Traffic accident detection alarm system and method based on intelligent vehicle-mounted terminal

Abstract

The invention provides a traffic accident detection and alarm system and method based on an intelligent vehicle-mounted terminal, and relates to the field of intelligent traffic. Then, the data analysis technology based on artificial intelligence is adopted to comprehensively analyze vehicle acceleration data and posture data in the vehicle running process to obtain the accident level of the vehicle, and finally, the intelligent vehicle-mounted terminal reports alarm information such as the accident level and positioning data of the vehicle. Therefore, the judgment standard can be dynamically adjusted according to actual data characteristics of different vehicle types and driving conditions, the accuracy of the finally obtained vehicle accident grade can be improved, and the rescue response speed and efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation, and more specifically, to a traffic accident detection and alarm system and method based on an intelligent vehicle-mounted terminal. Background Art

[0002] With the continuous development of social economy, the number of vehicles has continued to grow, road conditions have become increasingly complex, and the probability of traffic accidents has increased significantly. How to quickly discover, accurately locate and efficiently handle traffic accidents is of great significance to saving lives and restoring road traffic as soon as possible. Especially when the accident occurs in sparsely populated areas or involves illegal acts such as hit-and-run, being able to identify the accident in a timely and accurate manner and quickly report it to the police is not only crucial to rescuing the injured, but also provides a key guarantee for restoring the truth of the accident.

[0003] Patent CN113538901A discloses a traffic accident detection and alarm method based on an intelligent vehicle-mounted terminal, which uses a high-precision accelerometer and gyroscope built into the vehicle to calculate and correct the vehicle's posture and driving state, and thereby determines whether a traffic accident has occurred and the level of the accident. Specifically, the method determines the collision situation and its level by analyzing acceleration data, determines the rollover situation and its level by using posture data, and finally determines the level of the accident by combining the collision situation and the rollover situation.

[0004] However, this method has limitations in determining the collision level and rollover level that occur during vehicle operation. Specifically, the collision and rollover levels are determined by comparing the vehicle's acceleration data and posture data with preset thresholds. However, for different vehicle models and driving conditions, the same collision or rollover level may correspond to different acceleration values ​​or angle change rates. Fixed thresholds may not be able to fully adapt to various situations, which may lead to misjudgment of the collision level and rollover level, and directly lead to the final accident level being inaccurate. Inaccurate accident level will directly affect the speed and efficiency of rescue response.

[0005] Therefore, an optimized traffic accident detection and alarm solution based on intelligent vehicle-mounted terminals is expected. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present application provides a traffic accident detection and alarm system and method based on an intelligent vehicle-mounted terminal.

[0007] A traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal, comprising: The vehicle operation data acquisition module is used to obtain the vehicle acceleration data, posture data and positioning data of the vehicle during operation through the intelligent vehicle terminal; A vehicle accident grade result analysis module is used to comprehensively analyze the vehicle acceleration data and posture data of the vehicle during operation to obtain an accident grade classification result; The accident alarm information reporting module is used for the intelligent vehicle terminal to report the accident level classification result and the positioning data and other alarm information.

[0008] A traffic accident detection and alarm method based on an intelligent vehicle-mounted terminal, comprising: Obtain vehicle acceleration data, posture data, and positioning data during vehicle operation through an intelligent vehicle terminal; Comprehensively analyzing the vehicle acceleration data and posture data of the vehicle during operation to obtain an accident level classification result; The intelligent vehicle-mounted terminal reports the accident level classification result and the positioning data and other alarm information.

[0009] The present application has significant technical effects due to the adoption of the above technical solutions: the traffic accident detection alarm system and method based on the intelligent vehicle terminal provided by the present application first obtains the vehicle acceleration data, posture data and positioning data during the operation of the vehicle through the intelligent vehicle terminal, and then uses artificial intelligence-based data analysis technology to comprehensively analyze the vehicle acceleration data and posture data during the operation of the vehicle to obtain the vehicle's accident level, and finally the intelligent vehicle terminal reports the vehicle's accident level and positioning data and other alarm information. In this way, the judgment criteria can be dynamically adjusted according to the actual data characteristics of different vehicle models and driving conditions, which can improve the accuracy of the final vehicle accident level, thereby improving the rescue response speed and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 The system block diagram of the traffic accident detection and alarm system based on the intelligent vehicle-mounted terminal according to the embodiment of the present application.

