to the prompting terminal of the connected vehicle. The system determines if there are any pre-recorded playback sections along the current route. If so, it recommends videos and driving behavior data of other drivers on that route stored in the driving database. After the driver inputs the available driving time, the system estimates the time available for learning operation videos and recommends videos based on the estimated time. In-vehicle navigation apps include: Gaode Maps, Baidu Maps, Google Maps, etc.

Claims

1. A method for driving playback in an intelligent cabin, characterized in that, The method comprises: The playback recording system construction comprises: recorded playback acquisition; construction of a driving route identification system, which comprises: identification of a playback route, correct driving of the same route; identification of a playback scene and time, which comprises: establishment of a scene identification model, which comprises: establishment of a road condition identification neural network model based on a target detection method; road condition information of car driving includes road facility information and road traffic information; first, a large number of road images under the perspective of the car are collected by the camera in the car as a training set and a test set, and then image preprocessing and feature extraction are performed to obtain the candidate boxes of various road facility objects in the pictures; second, the candidate boxes extracted in the training set are manually labeled, and the object categories in the candidate boxes are used as labels; third, the candidate boxes are input into the classifier for training, the classifier learns the sample features of various labels, and a classification decision rule is established; finally, the test set is input into the classifier, and the classification test is performed according to the classification decision rule established by the classifier, the accuracy of the classification result is evaluated, the model parameters are adjusted according to the test result, the playback time is determined, and the key part is determined; automatically analyze and explain the driving content; intelligently recommend video playing time; protect the privacy of the driver; recommend operation video based on predicted road conditions.

2. The method of claim 1, wherein, The playback recording system construction comprises: The playback recording system construction comprises: When the driver drives the car to the predetermined route, the playback recording system loaded in the car will start recording, and when the driver drives away from the route, the recording will end, and then the recorded video will be uploaded to the system storage space; comprising: recorded playback acquisition; The recorded playback acquisition comprises: When a video playback request is received, a video to be played back is obtained; the video to be played back is divided into a plurality of video segments according to a preset time interval, and the initial state of the video segment is a to-be-processed state; whether there is a moving target in the plurality of video segments is judged in turn; if there is a moving target in the current video segment, the segment is marked as an initial motion segment, and the state of the current video segment is a moving target state; if it is detected that the moving target pauses, the segment is marked as a terminated motion segment, and the state of the current video segment is a non-moving target state; all video segments between the current video segment and the terminated motion segment are marked as predicted target state, and the video segments with the target state and the predicted target state are added to an analysis list; the video segments with the moving target in the analysis list are obtained, and the video segments with the moving target are added to a play list; when the number of video segments in the play list reaches a first preset threshold, the video segments in the play list are played back to provide input for subsequent video analysis.

3. The method of claim 1, wherein, The construction of the driving route identification system comprises: According to the historical driving trajectory points and historical roads and historical road information of the driver, a historical trajectory database is constructed for storing a plurality of historical driving trajectory points and historical roads and historical road information corresponding to the historical driving trajectory points, the historical driving trajectory points including a set of historical vectors corresponding to each road segment; GPS trajectory points uploaded by a driving vehicle are acquired; the GPS trajectory points are matched with the historical driving trajectory points in the historical trajectory database to obtain a set of GPS trajectory points uploaded by the vehicle and not matching the historical pre-stored roads; GPS trajectory points in the set that are useless for newly added historical road information are filtered out; the remaining GPS trajectory points in the set are saved as newly added historical driving trajectory points; a RoadMapReverseUpdate method is used to output newly added historical roads according to the newly added historical driving trajectory points; driving data of the vehicle during driving of the driver on the newly added historical roads is acquired; the driving data is processed to form vehicle driving data standard samples suitable for learning algorithms; The vehicle driving data standard samples are input into a pre-constructed learning model to output road information of the newly added historical roads; including: identifying a playback road segment; obtaining correct driving on the same road segment; The identification of the playback road segment specifically includes: recording geographical position information and time corresponding to the driving video while recording the driving video; combining the geographical position information into a driving trajectory for display on an electronic map; determining that the road segment is a playback road segment according to a national traffic accident frequency record if the frequency is greater than a preset threshold; determining the geographical position of the playback road segment according to the geographical position information corresponding to the current time of the driving video; The correct driving on the same road segment is obtained, specifically including: acquiring driving data of a current vehicle and driving road segment data of the current vehicle, the driving data including vehicle driving data and driver state data; performing quantitative processing on the driving data and the driving road segment data of the current vehicle to obtain driving data in a preset form; inputting the driving data in the preset form and the driving road segment data into a pre-created regression classification model to determine whether the current road segment and the playback road segment are the same road segment according to an output result of the regression classification model; if the current road segment and the playback road segment are the same road segment, recording driving data and driver corresponding driving habit data, and establishing a corresponding driving database according to the driving data and the driving habit data.

