Method and system for extracting abnormal scenes from smart car driving data based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based intelligent vehicle driving data abnormal scene extraction method and system, which relates to the field of data analysis, including retrieving basic information of each intelligent vehicle to obtain abnormal risk assessment value of each intelligent vehicle; matching to obtain driving data extraction ratio of each intelligent vehicle; obtaining driving data packets of each intelligent vehicle, statistically analyzing comprehensive scene data, and importing into an integrated AI cloud computing center after processing; the present invention obtains basic information of intelligent vehicles from multiple data transmission terminals through an artificial intelligence-based intelligent vehicle driving data abnormal scene extraction method, analyzes and evaluates abnormal risks, dynamically adjusts data extraction ratio, and efficiently identifies risky vehicles. Use professional modules to mine data features, and annotate abnormal scenes through cluster analysis to improve data analysis efficiency.

Description

Technical Field

[0001] This invention relates to the field of data analysis, specifically to a method and system for extracting abnormal scenarios from intelligent vehicle driving data based on artificial intelligence. Background Art

[0002] With the development of artificial intelligence technology, intelligent vehicles have become an important research direction in the automotive industry. Intelligent vehicles, equipped with various sensors (such as radar, cameras, and lidar) and advanced computing systems, can perceive, make decisions, and control their surroundings, thus achieving partial or fully autonomous driving capabilities. However, in complex real-world traffic environments, intelligent vehicles may encounter various abnormal scenarios, which are often the main causes of traffic accidents. Therefore, effectively extracting and identifying these abnormal scenarios is crucial for improving the safety and reliability of intelligent vehicles.

[0003] Existing technologies, such as the invention patent with announcement number CN113992533B, are a method for detecting and identifying anomalies in vehicle-mounted CAN bus data. It adopts a modified self-attention mechanism integrated with a multi-layer GRU network model architecture, utilizes the characteristics of the self-attention mechanism to enhance the temporal characteristics of the data, and uses the multi-layer GRU network to further extract multi-dimensional temporal data features, thereby improving the recognition accuracy of CAN bus data.

[0004] Existing technologies, such as the invention patent with publication number CN111448783B, are vehicle network anomaly detection systems and methods. The vehicle network anomaly detection system includes: a first communication unit that receives first unit data as communication unit data under a first protocol from a first network; an anomaly determination database containing information on anomaly determination rules; an anomaly determination unit that uses the anomaly determination rules to determine whether the first unit data is anomaly-free; a unit data conversion unit that extracts second unit data as communication unit data under a second protocol from the first unit data determined by the anomaly determination unit to be non-anomaly-free; and a second communication unit that sends the extracted second unit data to a second network. The first unit data includes transmission source information indicating a first device as the transmission source, and the second unit data includes a data identifier. If the anomaly determination unit determines that the first unit data is anomaly-free by comparing the combination of the transmission source information and the data identifier with the anomaly determination rules, it performs a prescribed anomaly response process.

[0005] As can be seen from the above technical solutions, in the field of anomaly scene detection, the existing technology relies on a relatively single data collection source and the data processing method requires a large amount of computing resources. However, in practical applications, different intelligent vehicles may face different anomaly scenarios, and the risks of anomaly scenarios for different intelligent vehicles are not the same. If only data from a single source is analyzed, the analysis results may not be universal. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for extracting abnormal driving scenarios from intelligent vehicle driving data based on artificial intelligence. To achieve the above objectives, this invention employs the following technical solution: A method and system for extracting abnormal driving scenarios from intelligent vehicle driving data based on artificial intelligence, comprising:

[0007] Basic information of each intelligent vehicle is retrieved from multiple data transmission terminals, analyzed and processed to obtain the abnormal risk assessment value of each intelligent vehicle, and the driving data extraction ratio of each intelligent vehicle is matched based on the abnormal risk assessment value of each intelligent vehicle.

[0008] Based on the driving data extraction ratio of each intelligent vehicle, driving data packages of each intelligent vehicle are obtained. At the same time, the time points marked by each driving anomaly are obtained. Comprehensive scene data of each driving anomaly time point is statistically analyzed from the driving data package, and after classification data processing, it is imported into the integrated AI cloud computing center.

[0009] The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs clustering processing to generate abnormal scene data of each smart car and stores it.

[0010] As a preferred technical solution, the basic information of each smart car is retrieved from multiple data transmission terminals. The specific process is as follows:

[0011] The integrated AI cloud computing center initiates connection requests to each data transmission terminal. After receiving the connection request, each data transmission terminal packages the basic information of each smart vehicle and sends it to the central server of the integrated AI cloud computing center.

[0012] The basic information of each smart car includes the accumulated mileage, accumulated driving time and accumulated number of driving failures of each smart car.

[0013] As a preferred technical solution, the analysis and processing to obtain the abnormal risk assessment value of each smart car specifically includes:

[0014] The basic information of each smart car is analyzed and processed to obtain a basic information average value set and a basic information median value set.

[0015] The basic information average value set includes the cumulative average driving mileage, the cumulative average driving time and the cumulative average number of driving failures.

[0016] The basic information median value set includes the median value of the cumulative driving mileage, the median value of the cumulative driving time, and the median value of the cumulative number of driving failures.

[0017] The basic information average value set and the basic information median value set are comprehensively analyzed and processed to obtain the basic information median adjusted average value set, including the cumulative median adjusted average mileage, the cumulative median adjusted average driving time and the cumulative median adjusted average number of failures.

[0018] The basic information of each smart car is compared and analyzed with the set of median-adjusted average values ​​of the basic information to obtain an abnormal risk assessment value of each smart car. The abnormal risk assessment value of each smart car is used to characterize the risk of abnormality occurring during the driving process of each smart car.

[0019] As a preferred technical solution, the driving data extraction ratio of each smart car is obtained by matching the abnormal risk assessment value of each smart car, specifically including:

[0020] Based on the abnormal risk assessment value of each smart car, a mapping and matching is performed with the driving data extraction ratio corresponding to each abnormal risk assessment value interval in the built-in database of the integrated AI cloud computing center to obtain the driving data extraction ratio of each smart car.

[0021] As a preferred technical solution, the method of obtaining each driving abnormality marking time point and collecting comprehensive scene data of each driving abnormality marking time point from the driving data packet specifically includes:

[0022] Identify the time points marked as driving abnormalities in the driving data packets of each smart car, and count the longitudinal abnormal time point data and horizontal abnormal time period data of each driving abnormality marked time point, and record them as comprehensive scene data of each driving abnormality marked time point.

[0023] The comprehensive scenario data at each driving abnormality marking time point includes environmental perception data and control system data at each driving abnormality marking time point.

[0024] The environmental perception data includes image data, point cloud data and radar data at each driving abnormality marking time point.

[0025] The control system data includes engine response time, braking deceleration, braking distance, suspension compression and suspension rebound speed at each driving abnormality mark time point.

[0026] As a preferred technical solution, the classified data is processed and then imported into the integrated AI cloud computing center, specifically including:

[0027] The image data, point cloud data, and radar data at each driving anomaly marking time point are imported into the local image feature extraction module, point cloud geometric feature extraction module, and radar feature extraction module of the AI ​​cloud computing center respectively, and corresponding data processing is performed to obtain the image feature data, point cloud geometric feature data, and radar feature data at each driving anomaly marking time point.