[0012] Figure 2 It is a block diagram of a vehicle accident level result analysis module in a traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal according to an embodiment of the present application.

[0013] Figure 3The present invention is a block diagram of a vehicle collision situation analysis unit in a traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal according to an embodiment of the present application.

[0014] Figure 4 The present invention is a block diagram of a vehicle rollover situation analysis unit in a traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal according to an embodiment of the present application.

[0015] Figure 5 The present invention is a block diagram of a vehicle accident level result generating unit in a traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal according to an embodiment of the present application.

[0016] Figure 6 The present invention is a flowchart of a traffic accident detection and alarm method based on an intelligent vehicle-mounted terminal according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0018] Based on the technical problems raised by the above background technology, the present application provides a traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal. Figure 1 FIG. 1 is a system block diagram of a traffic accident detection and alarm system based on an intelligent vehicle terminal according to an embodiment of the present application. Figure 1 As shown, in the traffic accident detection and alarm system 100 based on the intelligent vehicle terminal, it includes: a vehicle operation data acquisition module 110, which is used to obtain the vehicle acceleration data, posture data and positioning data of the vehicle during operation through the intelligent vehicle terminal; a vehicle accident level result analysis module 120, which is used to comprehensively analyze the vehicle acceleration data and posture data of the vehicle during operation to obtain an accident level classification result; an accident alarm information reporting module 130, which is used for the intelligent vehicle terminal to report the accident level classification result and the alarm information such as the positioning data.

[0019] In the embodiment of the present application, the vehicle operation data acquisition module 110 is used to obtain the vehicle acceleration data, attitude data and positioning data of the vehicle during operation through the intelligent vehicle terminal. It should be understood that the vehicle acceleration data contains information about the speed change of the vehicle during driving, specifically including the acceleration value, the acceleration change rate and the duration of the acceleration change. The acceleration value intuitively reflects the speed of the vehicle speed change. For example, the acceleration value will change significantly when the vehicle accelerates or brakes suddenly. The acceleration change rate reflects the speed of the acceleration change over time, which can assist in judging the stability of the vehicle acceleration or deceleration process. The duration of the acceleration change indicates the length of time the vehicle is in a specific acceleration state. For example, when the acceleration jumps sharply during a collision, its duration is short, while when the vehicle accelerates or decelerates normally, the duration of the acceleration change is relatively regular. The vehicle attitude data mainly includes the pitch angle, the roll angle and its change amplitude, and the heading angle change rate. The pitch angle represents the angle of rotation of the vehicle around the horizontal axis, reflecting the up and down tilt of the front or rear of the vehicle, such as the pitch angle changes when the vehicle climbs or descends. The roll angle is the angle at which the vehicle rotates around the longitudinal axis, which can reflect the vehicle's left and right tilt. The roll angle will change significantly when the vehicle makes a sharp turn or rolls over. The change amplitude of the pitch angle and roll angle reflects the change size of these two angles over a period of time, which is used to judge the severity of the vehicle's posture change. The heading angle change rate represents the speed of the vehicle's driving direction change, which can reflect the frequency and intensity of the vehicle's steering operation. The vehicle positioning data contains the vehicle's geographic location information, such as longitude and latitude data, and may also contain altitude information. In complex terrain areas such as mountainous areas, this helps to more accurately describe the vehicle's location. The positioning data will also have a positioning timestamp to record the time when the location information is obtained, which is convenient for tracking the vehicle's position change trajectory at different time points. In other words, obtaining acceleration data can determine whether the vehicle has collided and the severity of the collision. The acceleration will change abnormally during a collision. By analyzing the acceleration value, change rate and duration, the specific collision situation of the vehicle can be understood. The posture data is used to determine whether the vehicle has overturned and the severity of the overturn. By monitoring the pitch angle, roll angle and its change amplitude and the heading angle change rate, abnormal changes in the vehicle's posture can be detected in time and the vehicle's overturning can be evaluated. Positioning data can accurately determine the location of the accident. When an accident occurs, fast and accurate location information is crucial for rescue personnel to arrive at the scene quickly. It is one of the key elements for achieving timely rescue. In general, by comprehensively analyzing these data, we can fully understand the operating status of the vehicle, which can provide strong support for accurately judging the accident level and subsequent alarm and rescue work, thereby improving the efficiency and effectiveness of traffic accident handling.