4. The method of claim 1, wherein, The identification of the playback scene and time includes: real-time collection of environment images of the vehicle through a vehicle-mounted camera, input of the environment images into a scene recognition model to detect various road facilities and traffic conditions in the images, and real-time acquisition of current driving road segment data; then, output data of a driver driving behavior recognition model are acquired, and the number of various driving behaviors of the driver under different road condition information is counted; based on the statistical result, the correlation between various driving behaviors and non-driving behaviors of the driver and the road condition information is analyzed; the probability of the driver making a certain behavior under a certain road condition is calculated; including: establishing a scene recognition model; determining a playback duration; determining a key part to be intercepted; The determination of the playback duration specifically includes: According to the qualified video behavior record, the behavior duration record for the video is counted, including: according to the qualified video behavior record, the video behavior for the video and the start time and end time of each video behavior are determined; according to the start time and end time of each video behavior, the video duration corresponding to each video behavior is determined; the video duration corresponding to multiple video behaviors for the video is accumulated to obtain the behavior duration for the video; The judgment of the key part specifically includes: The method for acquiring the key part comprises: the sensor of the ECU collects vehicle parameter data in real time, and sends the collected vehicle parameter data to the microcontroller in the ECU; the microcontroller acquires the position data of the vehicle in real time, and collects the driving data of the vehicle, the driving data comprising: acceleration data and / or corner speed data of the vehicle; the microcontroller sends all or part of the vehicle parameter data, the driving data and the position data to the upper computer; the event analysis module determines the event in the driving process of the vehicle by analyzing the vehicle parameter data, and if the event is a pre-set abnormal event, the playback recording system is started; after the video recording is completed, the event video positioning module and the event video cutting module are used to determine the key part of the video; according to the existing national standard, the road traffic accident-prone sections are focused on.

5. The method of claim 1, wherein, The automatic analysis and explanation of the driving content comprises: The driving content explanation model is input by the acquired playback and vehicle driving data, the vehicle driving data comprising vehicle driving data and driver state data, the driving operation of different drivers facing the same road section is analyzed, and the number of various driving behaviors of the driver facing the same road condition information is counted; A support vector machine (SVM) classifier is trained according to known driving behaviors and non-driving behavior samples to obtain the support vector machine (SVM) classifier capable of distinguishing driving behaviors and non-driving behaviors; The behavior of the driver is distinguished as a driving behavior or a non-driving behavior according to the SVM classifier; Based on the statistical results and the classification results of the SVM classifier, the correlation of various driving behaviors and non-driving behaviors of the driver with the road condition information is analyzed, and the probability of the driver making a certain behavior under a certain road condition is calculated; when the probability of different drivers making a certain behavior under a certain road condition exceeds a pre-set threshold, it is determined that the behavior is the correct operation under the road condition, and the behavior playback and driving data are sent to the video recommendation system.

6. The method of claim 1, wherein, The intelligent recommendation of the video playing time comprises: When the driver plans the current navigation route, the driver is prompted about the special road section that will be encountered in the current journey according to the accessed big data, and whether the driver wants to learn the correct operation is prompted, and the available time of the current navigation route is input, the system automatically judges the time available for learning, and determines the video playing time according to the length of the time; according to the vehicle driving data obtained by the ECU, if the vehicle speed is 0, it indicates that the vehicle is in a stationary state, and the video is pushed to the driver at this time.

7. The method of claim 1, wherein, The protection of the privacy of the driver comprises: The privacy information of the driver is encrypted by homomorphic encryption, data operation is carried out without decrypting the encrypted privacy information, and the operation result is decrypted by a decryption algorithm to obtain the same result as the plaintext operation.

8. The method of claim 1, wherein, The operation video is recommended based on the predicted road condition, and the operation video recommendation based on the predicted road condition comprises: The driver inputs the navigation destination, the vehicle-mounted computer ECU logs in the vehicle-mounted navigation app through a wireless network to query traffic conditions, collects traffic control announcement information from the official website of the traffic department, 122 accident alarm personnel announcement information, information broadcast by the local traffic radio station, and sends the special road section and traffic conditions existing in the current driving route to the prompt terminal of the vehicle connected with itself according to the query request sent by the prompt terminal; the system judges whether there is a playback road section that has been recorded in the current driving road section, if there is, the system recommends the correct driving operation video and driving behavior data of other drivers in the road section stored in the driving database to the driver, and after the driver inputs the available time for the current driving, the system estimates the time available for learning the operation video of the current driving, and recommends the video to the driver according to the estimated time.

Images