[0028] The control system data at each driving abnormality marking time point is compared and analyzed with the control system data verification set pre-stored in the local database to obtain the control system state characteristic value at each driving abnormality marking time point. The control system state characteristic value at each driving abnormality marking time point is used to confirm whether the control system at each driving abnormality marking time point is normal. If the control system state characteristic value at a certain driving abnormality marking time point is greater than or equal to a preset control system state abnormality threshold, then the control system at the driving abnormality marking time point is determined to be abnormal. If the control system state characteristic value at a certain driving abnormality marking time point is less than the control system state abnormality threshold, then the control system at the driving abnormality marking time point is determined to be normal.

[0029] The statistical control system marks the time point of each driving anomaly during the abnormality, records it as the time point marked by each abnormal control system, and stores the image feature data, point cloud geometric feature data and radar feature data of each abnormal control system marking time point in the database under the abnormal control system label.

[0030] The control system data verification set includes engine response time threshold, braking deceleration threshold, braking distance threshold, suspension compression threshold, and suspension rebound speed threshold.

[0031] The image feature data, point cloud geometric feature data, and radar feature data of each driving abnormality marking time point when the control system is normal are collectively recorded as the processed comprehensive scene data of each driving abnormality marking time point.

[0032] As a preferred technical solution, the integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs clustering processing to generate abnormal scene data of each smart car and stores it. The specific processing conditions are:

[0033] The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs K-means clustering processing and analysis to obtain multiple abnormal scene feature clusters at each driving abnormality marking time point, and obtains the data features of each abnormal cluster at each driving abnormality marking time point through iterative updating. The data features of each abnormal cluster at each driving abnormality marking time point are matched with the corresponding abnormal scene labels pre-stored in the integrated AI cloud computing center to obtain the abnormal scene labels of each driving abnormality marking time point.

[0034] Based on the abnormal scenario labels at each driving abnormality marker time point, the processed comprehensive scenario data at each driving abnormality marker time point is stored in the database under the corresponding abnormal scenario label.

[0035] As a preferred technical solution, it also includes detecting the performance parameters of the transmission channel from each data transmission terminal to the integrated AI cloud computing center, processing and analyzing to generate data quality repair factors and performing data quality repair. Specifically, it includes:

[0036] The transmission channel performance parameters include the real-time signal strength, interface rate, real-time delay and real-time signal-to-noise ratio of each data transmission terminal.

[0037] The transmission channel performance verification parameters are retrieved from the built-in database of the integrated AI cloud computing center, including real-time signal strength verification value, interface rate verification value, real-time delay verification value and real-time signal-to-noise ratio verification value.

[0038] A comprehensive comparison and analysis is performed on the transmission channel performance parameters and the transmission channel performance verification parameters to obtain the data quality impact value of each transmission channel, and the data quality impact value of each transmission channel is mapped and matched with the data quality repair factor corresponding to each data quality impact value interval pre-stored in the built-in database of the integrated AI cloud computing center to obtain the data quality repair factor of each transmission channel. The data quality impact value of each transmission channel is used to characterize the impact of the performance of each transmission channel on the data quality, and data quality repair is performed based on the data transmission repair factor.

[0039] As a preferred technical solution, the analysis and processing obtains the abnormal risk assessment value of each smart car, and the specific processing method is:

[0040] ;

[0041] in, For the Anomaly risk assessment value for an intelligent vehicle For the The cumulative mileage of a smart car For the The cumulative driving time of each smart car For the The cumulative number of malfunctions during the operation of an intelligent vehicle. The median adjusted average mileage for driving is The median adjusted average duration of driving is calculated. The median number of failures is the cumulative number of failures during driving. The weighting factor is the cumulative mileage. As a weighting factor for cumulative driving time, The weighting factor is the average number of cumulative failures during driving. Number the smart car. , The `softplus` function is a built-in function in Python, representing the total number of intelligent vehicles. .

[0042] An AI-based intelligent vehicle driving data anomaly scene extraction system includes:

[0043] The abnormal risk assessment module is used to retrieve the basic information of each smart car from multiple data transmission terminals, analyze and process it to obtain the abnormal risk assessment value of each smart car, and based on the abnormal risk assessment value of each smart car, match and obtain the driving data extraction ratio of each smart car.

[0044] The data extraction module is used to obtain the driving data packet of each smart car based on the driving data extraction ratio of each smart car, and at the same time obtain the time point of each driving abnormality mark, collect the comprehensive scene data of each driving abnormality mark time point from the driving data packet, and import it into the integrated AI cloud computing center after classified data processing.

[0045] The scene extraction module is used to enable the integrated AI cloud computing center to receive the processed comprehensive scene data of each driving abnormality marking time point, perform clustering processing and analysis to generate abnormal scene data of each smart car and store it.

[0046] Compared with the prior art, the embodiments of the present invention have at least the following beneficial effects:

[0047] (1) The present invention provides an artificial intelligence-based method and system for extracting abnormal scenarios from smart car driving data. By retrieving basic information of smart cars from multiple data transmission terminals, comprehensive data is obtained, and the basic information of smart cars is analyzed and processed to obtain abnormal risk assessment values. The driving data extraction ratio is dynamically adjusted, and data collection resources are effectively allocated to improve efficiency.

[0048] (2) This invention comprehensively analyzes the driving environment by statistically integrating scene data, including environmental perception data and control system data. It uses an image feature extraction module, a point cloud geometric feature extraction module, and a radar data processing module to classify and process the data, deeply mining the feature information within the data. Through clustering analysis, different abnormal scenes can be identified, and each scene can be labeled for easy subsequent data storage and analysis.

[0049] (3) The present invention detects the performance parameters of the data transmission channel and generates a data quality repair factor to ensure the integrity and accuracy of data during transmission, reduce maintenance costs, and improve data processing efficiency.

[0050] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the method of the present invention.

[0052] Figure 2Schematic diagram of the system module of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inner", "around", etc., which indicate orientation or positional relationship, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting this invention.

[0055] See also Figure 1 As shown, this embodiment of the invention provides a method for extracting abnormal scenarios from intelligent vehicle driving data based on artificial intelligence, including:

[0056] Basic information of each intelligent vehicle is retrieved from multiple data transmission terminals, analyzed and processed to obtain the abnormal risk assessment value of each intelligent vehicle, and the driving data extraction ratio of each intelligent vehicle is matched based on the abnormal risk assessment value of each intelligent vehicle.

[0057] The specific process of retrieving basic information of each intelligent vehicle from multiple data transmission terminals is as follows:

[0058] The integrated AI cloud computing center initiates connection requests to each data transmission terminal. After receiving the connection request, each data transmission terminal packages the basic information of each smart vehicle and sends it to the central server of the integrated AI cloud computing center.

[0059] The integrated AI cloud computing center mentioned above is an integrated cloud processor used in this embodiment of the invention for receiving, sending, calculating, retrieving, and storing data.

[0060] In this embodiment of the invention, each data transmission terminal is a data transmission port pre-configured for each intelligent vehicle, used to send basic information and driving data packets of each intelligent vehicle.

[0061] The basic information of each intelligent vehicle includes the cumulative mileage, cumulative driving time, and cumulative number of driving failures, which are obtained by retrieving the driving logs of each intelligent vehicle.

[0062] Accumulated mileage refers to the total distance a smart car has traveled since it left the factory until the current time. Accumulated mileage can be used to assess the intensity of vehicle usage.

[0063] Accumulated driving time refers to the total driving time of a smart car from the time it leaves the factory until the current time. Accumulated time allows us to understand the vehicle's average operating time and idle time, thereby assessing its operational efficiency. Prolonged operation can lead to the aging of vehicle components; accumulated time helps evaluate the overall condition of the vehicle.