[0020] In the embodiment of the present application, the vehicle accident level result analysis module 120 is used to comprehensively analyze the vehicle acceleration data and posture data of the vehicle during operation to obtain the accident level classification result. Specifically, Figure 2 FIG. 1 is a block diagram of a vehicle accident level result analysis module in a traffic accident detection and alarm system based on an intelligent vehicle terminal according to an embodiment of the present application. Figure 2 As shown, the vehicle accident level result analysis module 120 includes: a vehicle collision situation analysis unit 121, which is used to perform acceleration analysis on the vehicle acceleration data of the vehicle during operation to obtain a vehicle collision situation reference data fusion feature vector; a vehicle rollover situation analysis unit 122, which is used to perform posture analysis on the posture data of the vehicle during operation to obtain a vehicle rollover situation reference data feature vector; a vehicle comprehensive situation fusion generation unit 123, which is used to fuse the vehicle collision situation reference data fusion feature vector and the vehicle rollover situation reference data feature vector to obtain a vehicle situation fusion feature vector; a vehicle accident level result generation unit 124, which is used to obtain the accident level classification result based on the vehicle situation fusion feature vector, and the accident level classification result is used to represent the accident level of the vehicle accident.

[0021] In the embodiment of the present application, the vehicle collision situation analysis unit 121 is used to perform acceleration analysis on the vehicle acceleration data of the vehicle during operation to obtain a vehicle collision situation reference data fusion feature vector. Specifically, Figure 3 FIG. 1 is a block diagram of a vehicle collision situation analysis unit in a traffic accident detection and alarm system based on an intelligent vehicle terminal according to an embodiment of the present application. Figure 3 As shown, the vehicle collision situation analysis unit 121 includes: a vehicle acceleration data sorting subunit 1211, which is used to sort the vehicle acceleration data of the vehicle during operation to obtain a vehicle acceleration time series input vector; a vehicle acceleration time series feature generation subunit 1212, which is used to input the vehicle acceleration time series input vector into a time series feature capturer to obtain a vehicle acceleration time series feature vector; a vehicle acceleration difference calculation subunit 1213, which is used to calculate the acceleration difference between two adjacent positions in the vehicle acceleration time series input vector to obtain a vehicle acceleration difference time series input vector; a vehicle acceleration difference time series feature generation subunit 1214, which is used to input the vehicle acceleration difference time series input vector into the time series feature capturer to obtain a vehicle acceleration difference time series feature vector; and a vehicle collision situation related feature fusion subunit 1215, which is used to fuse the vehicle acceleration time series feature vector and the vehicle acceleration difference time series feature vector to obtain the vehicle collision situation reference data fusion feature vector.

[0022] In an embodiment of the present application, the vehicle acceleration data sorting subunit 1211 is used to sort the vehicle acceleration data of the vehicle during operation to obtain a vehicle acceleration time series input vector. It should be understood that factors such as different vehicle models, sensor accuracy and installation location will cause differences in the collected original acceleration data. By sorting these data, data standardization can be achieved to ensure that data from different sources can be compared and analyzed on the same basis. And considering that the operating state of the vehicle is a dynamically changing process, the order of change of acceleration over time contains important information. Sorting into an acceleration time series input vector can clearly construct the time series relationship between the data and reflect the change of acceleration at different time points. In other words, this time series data helps to analyze the dynamic characteristics of the vehicle collision process, and can better capture the change law of acceleration before and after the collision, so as to more accurately judge the occurrence of the collision and the severity of the collision.