[0064] The cumulative number of driving failures refers to the total number of failures that have occurred in a smart car since it was manufactured. The number of failures can be used to assess the reliability of the vehicle.

[0065] The analysis and processing to obtain the abnormal risk assessment value of each smart car specifically includes:

[0066] The basic information of each smart car is analyzed and processed to obtain a basic information average value set and a basic information median value set.

[0067] The basic information average value set includes the cumulative average driving mileage, the cumulative average driving time and the cumulative average number of driving failures.

[0068] The basic information median value set includes the median value of the cumulative driving mileage, the median value of the cumulative driving time, and the median value of the cumulative number of driving failures.

[0069] The basic information average value set and the basic information median value set are comprehensively analyzed and processed to obtain the basic information median adjusted average value set, including the cumulative median adjusted average mileage, the cumulative median adjusted average driving time and the cumulative median adjusted average number of failures.

[0070] The basic information of each smart car is compared and analyzed with the set of median-adjusted average values ​​of the basic information to obtain an abnormal risk assessment value of each smart car. The abnormal risk assessment value of each smart car is used to characterize the risk of abnormality occurring during the driving process of each smart car.

[0071] The analysis and processing yields the anomaly risk assessment value for each intelligent vehicle. The specific processing method is as follows:

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] in, For the Anomaly risk assessment value for an intelligent vehicle For the The cumulative mileage of a smart car For the The cumulative driving time of each smart car For the The cumulative number of malfunctions during the operation of an intelligent vehicle. is the average accumulated mileage, is the average cumulative driving time, is the average number of cumulative failures during driving, is the median value of the accumulated mileage. is the median value of the cumulative driving time, is the median value of the cumulative number of driving failures, The median adjusted average mileage for driving is The median adjusted average duration of driving is calculated. The median number of failures is the cumulative number of failures during driving. The weighting factor is the cumulative mileage. As a weighting factor for cumulative driving time, The weighting factor is the average number of cumulative failures during driving. Adjust the scale factor for the mean, Adjust the scale factor for the median value, Number the smart car. , The `softplus` function is a built-in function in Python, representing the total number of intelligent vehicles. .

[0077] It should be noted that the mean value adjustment coefficient and the median value adjustment coefficient are both between 0 and 1 and meet the following requirements: The average value adjustment coefficient is an influencing factor of the median-adjusted average value of each basic information preset in the integrated AI cloud computing center, which represents a numerical value indicating the degree of influence of the average value on the median-adjusted average value of each basic information. The median value adjustment coefficient is an influencing factor of the median-adjusted average value of each basic information preset in the integrated AI cloud computing center, which represents a numerical value indicating the degree of influence of the median value on the median-adjusted average value of each basic information. When used, the average value adjustment coefficient and the median value adjustment coefficient can be directly obtained from the integrated AI cloud computing center, and the corresponding relationship is a pre-set mapping relationship. For example, after calculating the average value and median value of the basic information of each smart car involved in the embodiment of this invention, including the accumulated mileage, accumulated driving time and accumulated number of driving failures of each smart car, the mapping set is input for mapping matching to obtain the average value adjustment coefficient and the median value adjustment coefficient involved in the embodiment of this invention, wherein the mapping relationship is one-to-one correspondence.

[0078] It should be noted that the weight factor of the accumulated mileage, the accumulated driving time and the accumulated average number of failures are all within the range of 0 to 1 and meet the following requirements: The weight factor of the accumulated mileage is an influencing factor of the abnormal risk assessment value of each smart car preset in the integrated AI cloud computing center, which indicates the degree of influence of the accumulated mileage on the abnormal risk assessment value of each smart car; the weight factor of the accumulated driving time is an influencing factor of the abnormal risk assessment value of each smart car preset in the integrated AI cloud computing center, which indicates the degree of influence of the accumulated driving time on the abnormal risk assessment value of each smart car; the weight factor of the accumulated average number of failures is an influencing factor of the abnormal risk assessment value of each smart car preset in the integrated AI cloud computing center, which indicates the influence of the accumulated average number of failures on the abnormal risk assessment value of each smart car degree; when in use, the accumulated mileage weight factor, accumulated driving time weight factor and accumulated average number of failures weight factor can be directly obtained from the integrated AI cloud computing center, and the acquisition method is a preset mapping relationship. For example: the basic information of each smart car involved in the embodiment of this invention, including the accumulated mileage, accumulated driving time and accumulated number of failures of each smart car, is input into the preset mapping set in the database for mapping and matching to obtain the accumulated mileage weight factor, accumulated driving time weight factor and accumulated average number of failures weight factor involved in the embodiment of this invention, wherein the mapping relationship is one-to-one correspondence.

[0079] It should also be noted that the basic information of each smart car includes the cumulative mileage, cumulative driving time and cumulative number of driving failures of each smart car. There is a certain correlation between these parameters and they are inseparable from each other. Generally, the cumulative mileage and cumulative driving time are directly related. The longer the mileage of a vehicle, the more time it usually takes. The relationship between these two parameters is also affected by the average driving speed of the vehicle. The faster the speed, the more mileage is accumulated in the same time; the slower the speed, the less mileage is accumulated in the same time. As the cumulative mileage increases, vehicle components may be more prone to failure due to wear, so these two parameters show a positive correlation. Long-term operation may cause vehicle components to overheat or fatigue, thereby increasing the number of failures. In the absence of other factors, the cumulative mileage and cumulative driving time are usually positively correlated with the cumulative number of driving failures, that is, as the mileage or time increases, the number of failures may also increase.

[0080] The driving data extraction ratio of each smart car is obtained by matching the abnormal risk assessment value of each smart car, specifically including:

[0081] Based on the abnormal risk assessment value of each smart car, a mapping and matching is performed with the driving data extraction ratio corresponding to each abnormal risk assessment value interval in the built-in database of the integrated AI cloud computing center to obtain the driving data extraction ratio of each smart car.

[0082] The driving data extraction ratio of each intelligent vehicle is used to determine the amount of driving data to be extracted from each intelligent vehicle. For example, if the driving data extraction ratio corresponding to the abnormal risk assessment value of a certain intelligent vehicle is 20%, then 20% of the driving data of that intelligent vehicle will be selected.

[0083] By matching the abnormal risk assessment values ​​of each intelligent vehicle with the corresponding driving data extraction ratio, different data extraction strategies are formulated for different vehicle risk levels. High-risk vehicles will be allocated a higher data extraction ratio to ensure that more resources are invested in high-risk intelligent vehicles, while reducing unnecessary data collection and lowering data storage and processing costs.

[0084] It should be noted that the embodiment of the present invention also includes detecting the transmission channel performance parameters from each data transmission terminal to the integrated AI cloud computing center, processing and analyzing to generate a data quality repair factor, specifically including:

[0085] The transmission channel performance parameters include real-time signal strength, interface rate, real-time delay and real-time signal-to-noise ratio.

[0086] Real-time signal strength refers to the signal level in a transmission channel, usually expressed in decibels (dB). Signal strength directly affects transmission distance and interference resistance. Higher signal strength allows for longer transmission distances and stronger resistance to external interference. Insufficient signal strength may lead to data transmission errors or interruptions.

[0087] Interface rate refers to the transmission rate of a data transmission channel, usually expressed in bits per second (bps) or megabits per second (Mbps). The interface rate determines the speed of data transmission; the higher the rate, the greater the amount of data that can be transmitted per unit of time.

[0088] Real-time latency refers to the time required for data to travel from the sender to the receiver, typically measured in milliseconds (ms). Lower latency results in better real-time performance. High latency can degrade the real-time performance of data transmission, impacting application performance.

[0089] The real-time signal-to-noise ratio (SNR) is the ratio of signal power to noise power, usually expressed in decibels (dB). SNR is an important parameter for measuring signal quality. A higher SNR means a greater proportion of effective information in the signal relative to noise, resulting in higher reliability of data transmission. A low SNR indicates a high noise level, which may lead to data transmission errors or packet loss.

[0090] The transmission channel performance verification parameters are retrieved from the built-in database of the integrated AI cloud computing center, including real-time signal strength verification value, interface rate verification value, real-time delay verification value and real-time signal-to-noise ratio verification value.

[0091] The performance parameters of each transmission channel are comprehensively compared and analyzed with the performance verification parameters of the transmission channel to obtain the data quality impact value of each transmission channel, and the data quality repair factors corresponding to each data quality impact value interval pre-stored in the built-in database of the integrated AI cloud computing center are mapped and matched to obtain the data quality repair factor of each transmission channel. The larger the data quality impact value, the larger the data quality repair factor. The data quality repair factor of each transmission channel is used to repair the data loss incurred during the transmission process from each data transmission terminal to the integrated AI cloud computing center.

[0092] Obtaining the data quality impact value of each transmission channel specifically includes:

[0093] ;

[0094] in, For the The data quality impact value of each transmission channel, For the Real-time signal strength of each transmission channel, For the The interface rate of each transmission channel, For the Real-time delay of each transmission channel, For the Real-time signal-to-noise ratio of each transmission channel, is the real-time signal strength verification value, is the interface rate check value, is the real-time delay check value, is the real-time signal-to-noise ratio verification value, is the transmission channel number, , is the total number of transmission channels, is the real-time delay weight factor, is the real-time signal strength weight factor, is the interface rate weight factor, is the real-time signal-to-noise ratio weighting factor.

[0095] It should be noted that the real-time delay weight factor, real-time signal strength weight factor, interface rate weight factor and real-time signal-to-noise ratio weight factor all have a value range between 0 and 1 and meet the following requirements: The real-time delay weight factor is the impact factor corresponding to the data quality impact value of each transmission channel preset in the integrated AI cloud computing center, which indicates the numerical value of the influence degree of real-time delay on the data quality impact value of each transmission channel. The real-time signal strength weight factor is the impact factor corresponding to the data quality impact value of each transmission channel preset in the integrated AI cloud computing center, which indicates the numerical value of the influence degree of real-time signal strength on the data quality impact value of each transmission channel. The interface rate weight factor is the impact factor corresponding to the data quality impact value of each transmission channel preset in the integrated AI cloud computing center, which indicates the numerical value of the influence degree of interface rate on the data quality impact value of each transmission channel. The real-time signal-to-noise ratio weight factor is the impact factor corresponding to the data quality impact value of each transmission channel preset in the integrated AI cloud computing center. The impact factor represents a numerical value indicating the degree of influence of the real-time signal-to-noise ratio on the data quality impact value of each transmission channel. When used, the real-time delay weight factor, real-time signal strength weight factor, interface rate weight factor, and real-time signal-to-noise ratio weight factor can be directly obtained from the integrated AI cloud computing center. The acquisition method is a preset mapping relationship. For example, the data quality impact value of each transmission channel involved in the embodiment of this invention, such as the transmission channel performance parameters including real-time signal strength, interface rate, real-time delay, and real-time signal-to-noise ratio, is input into a preset mapping set in the database for mapping and matching to obtain the real-time delay weight factor, real-time signal strength weight factor, interface rate weight factor, and real-time signal-to-noise ratio weight factor involved in the embodiment of this invention, wherein the mapping relationship is one-to-one correspondence.

[0096] It should be noted that the transmission channel performance parameters include real-time signal strength, interface rate, real-time latency, and real-time signal-to-noise ratio (SNR). These parameters are closely related and inseparable. Real-time signal strength is a component of the real-time SNR. Higher signal strength generally results in a higher SNR because the ratio of signal power to noise power increases. Real-time signal strength can affect the interface rate. If the signal strength is sufficiently high, the transmission channel can support higher data rates. Conversely, if the signal strength decreases, the transmission rate may decrease because the system may need to reduce the rate to ensure data reliability. The real-time SNR is a key factor affecting the interface rate. A high SNR supports higher data rates because the signal contains more valid information and has a lower error rate. A low SNR may lead to data transmission errors, requiring data retransmission, which reduces the actual effective transmission rate. Real-time signal strength and real-time SNR are interrelated and together determine signal clarity and transmission quality. The interface rate is affected by real-time signal strength and real-time SNR because these parameters determine the capacity and reliability of the transmission channel. Real-time latency can be affected by all other parameters because it reflects the time required for data to travel from the sender to the receiver, which is influenced by signal quality, transmission rate, and network conditions.

[0097] The loss of the repaired data during transmission from various data transmission terminals to the integrated AI cloud computing center is specifically described as follows:

[0098] Based on the data quality repair factor of each transmission channel, the data quality repair parameter set corresponding to the data quality repair factor of each transmission channel pre-stored in the integrated AI cloud computing center is matched to obtain the data quality repair parameter set of each transmission channel, and the data quality repair parameter set includes a data cleaning ratio adjustment value, a data filling ratio adjustment value and a data deduplication ratio adjustment value.

[0099] The larger the data quality repair factor, the larger the set of data quality repair parameters, including the data cleaning ratio adjustment value, the data filling ratio adjustment value, and the data deduplication ratio adjustment value.

[0100] Based on the data quality repair parameter set of each transmission channel, corresponding data quality repair is performed on the basic information and driving data packets of each smart car after transmission. For example, the data cleaning ratio adjustment value of a certain transmission channel is 30%, the data filling ratio adjustment value is 20%, and the data deduplication ratio adjustment value is 15%. Then, the basic information and driving data packets of each smart car transmitted by this transmission channel are cleaned, filled, and deduplicated in corresponding proportions.

[0101] Based on the driving data extraction ratio of each intelligent vehicle, driving data packages of each intelligent vehicle are obtained. At the same time, the time points marked by each driving anomaly are obtained. Comprehensive scene data of each driving anomaly time point is statistically analyzed from the driving data package, and after classification data processing, it is imported into the integrated AI cloud computing center.

[0102] The process of obtaining the time points of each driving anomaly marker and statistically analyzing the comprehensive scene data of each driving anomaly marker time point from the driving data packet specifically includes:

[0103] Identify the time points marked as driving abnormalities in the driving data packets of each smart car, integrate the time points marked as driving abnormalities in the driving data packets of each smart car into each driving abnormality marked time point, count the longitudinal abnormal time point data and the horizontal abnormal time period data of each driving abnormality marked time point, and comprehensively record them as the comprehensive scene data of each driving abnormality marked time point.

[0104] The longitudinal abnormal time point data of each driving abnormality marker time point is used to locate the position of each driving abnormality marker time point in the driving data packet.

[0105] The horizontal abnormal time period data refers to locating the position of each driving abnormality mark time point in the driving data packet, and intercepting the data contained in the time period before and after each driving abnormality mark time point with the preset time period length of each smart car, and recording this section of data as the comprehensive scene data of each driving abnormality mark time point.