[0023] In an embodiment of the present application, the vehicle acceleration time series feature generation subunit 1212 is used to input the vehicle acceleration time series input vector into a time series feature capturer to obtain a vehicle acceleration time series feature vector. It should be understood that although the vehicle acceleration time series input vector contains acceleration change data over time, the key features of these data may not be intuitive. For example, when a collision occurs, there may be some subtle features in the acceleration change pattern, such as fluctuations of a specific frequency or short-term abnormal changes. In order to deeply explore the information hidden in the data sequence, find the characteristic patterns related to traffic accidents, and thus more accurately judge the collision situation, in this application, it is necessary to input the vehicle acceleration time series input vector into a time series feature capturer for processing. In particular, in this application, the time series feature capturer is a convolutional neural network model comprising a first convolution layer and a second convolution layer, wherein the first convolution layer uses a one-dimensional convolution kernel of a first scale, and the second convolution layer uses a one-dimensional convolution kernel of a second scale, and the first scale is different from the second scale. Convolution kernels of different scales can scan and extract features from the acceleration time series input vector at different granularities. Smaller-scale convolution kernels can capture local details in the data, such as small fluctuations in acceleration in a short period of time, which is very helpful for discovering subtle changes at the moment of collision; larger-scale convolution kernels can focus on the overall trend and long-distance dependencies of the data, such as the trend of acceleration changes over a period of time, to determine whether the vehicle is continuously accelerating, decelerating, or experiencing abnormal speed change patterns. The integration of multi-scale features can provide a more comprehensive understanding of the acceleration data. In other words, after capturing the time series features, the extracted acceleration time series feature vector has a stronger expressive ability and can better represent the key information in the vehicle acceleration data.

[0024] In the embodiment of the present application, the vehicle acceleration difference calculation subunit 1213 is used to calculate the acceleration difference between two adjacent positions in the vehicle acceleration time series input vector to obtain the vehicle acceleration difference time series input vector. It should be understood that the change trend of the acceleration is crucial to determine the vehicle state during the driving process. Although the original vehicle acceleration time series input vector contains the acceleration value at each moment, it is difficult to intuitively reflect the speed and direction of the acceleration change. By calculating the acceleration difference between adjacent positions to obtain the vehicle acceleration difference time series input vector, it can clearly show how the acceleration changes over time. For example, a positive acceleration difference indicates that the acceleration is increasing, and a negative value indicates that the acceleration is decreasing, and its absolute value reflects the severity of the acceleration change. When a traffic accident occurs, the acceleration usually changes sharply, and the acceleration difference time series input vector can more keenly capture this change trend, which helps to capture the occurrence of the accident more accurately. And the acceleration difference time series input vector can provide information about the rate of change of acceleration, which helps to better assess the severity of the accident.

[0025] In the embodiment of the present application, the vehicle acceleration difference time series feature generation subunit 1214 is used to input the vehicle acceleration difference time series input vector into the time series feature capturer to obtain the vehicle acceleration difference time series feature vector. It should be understood that although the vehicle acceleration difference time series input vector reflects the change of acceleration, there may still be some hidden deep features that are more valuable for judging traffic accidents. The time series feature capturer has a powerful feature extraction capability. By processing the acceleration difference time series input vector, these deep features, such as change patterns, periodic laws, etc., can be mined. These deep features can provide a more accurate basis for judging the running state of the vehicle and whether an accident has occurred, and improve the accuracy of traffic accident detection. Moreover, through the time series feature capturer, the acceleration difference time series input vector can be subjected to complex nonlinear transformations to extract features that adapt to various complex situations. These features can more comprehensively reflect the characteristics of the vehicle's acceleration difference under different driving conditions, enhance the system's ability to judge traffic accidents in complex scenarios, and help reduce misjudgments and missed judgments.

[0026] In the embodiment of the present application, the vehicle collision situation related feature fusion subunit 1215 is used to fuse the vehicle acceleration time series feature vector and the vehicle acceleration difference time series feature vector to obtain the vehicle collision situation reference data fusion feature vector. It should be understood that the vehicle acceleration time series feature vector contains the actual size information of the vehicle's acceleration at each moment, which can intuitively reflect the speed change during the vehicle's driving process. The vehicle acceleration difference time series feature vector focuses on the change trend and change rate of acceleration, highlighting the difference in acceleration at adjacent moments. The two carry different information, and after fusion, information complementation can be achieved. For example, when judging a vehicle collision, the acceleration time series feature vector can provide the absolute value of the acceleration at the moment of collision, so that the model knows the strength of the collision; the acceleration difference time series feature vector can reflect the speed of acceleration change before and after the collision, and supplement the dynamic information of acceleration change during the collision process. The combination of the two can more comprehensively present the change of acceleration when the vehicle collides. That is, the fused vehicle collision situation reference data fusion feature vector contains richer and more comprehensive information, which can more accurately identify whether the vehicle collides. When faced with complex driving conditions and slight changes in acceleration, misjudgment or missed judgment may occur based on a single feature vector. The fused feature vector comprehensively considers the acceleration size and rate of change, and can more accurately capture the collision characteristics, reduce the probability of misjudgment and missed judgment, and make the collision judgment result more in line with the actual situation.