[0106] The preset time period length for each intelligent vehicle specifically includes:

[0107] Based on the abnormal risk assessment value of each smart car, the preset time period length corresponding to each abnormal risk assessment value interval in the database is mapped and matched to obtain the preset time period length of each smart car. The higher the abnormal risk assessment value of a smart car, the longer the corresponding preset time period length.

[0108] The comprehensive scenario data at each driving abnormality marking time point includes environmental perception data and control system data at each driving abnormality marking time point.

[0109] The environmental perception data includes image data, point cloud data and radar data at each driving abnormality marking time point.

[0110] The image data includes front, rear, side, and overhead and upward view images of each smart vehicle acquired through dashcams and external integrated cameras.

[0111] The point cloud data includes the 3D edge coordinates of each object in the scene, laser intensity, and the reflectivity of the laser pulse, acquired through a LiDAR sensor. The 3D edge coordinates refer to the spatial position of each point on the edge of an object in the scene, typically represented by (X, Y, Z). This is used to determine the exact location of an object in 3D space (i.e., in abnormal scenes). The 3D shape and structure of the object can be reconstructed using the edge coordinates. Laser intensity refers to the intensity value returned after the laser pulse emitted by the LiDAR is reflected by an object. Different materials reflect different laser intensities, used to identify the object's material. Laser intensity can also reflect the smoothness or roughness of the object's surface. The reflectivity of the laser pulse refers to the proportion of light reflected back to the sensor after the laser pulse interacts with the object's surface. Different objects have different reflectivities, used for object classification and obstacle detection.

[0112] The radar data includes range information, velocity information, and radar cross-section (RCS). Range information is calculated by transmitting pulses and measuring the time it takes for them to reflect back, and includes the straight-line distances between each intelligent vehicle and other objects in the scene. Velocity information is measured using a laser velocimeter, including the absolute and relative velocities of each intelligent vehicle and other objects in the scene. Radar cross-section is the ability of an object to reflect radar signals, used to estimate the size and shape of objects in the scene.

[0113] The control system data includes engine response time, braking deceleration, braking distance, suspension compression and suspension rebound speed at each driving abnormality marking time point, which is obtained through the on-board diagnostic system.

[0114] Engine response time is the time it takes from the issuance of an acceleration request (for example, by pressing the accelerator pedal) to the start of the increase in engine output power. Engine response time is an important indicator of engine dynamic performance.

[0115] Braking deceleration is the rate at which a vehicle loses speed during braking, typically measured in meters per second squared (m / s²). Braking deceleration reflects the effectiveness of the braking system. A higher braking deceleration can slow or stop a vehicle in a shorter distance, which is crucial for safe stopping in emergency situations.

[0116] Braking distance refers to the distance a vehicle travels from a given speed to a complete stop. It's a crucial indicator of vehicle safety. Short braking distances provide more time and space for reaction in emergency situations, thus reducing the likelihood of accidents.

[0117] Suspension compression refers to the degree of compression of a vehicle's suspension system under load (such as vehicle weight, acceleration, or braking). It primarily reflects the deformation of the suspension system. In this embodiment of the invention, the degree of suspension compression is measured by the compression length of the spring, in millimeters (mm). Suspension compression affects a vehicle's driving stability, comfort, and handling. Appropriate suspension compression ensures good contact between the wheels and the ground, improving grip.

[0118] Suspension rebound speed refers to how quickly the suspension system returns to its original position after a load is released. Suspension rebound speed affects a vehicle's dynamic behavior and ride comfort. Faster rebound reduces vehicle oscillation on bumpy roads, improving handling and stability.

[0119] The classified data is then processed and imported into the integrated AI cloud computing center, specifically including:

[0120] The image data, point cloud data, and radar data at each driving anomaly marking time point are imported into the local image feature extraction module, point cloud geometric feature extraction module, and radar data processing module of the AI ​​cloud computing center respectively, and corresponding data processing is performed to obtain the image feature data, point cloud geometric feature data, and radar feature data at each driving anomaly marking time point.

[0121] The obtaining of image feature data, point cloud geometric feature data, and radar feature data at each driving abnormality marking time point specifically includes:

[0122] After preprocessing the image data at each driving anomaly marking time point, the image feature extraction module uses the ORB feature extraction algorithm to encode the identified image features (such as traffic lights, pedestrians, vehicles, and road traffic signs) into feature vectors. The PCA feature selection algorithm is used to filter the feature vectors and summarize them into the image feature data at each driving anomaly marking time point.

[0123] After preprocessing the point cloud data of each driving abnormality marking time point, the point cloud geometric feature extraction module uses PCL to calculate the surface normal, curvature and other geometric features of the point cloud, and summarizes them into the point cloud geometric feature data of each driving abnormality marking time point.

[0124] After pre-processing the radar data at each driving anomaly marking time point, the radar data processing module uses the CFAR algorithm to identify moving objects in the radar data and perform path tracking. It extracts the motion data of each moving object, including the average speed, direction, and distance of movement, and summarizes it as radar feature data.

[0125] It should be noted that the ORB feature extraction algorithm (Oriented FAST and Rotated BRIEF) is an algorithm used for feature detection and description in computer vision. The ORB feature extraction algorithm includes FAST keypoint detection and BRIEF feature descriptors, specifically including:

[0126] FAST (Features from Accelerated Segment Test) is an algorithm for detecting keypoints (feature points) in an image. It determines whether a pixel is a keypoint by comparing it to its surrounding pixels. If a pixel is brighter or darker than a sufficient number of pixels in its neighborhood, it is marked as a keypoint. ORB uses the FAST algorithm to quickly detect keypoints in an image.

[0127] BRIEF (Binary Robust Independent Elementary Features) is a feature descriptor that generates a binary string (a sequence of 0s and 1s) by comparing the brightness of a pair of pixels around a keypoint. This string is the descriptor of the keypoint and is used in the subsequent matching process.

[0128] To make the BRIEF descriptor rotation-invariant, the ORB algorithm first computes the orientation of each keypoint. This is typically done by computing the histogram of gradient orientations within the keypoint's neighborhood and finding the dominant orientation. The pixel pairs used to compare the BRIEF descriptors are rotated according to this orientation, ensuring that the descriptor is insensitive to image rotation.

[0129] ORB achieves scale invariance by building a pyramid of images and applying FAST to detect key points at different scales. This allows key points to be detected at different scales, enabling the algorithm to handle image features of different sizes.

[0130] The PCA feature selection algorithm (Principal Component Analysis) is a statistical method that transforms a set of potentially correlated variables into a set of linearly uncorrelated variables through orthogonal transformations. These new variables are called principal components. PCA is mainly used for data dimensionality reduction, that is, reducing the dimensionality of data while preserving as much original information as possible. In feature selection, PCA can help identify and select the most important features. The basic steps of PCA include:

[0131] Step 1: Standardize the data. Since PCA is greatly affected by the data scale, it is usually necessary to standardize the data first.

[0132] Step 2: Calculate the covariance matrix. The covariance matrix describes the correlation between the features in the data. If the covariance of two features is large, it means that they are likely to change together.

[0133] Step 3: Calculate the eigenvalues ​​and eigenvectors of the covariance matrix. By calculating the eigenvalues ​​and eigenvectors of the covariance matrix, we can find the main directions of the data (i.e., principal components).

[0134] Step 4: Select the principal components. The size of the eigenvalue represents the variance of the data in the direction of the corresponding eigenvector. Select the eigenvectors corresponding to the largest eigenvalues; these eigenvectors are the principal components of the data.

[0135] Step 5: Form the projection matrix. Combine the selected eigenvectors into a new matrix called the projection matrix. This matrix will be used to project the original data into a lower-dimensional space.

[0136] Step 6: Transform the data. Use the projection matrix to transform the original data into the new space.

[0137] In feature selection, PCA can help identify which principal components contain most of the information in the data.

[0138] By observing the size of the eigenvalues ​​of each principal component for the data, we can determine which original features contribute the most to the variance of the data. If the first few principal components can explain most of the variance of the data, then we can use these principal components to replace the original features, thereby achieving data dimensionality reduction.

[0139] The PCL (Point Cloud Library) is an open source library for 2D / 3D image and point cloud processing.

[0140] The CFAR (Constant False Alarm Rate) algorithm is a detection technique used in radar and sonar signal processing. It adaptively adjusts the detection threshold to maintain a predetermined false alarm rate, even in the presence of changing ambient noise or interference. The CFAR algorithm is widely used for target detection in radar systems to ensure stable identification of true targets amidst complex background noise.

[0141] Image feature data is used to identify key visual elements in the driving environment and detect abnormal events (such as vehicle lane deviation, collision, etc.).

[0142] Point cloud geometric feature data is used to provide three-dimensional structural information about the surrounding environment. Geometric features such as surface normals and curvature are used to identify and classify obstacles on the road, such as pedestrians and vehicles. The surface normal is a vector representing the surface direction of each point in the point cloud. The curvature describes the degree of curvature of the point cloud surface and is used to identify the shape of objects.

[0143] Radar data obtains dynamic information about objects in the scene through moving object recognition and path tracking, including the average speed, direction, and distance of each object. This information is mainly used to analyze the vehicle's speed status and relative position, and to determine abnormal driving conditions such as whether there is a rear-end collision and whether a safe distance is maintained.

[0144] The control system data at each driving abnormality marking time point is compared and analyzed with a control system data verification set pre-stored in a local database. The control system data verification set includes an engine response time threshold, a braking deceleration threshold, a braking distance threshold, a suspension compression threshold, and a suspension rebound speed threshold. The control system state characteristic value at each driving abnormality marking time point is obtained. The control system state characteristic value at each driving abnormality marking time point is used to confirm whether the control system at each driving abnormality marking time point is normal. The specific processing method is as follows:

[0145] ;

[0146] in, For the The control system state characteristic value at the driving abnormality marking time point, For the The engine response time at each driving abnormality mark time point, For the The braking deceleration at the time point of the driving abnormality mark, For the The braking distance at the time point of driving abnormality mark, For the The suspension compression amount at the time point of the driving abnormality mark, For the The suspension rebound speed at the driving abnormality mark time point, is the engine response time threshold, is the braking deceleration threshold, is the braking distance threshold, is the suspension compression threshold, is the suspension rebound speed threshold, Mark the time point number for driving abnormality, , The total number of time points marked for driving abnormality, is the transmission channel number, , is the total number of transmission channels, is the engine response time weight factor, is the braking distance weight factor, is the suspension compression weight factor, is the braking deceleration weight factor, is the suspension rebound velocity weight factor.

[0147] It should be noted that the engine response time weight factor, braking distance weight factor, suspension compression weight factor, braking deceleration weight factor and suspension rebound speed weight factor all have a value range between 0 and 1 and meet the requirements. The engine response time weight factor is an influence factor corresponding to the control system state characteristic value of each driving abnormality mark time point preset in the integrated AI cloud computing center, which represents the numerical value of the influence degree of the engine response time on the control system state characteristic value of each driving abnormality mark time point. The braking distance weight factor is an influence factor corresponding to the control system state characteristic value of each driving abnormality mark time point preset in the integrated AI cloud computing center, which represents the numerical value of the influence degree of the braking distance on the control system state characteristic value of each driving abnormality mark time point. The suspension compression weight factor is an influence factor corresponding to the control system state characteristic value of each driving abnormality mark time point preset in the integrated AI cloud computing center, which represents the numerical value of the influence degree of the suspension compression on the control system state characteristic value of each driving abnormality mark time point. The braking deceleration weight factor is an influence factor corresponding to the control system state characteristic value of each driving abnormality mark time point preset in the integrated AI cloud computing center, which represents the control deceleration on the braking deceleration at each driving abnormality mark time point. The suspension rebound speed weighting factor is an influence factor corresponding to the control system state characteristic value at each driving abnormality marking time point preset in the integrated AI cloud computing center. It represents the numerical value of the influence of the suspension rebound speed on the control system state characteristic value at each driving abnormality marking time point. When used, the engine response time weighting factor, braking distance weighting factor, suspension compression weighting factor, braking deceleration weighting factor, and suspension rebound speed weighting factor can be directly obtained from the integrated AI cloud computing center. The acquisition method is a pre-set mapping relationship. For example, the engine response time, braking distance, suspension compression, braking deceleration, and suspension rebound speed involved in the embodiment of the present invention are input into a preset mapping set in the database for mapping and matching to obtain the engine response time weighting factor, braking distance weighting factor, suspension compression weighting factor, braking deceleration weighting factor, and suspension rebound speed weighting factor involved in the embodiment of the present invention. The mapping relationship is one-to-one correspondence.

[0148] It should be noted that the control system data includes engine response time, brake deceleration, braking distance, suspension compression, and suspension rebound speed at each driving anomaly marker time point. These parameters are closely related. Engine response time affects vehicle acceleration performance, while brake deceleration affects deceleration performance. In maneuvers such as emergency avoidance or rapid lane changes, the vehicle may need to decelerate rapidly and then accelerate immediately. Therefore, these two parameters jointly determine the vehicle's overall dynamic responsiveness. Braking deceleration directly determines braking distance. The greater the braking deceleration, the shorter the braking distance required from the same initial speed. These two parameters are core indicators for braking performance evaluation and are directly related. During braking, the suspension system may compress due to the forward shift of the vehicle's center of gravity, which affects tire contact with the road and, in turn, braking distance. Excessive suspension compression may increase braking distance. The suspension rebound speed affects how quickly the suspension recovers from a compressed state, which is related to vehicle stability and comfort on continuously bumpy roads. Greater suspension compression may require a faster rebound speed to maintain vehicle dynamic balance. The performance of the suspension system not only affects the comfort of the vehicle, but also affects the vehicle's posture control during braking. If the suspension system cannot effectively control the vehicle's posture, it may increase the braking distance and reduce braking efficiency during braking.

[0149] If the control system state characteristic value at a certain driving abnormality marking time point is greater than or equal to the control system state abnormality threshold, the control system at the driving abnormality marking time point is judged to be abnormal; if the control system state characteristic value at a certain driving abnormality marking time point is less than the abnormality threshold, the control system at the driving abnormality marking time point is judged to be normal.

[0150] The time points of each driving abnormality marking of the abnormal control system are counted and recorded as the time points of each abnormal control system marking, and the image feature data, point cloud geometric feature data and radar feature data of each abnormal control system marking time point are stored in the database under the abnormal control system label.

[0151] It should be noted that abnormal scenarios caused by abnormalities in the control system itself are considered internal factors and should be stored separately. Furthermore, combining the data from each driving anomaly marker time point in the abnormal control system with other data for deep data cluster analysis may lead to overfitting during the analysis, affecting the results.

[0152] The abnormal state threshold is processed as follows: in a specific embodiment, the abnormal state threshold is set directly in the database during the development of the artificial intelligence-based intelligent vehicle driving data abnormal scene extraction system involved in the embodiment of the present invention. There are many methods for setting the abnormal state threshold, such as obtaining it through statistical analysis, including counting the control system state characteristic values ​​under the historical normal control system state (for example, the engine response time is less than or equal to 0.5 seconds), and averaging the control system state characteristic values ​​obtained multiple times to obtain the abnormal state threshold.