[0027] In the embodiment of the present application, the vehicle rollover analysis unit 122 is used to perform posture analysis on the posture data of the vehicle during operation to obtain a vehicle rollover reference data feature vector. Specifically, Figure 4 FIG. 1 is a block diagram of a vehicle rollover analysis unit in a traffic accident detection and alarm system based on an intelligent vehicle terminal according to an embodiment of the present application. Figure 4As shown, the vehicle rollover situation analysis unit 122 includes: a vehicle posture data sorting subunit 1221, which is used to sort the posture data of the vehicle during operation to obtain a pitch angle timing input vector, a heading angle timing input vector and a roll angle timing input vector; a vehicle posture timing feature encoding subunit 1222, which is used to input the pitch angle timing input vector, the heading angle timing input vector and the roll angle timing input vector into a posture timing feature extractor to obtain a pitch angle timing feature vector, a heading angle timing feature vector and a roll angle timing feature vector; The vehicle attitude timing feature integration subunit 1223 is used to arrange the pitch angle timing feature vector, the heading angle timing feature vector and the roll angle timing feature vector in two dimensions to obtain the attitude timing input matrix; the vehicle attitude key feature capture subunit 1224 is used to input the attitude timing input matrix into the attitude key feature capturer to obtain the vehicle rollover condition reference data feature matrix; the vehicle attitude key feature dimensionality reduction subunit 1225 is used to expand the vehicle rollover condition reference data feature matrix to obtain the vehicle rollover condition reference data feature vector.

[0028] In the embodiment of the present application, the vehicle posture data sorting subunit 1221 is used to sort the posture data of the vehicle during operation to obtain the pitch angle time series input vector, the heading angle time series input vector and the roll angle time series input vector. It should be understood that the collection method and format of the posture data during the operation of the vehicle are various, and the original data is relatively messy and difficult to be directly processed by the subsequent model used for accident detection. Sorting the posture data into the pitch angle, heading angle and roll angle time series input vector can convert the data into a unified and standardized format, so that the model can effectively read and analyze these data, extract key information therein, and judge whether the vehicle has overturned or other accidents. That is, the sorted posture data is presented in a clear time series input vector, which can more accurately reflect the changes in the vehicle posture. For example, when judging whether the vehicle has overturned, the change trajectory of the pitch angle and the roll angle can be more accurately tracked, and abnormal angle changes can be discovered in time. Compared with the judgment based on the messy data when not sorted, the accuracy of the overturning accident detection is greatly improved, and the occurrence of misjudgment and missed judgment is reduced.

[0029] In an embodiment of the present application, the vehicle posture time series feature encoding subunit 1222 is used to input the pitch angle time series input vector, the heading angle time series input vector and the roll angle time series input vector into the posture time series feature extractor to obtain the pitch angle time series feature vector, the heading angle time series feature vector and the roll angle time series feature vector. In particular, the posture time series feature extractor in the present application is a convolutional neural network model including a fully connected layer and a one-dimensional convolution layer. It should be understood that the posture changes of the vehicle during driving are complex and diverse, and are affected by many factors. It is difficult to comprehensively and accurately judge the actual posture status of the vehicle only by relying on the original time series data. The posture time series feature extractor can perform complex nonlinear transformations on the input time series vectors and learn the laws and characteristics of vehicle posture changes under different driving scenarios. For example, under different road conditions, vehicle speeds and driving operations, the characteristics of vehicle posture changes are different. Through the learning and processing of the feature extractor, it can better adapt to these complex situations, accurately identify abnormal posture changes, and provide a more reliable basis for accident judgment. Specifically, the one-dimensional convolution layer in the posture temporal feature extractor structure can perform convolution operations along the time dimension of the temporal data. For the pitch angle, heading angle, and roll angle temporal input vectors, it can perform weighted summation of the data at each time point and its adjacent time points to extract local temporal features; the fully connected layer will integrate the local features extracted by the one-dimensional convolution layer, consider the relationship between the feature information of all time points, that is, it will perform weighted calculations on the features after convolution processing, and combine different local features to form a more global and representative feature vector. In other words, the one-dimensional convolution layer focuses on extracting the local temporal features of the input vector and pays attention to the details of the posture changes in a short period of time; the fully connected layer integrates these local features from a global perspective and comprehensively considers the posture changes in the entire time series. This combination allows the extracted features to contain both the detailed information of the instantaneous changes in the vehicle posture and the overall change trend over a long period of time, which can more comprehensively describe the changes in the vehicle posture and improve the ability to capture accident-related posture features.

[0030] In an embodiment of the present application, the vehicle attitude timing feature integration subunit 1223 is used to arrange the pitch angle timing feature vector, the heading angle timing feature vector and the roll angle timing feature vector in two dimensions to obtain an attitude timing input matrix. Accordingly, considering that the pitch angle, heading angle and roll angle of the vehicle do not change in isolation, there is a certain correlation between them. Through the two-dimensional arrangement, the relationship between these features can be intuitively displayed in the data structure, so that the model can better capture the mutual influence and synergy between the changes in different attitude angles when processing data, so as to more comprehensively analyze the changes in the attitude of the vehicle. And compared to processing three vectors separately, after arranging them into a matrix, the efficiency of matrix operations can be used for batch processing. For example, when performing operations such as convolution, data in matrix form can more conveniently apply related algorithms to improve data processing speed. At the same time, this structured data organization method helps the model to more accurately extract complex attitude features, thereby improving the accuracy of judging accident situations such as vehicle rollover.

[0031] In the embodiment of the present application, the vehicle posture key feature capture subunit 1224 is used to input the posture time series input matrix into the posture key feature capturer to obtain the vehicle rollover reference data feature matrix. In particular, the posture key feature capturer in the present application is a convolutional neural network model using a spatial attention mechanism. It should be understood that although the posture time series input matrix integrates the feature information of different posture angles (pitch angle, heading angle, roll angle) of the vehicle at different time points, it may contain a large amount of redundant information and noise. The posture key feature capturer can perform deep processing on this matrix, and filter and extract key features closely related to the vehicle rollover situation. For example, in the process of normal driving and rollover of the vehicle, the change of posture angle has specific patterns and features. Through the capturer, these key patterns can be accurately identified, while ignoring those irrelevant information, providing a strong basis for accurately judging the vehicle rollover situation. Specifically, the posture key feature capturer uses a spatial attention mechanism to automatically focus on the most critical areas and features in the posture time series input matrix for judging the vehicle rollover. In the matrix, elements at different positions may have different importance for rollover judgment. The spatial attention mechanism can assign different weights according to the importance of the elements. For example, when a vehicle is about to roll over, the changes in certain posture angles within a specific time period may be key signals. The attention mechanism will increase the focus on these key areas and highlight these important features, thereby improving the sensitivity and recognition accuracy of vehicle rollover. Moreover, convolutional neural networks have unique advantages in processing data with spatial structures. For the posture time series input matrix, the convolution layer can perform sliding convolution operations on the matrix through the convolution kernel to extract local feature patterns. These local features can reflect the change relationship between posture angles at adjacent time points and between different angles. Through multi-layer convolution and pooling operations, the network can gradually extract more advanced and abstract features, and comprehensively describe the posture changes of the vehicle from micro to macro. This hierarchical feature extraction method enables the model to capture complex posture features and improve the ability to judge vehicle rollover.

[0032] In an embodiment of the present application, the vehicle posture key feature dimensionality reduction subunit 1225 is used to expand the vehicle rollover condition reference data feature matrix to obtain the vehicle rollover condition reference data feature vector. It should be understood that the elements in the vehicle rollover condition reference data feature matrix contain vehicle posture information of different dimensions, but this information is not conducive to direct comprehensive analysis in matrix form. After expanding into a feature vector, the features of each dimension can be integrated into one vector, which is convenient for subsequent models to comprehensively evaluate the vehicle's rollover condition as a whole, comprehensively consider the relationship between different posture features, and thus more accurately judge the accident situation.