[0153] The image feature data, point cloud geometric feature data, and radar feature data of each driving anomaly marker time point in the normal control system are collectively recorded as the processed integrated scene data of each driving anomaly marker time point.

[0154] The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs clustering processing and analysis to generate abnormal scene data of each smart car and stores it.

[0155] The integrated AI cloud computing center receives the processed comprehensive scene data from each driving anomaly marker time point, performs clustering analysis to generate anomaly scene data for each intelligent vehicle, and stores the data. The specific processing conditions are as follows:

[0156] The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, standardizes the processed comprehensive scene data of each driving abnormality marking time point, performs K-means clustering analysis to obtain multiple abnormal scene feature clusters of each driving abnormality marking time point, obtains the data features of each abnormal cluster of each driving abnormality marking time point through iterative updating, and matches the data features of each abnormal cluster of each driving abnormality marking time point with the corresponding abnormal scene labels pre-stored in the integrated AI cloud computing center to obtain each abnormal scene label of each driving abnormality marking time point.

[0157] The standardization process for the processed integrated scene data at each driving anomaly marker time point specifically includes: calculating the mean and standard deviation of each feature; and standardizing the processed integrated scene data at each driving anomaly marker time point based on the mean and standard deviation using Z-score standardization, with the specific expression as follows:

[0158]

[0159] in, For the The standardized processing results of the features, For the Features, For the The feature mean, For the Standard deviation of each characteristic For feature numbering, , The total number of features is denoted as the dataset for each driving anomaly marker time point after processing.

[0160] K-means clustering is a typical unsupervised machine learning method used to divide a dataset of driving anomaly markers into clusters composed of similar elements. Data points in the dataset of driving anomaly markers are assigned to K clusters such that the sum of the squared distances between each data point and its assigned cluster center is minimized. The following are the basic steps and principles of K-means clustering:

[0161] Step 1: Randomly select K data points as the initial cluster centers (centroids).

[0162] Step 2: For each data point, calculate its distance to each cluster center and assign it to the cluster represented by the nearest cluster center.

[0163] Step 3: Recalculate the center of each cluster, usually by taking the mean of all points within the cluster as the new cluster center.

[0164] Step 4: Repeat steps 2 and 3 until the stopping condition is met. In this embodiment of the invention, the stopping condition is reaching a preset number of iterations.

[0165] In the embodiment of the present invention, Euclidean distance is used to measure the similarity between data points. In other embodiments of the invention, other distance measurement methods may also be used, such as Manhattan distance or cosine similarity.

[0166] The K-means algorithm aims to minimize the squared error within a cluster, and its objective function can be expressed as:

[0167] ;

[0168] in, Let K-means be the objective function. For the Clusters, for The first in data points, for The center, Cluster number, , The total number of clusters, for Data point numbering, , for The total number of data points. It should be noted that the K value is preset; in this embodiment of the invention, the K value is set to the number of abnormal scene labels.

[0169] Compare the features of each abnormal cluster data with the feature templates corresponding to each abnormal scene label pre-stored in the integrated AI cloud computing center. In this embodiment of the present invention, the Euclidean distance is used to calculate the similarity between the current abnormal cluster data feature and a feature in the feature template corresponding to the pre-stored abnormal scene label, specifically including:

[0170] Let the eigenvector expression of a certain abnormal cluster data feature be: .

[0171] Let the feature vector expression of a certain feature of the feature template corresponding to a certain abnormal scene label be: .

[0172] ;

[0173] in, This represents the similarity between the features of this abnormal cluster data and a certain feature in the feature template corresponding to a certain abnormal scene label. These are the indices of each vector in the eigenvector expression. , is the total number of vectors.

[0174] Based on the similarity calculation results, a machine learning classifier is used to determine whether each abnormal cluster data feature matches each abnormal scene label. If the classifier determines that there is a match, the corresponding abnormal scene label is assigned to the current abnormal cluster data feature.

[0175] The Euclidean distance refers to the real distance between two points in an m-dimensional space, or the natural length of a vector (ie, the distance from the point to the origin).

[0176] Based on the abnormal scenario labels at each driving abnormality marker time point, the processed comprehensive scenario data at each driving abnormality marker time point is stored in the database under the corresponding abnormal scenario label.

[0177] For example, if the abnormal scene labels of a certain driving abnormality marking time point include "rear-end collision", "sudden braking" and "collision", the processed comprehensive scene data of the driving abnormality marking time point will be stored in the database under the "rear-end collision", "sudden braking" and "collision" labels, and put into the integrated AI cloud computing center for learning and updating. When scene data similar to the driving abnormality marking time point appears again, the integrated AI cloud computing center can identify and mark the abnormal scene.

[0178] It should also be noted that the abnormal scene labels for a certain driving abnormality marker time point include "rear-end collision", "sudden braking" and "collision". For example, the "rear-end collision" abnormal scene label features include speed change features, following distance features, acceleration change features and the degree of deformation of the rear of the vehicle in front.

[0179] In this embodiment, if Figure 2 As shown, the present invention provides an artificial intelligence-based intelligent vehicle driving data abnormal scene extraction system, which specifically includes:

[0180] The abnormal risk assessment module is used to retrieve the basic information of each smart car from multiple data transmission terminals, analyze and process it to obtain the abnormal risk assessment value of each smart car, and based on the abnormal risk assessment value of each smart car, match and obtain the driving data extraction ratio of each smart car.

[0181] The data extraction module is used to obtain the driving data packet of each smart car based on the driving data extraction ratio of each smart car, and at the same time obtain the time point of each driving abnormality mark, collect the comprehensive scene data of each driving abnormality mark time point from the driving data packet, and import it into the integrated AI cloud computing center after classified data processing.

[0182] The scene extraction module is used to enable the integrated AI cloud computing center to receive the processed comprehensive scene data of each driving abnormality marking time point, perform clustering processing and analysis to generate abnormal scene data of each smart car and store it.

[0183] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0184] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the contents of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention. As long as they do not deviate from the structure of the present invention or exceed the scope defined by the present invention, they should fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based method for extracting abnormal scenes from smart car driving data, characterized in that: include: Retrieving basic information of each smart car from multiple data transmission terminals, analyzing and processing to obtain abnormal risk assessment values ​​of each smart car, and matching to obtain driving data extraction ratios of each smart car based on the abnormal risk assessment values ​​of each smart car; Based on the driving data extraction ratio of each smart car, the driving data packet of each smart car is obtained, and at the same time, the time point of each driving abnormality mark is obtained. The comprehensive scene data of each driving abnormality mark time point is counted from the driving data packet, and the classified data is processed and then imported into the integrated AI cloud computing center; The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs clustering processing to generate abnormal scene data of each smart car and stores it; The classified data is then processed and imported into the integrated AI cloud computing center, specifically including: The image data, point cloud data and radar data at each driving abnormality marking time point are respectively imported into the local image feature extraction module, point cloud geometric feature extraction module and radar feature extraction module of the AI ​​cloud computing center, and the corresponding data processing is performed to obtain the image feature data, point cloud geometric feature data and radar feature data at each driving abnormality marking time point; Compare and analyze the control system data at each driving abnormality marking time point with the control system data check set pre-stored in the local database to obtain the control system state characteristic value at each driving abnormality marking time point, wherein the control system state characteristic value at each driving abnormality marking time point is used to confirm whether the control system at each driving abnormality marking time point is normal. If the control system state characteristic value at a certain driving abnormality marking time point is greater than or equal to a preset control system state abnormality threshold, the control system at the driving abnormality marking time point is determined to be abnormal. If the control system state characteristic value at a certain driving abnormality marking time point is less than the control system state abnormality threshold, the control system at the driving abnormality marking time point is determined to be normal. The time points of each driving abnormality marking when the statistical control system is abnormal are recorded as the marking time points of each abnormal control system, and the image feature data, point cloud geometric feature data and radar feature data of each abnormal control system marking time point are stored in the database under the abnormal control system label; The control system data verification set includes an engine response time threshold, a brake deceleration threshold, a brake distance threshold, a suspension compression threshold, and a suspension rebound speed threshold; The image feature data, point cloud geometric feature data, and radar feature data at each driving abnormality marking time point when the control system is normal are collectively recorded as the processed comprehensive scene data at each driving abnormality marking time point.

2. The method for extracting abnormal scenes from smart car driving data based on artificial intelligence according to claim 1 is characterized by: The specific process of retrieving basic information of each smart car from multiple data transmission terminals is as follows: A connection request is initiated from the integrated AI cloud computing center to each data transmission terminal. After receiving the connection request, each data transmission terminal packages the basic information of each smart car and sends it to the central server of the integrated AI cloud computing center; The basic information of each smart car includes the accumulated mileage, accumulated driving time and accumulated number of driving failures of each smart car.

3. The method for extracting abnormal scenes from smart car driving data based on artificial intelligence according to claim 1 is characterized by: The analysis and processing obtains the abnormal risk assessment value of each smart car, specifically including: Analyze and process the basic information of each smart car to obtain a basic information average value set and a basic information median value set; The basic information average value set includes the cumulative average driving mileage, the cumulative average driving time and the cumulative average number of driving failures; The basic information median value set includes the median value of the cumulative driving mileage, the median value of the cumulative driving time, and the median value of the cumulative number of driving failures; Comprehensively analyzing and processing the basic information average value set and the basic information median value set to obtain the basic information median adjusted average value set, including the cumulative median adjusted average mileage, the cumulative median adjusted average driving time, and the cumulative median adjusted average number of failures; The basic information of each smart car is compared and analyzed with the set of adjusted median average values ​​of the basic information to obtain an abnormal risk assessment value of each smart car, and the abnormal risk assessment value of each smart car is used to characterize the risk of abnormality of each smart car during driving.

4. The method for extracting abnormal scenes from driving data of smart cars based on artificial intelligence according to claim 1 is characterized by: The driving data extraction ratio of each smart car is obtained by matching the abnormal risk assessment value of each smart car, specifically including: Based on the abnormal risk assessment value of each smart car, the driving data extraction ratio corresponding to each abnormal risk assessment value interval in the built-in database of the integrated AI cloud computing center is mapped and matched to obtain the driving data extraction ratio of each smart car.

5. The method for extracting abnormal scenes from smart car driving data based on artificial intelligence according to claim 1 is characterized by: The step of obtaining each driving abnormality marking time point and collecting comprehensive scene data of each driving abnormality marking time point from the driving data packet specifically includes: Identify the time points marked as driving abnormalities in the driving data packets of each smart car, and count the longitudinal abnormal time point data and the lateral abnormal time period data of each driving abnormality marked time point, and record them as the comprehensive scene data of each driving abnormality marked time point; The comprehensive scene data at each driving abnormality marking time point includes environmental perception data and control system data at each driving abnormality marking time point; The environmental perception data includes image data, point cloud data and radar data at each driving abnormality marking time point; The control system data includes engine response time, braking deceleration, braking distance, suspension compression and suspension rebound speed at each driving abnormality marking time point.

6. The method for extracting abnormal scenes from smart car driving data based on artificial intelligence according to claim 1 is characterized by: The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs clustering processing to generate abnormal scene data of each smart car and stores it. The specific processing conditions are: The integrated AI cloud computing center receives the processed comprehensive scene data of each driving abnormality marking time point, performs K-means clustering processing and analysis to obtain multiple abnormal scene feature clusters of each driving abnormality marking time point, obtains each abnormal cluster data feature of each driving abnormality marking time point through iterative updating, and matches each abnormal cluster data feature of each driving abnormality marking time point with the corresponding abnormal scene label pre-stored in the integrated AI cloud computing center to obtain each abnormal scene label of each driving abnormality marking time point; Based on each abnormal scene label at each driving abnormality marking time point, the processed comprehensive scene data at each driving abnormality marking time point is stored in a database under each corresponding abnormal scene label.

7. The method for extracting abnormal scenes from driving data of smart cars based on artificial intelligence according to claim 1 is characterized by: It also includes detecting the transmission channel performance parameters from each data transmission terminal to the integrated AI cloud computing center, processing and analyzing to generate data quality repair factors and performing data quality repair, specifically including: The transmission channel performance parameters include the real-time signal strength, interface rate, real-time delay and real-time signal-to-noise ratio of each data transmission terminal; Retrieve transmission channel performance verification parameters from the built-in database of the integrated AI cloud computing center, including real-time signal strength verification value, interface rate verification value, real-time delay verification value, and real-time signal-to-noise ratio verification value; A comprehensive comparison and analysis is performed on the transmission channel performance parameters and the transmission channel performance verification parameters to obtain the data quality impact value of each transmission channel, and the data quality impact value of each transmission channel is mapped and matched with the data quality repair factors corresponding to each data quality impact value interval pre-stored in the built-in database of the integrated AI cloud computing center to obtain the data quality repair factor of each transmission channel. The data quality impact value of each transmission channel is used to characterize the impact of the performance of each transmission channel on the data quality, and data quality repair is performed based on the data transmission repair factor.

8. The method for extracting abnormal scenes from driving data of a smart car based on artificial intelligence according to claim 3 is characterized by: The analysis and processing obtains the abnormal risk assessment value of each smart car, and the specific processing method is: ; in, For the The abnormal risk assessment value of a smart car, For the The accumulated mileage of a smart car, For the The cumulative driving time of a smart car, For the The cumulative number of driving failures of a smart car, The average mileage adjusted for the cumulative median mileage driven is The average time for driving is adjusted for the cumulative median. The cumulative median adjusted average number of failures for driving, is the weight factor of accumulated mileage, is the weight factor of the accumulated driving time, is the weight factor of the average number of cumulative failures during driving, Number the smart car. , is the total number of smart cars. The softplus function is a built-in function in Python. .

9. An artificial intelligence-based intelligent vehicle driving data abnormal scene extraction system, used to execute the method according to any one of claims 1 to 8, characterized in that: include: The abnormal risk assessment module is used to retrieve the basic information of each smart car from multiple data transmission terminals, analyze and process to obtain the abnormal risk assessment value of each smart car, and match and obtain the driving data extraction ratio of each smart car based on the abnormal risk assessment value of each smart car; A data extraction module is used to obtain the driving data packets of each smart car based on the driving data extraction ratio of each smart car, and at the same time obtain the time point of each driving abnormality mark, collect the comprehensive scene data of each driving abnormality mark time point from the driving data packet, and import it into the integrated AI cloud computing center after classified data processing; The scene extraction module is used to enable the integrated AI cloud computing center to receive the processed comprehensive scene data of each driving abnormality marking time point, perform clustering processing and analysis to generate abnormal scene data of each smart car and store it.

Images