[0033] In the embodiment of the present application, the vehicle comprehensive situation fusion generation unit 123 is used to fuse the vehicle collision situation reference data fusion feature vector and the vehicle rollover situation reference data feature vector to obtain the vehicle situation fusion feature vector. It should be understood that the vehicle collision situation reference data fusion feature vector mainly focuses on acceleration related information, reflecting the speed change and related dynamic characteristics when the vehicle collides; while the vehicle rollover situation reference data feature vector focuses on the vehicle posture angle change information, reflecting the abnormality of the vehicle in the spatial posture. These two sets of vectors describe the situation of the vehicle in the accident from different dimensions, and it is impossible to fully grasp the actual situation of the vehicle accident based on a single vector. Fusion of them can achieve information complementarity, integrate the key features of collision and rollover, and provide more complete data support for accident judgment. That is, the fused vehicle situation fusion feature vector contains richer and more comprehensive accident related information. By comprehensively considering the characteristics of collision and rollover, the system can more accurately judge the severity of the accident and reduce misjudgment and missed judgment. For example, in complex situations such as a minor collision with a large rollover of the vehicle, or a severe collision with a small rollover, the fused feature vector can more accurately reflect the actual situation, thereby giving a more realistic accident level.

[0034] In the embodiment of the present application, the vehicle accident level result generating unit 124 is used to obtain the accident level classification result based on the vehicle situation fusion feature vector, and the accident level classification result is used to indicate the accident level of the vehicle accident. Specifically, Figure 5 FIG. 1 is a block diagram of a vehicle accident level result generating unit in a traffic accident detection and alarm system based on an intelligent vehicle terminal according to an embodiment of the present application. Figure 5 As shown, the vehicle accident level result generating unit 124 includes: a vehicle condition fusion feature optimization subunit 1241, used for performing latent space mapping of the vehicle condition fusion feature vector based on low-rank feature extraction to obtain an optimized vehicle condition fusion feature vector; a vehicle condition fusion feature parsing subunit 1242, used for inputting the optimized vehicle condition fusion feature vector into an accident level classifier to obtain the accident level classification result.

[0035] In an embodiment of the present application, the vehicle situation fusion feature optimization subunit 1241 is used to perform a latent space mapping based on low-rank feature extraction on the vehicle situation fusion feature vector to obtain an optimized vehicle situation fusion feature vector. In particular, considering that there may be a distribution bias in vehicle accident data, for example, some types of accidents appear more frequently in the data set, while other types of accidents appear less frequently, the data-driven model tends to learn the patterns that appear frequently in the data, while ignoring those uncommon patterns. When the model encounters new data with a different distribution from the training data, its generalization ability will be limited, and it will not be able to accurately classify new accident types. At the same time, the data-driven feature learning method mainly relies on the data itself to learn features, and less consideration is given to prior knowledge and domain expert experience. In the classification of vehicle accident levels, domain experts may summarize some important features and rules based on years of accident handling experience, but these experiences cannot be directly integrated into the data-driven model. For example, experts may know that certain specific acceleration change patterns or attitude angle combinations usually correspond to serious accidents, but the model cannot make accurate judgments due to the lack of such knowledge, resulting in misjudgment. Based on this, the present application needs to perform latent space mapping on the vehicle condition fusion feature vector based on low-rank feature extraction to obtain an optimized vehicle condition fusion feature vector.

[0036] Specifically, in the embodiment of the present application, the vehicle condition fusion feature vector is subjected to a latent space mapping based on low-rank feature extraction to obtain an optimized vehicle condition fusion feature vector, including: extracting a vehicle domain prior knowledge constraint matrix; performing low-rank feature extraction on the vehicle domain prior knowledge constraint matrix to obtain a set of vehicle domain core prior information encoding vectors. The process can be expressed by the formula: ;in, represents the set of core prior information encoding vectors in the vehicle domain, Respectively represent the first, second, and A vehicle domain core prior information encoding vector, represents the transpose operation, represents low-rank feature extraction, represents the vehicle domain prior knowledge constraint matrix, represents the vehicle domain diagonal matrix, Respectively represent the first and second diagonal matrix of the vehicle domain The value of the position; The vehicle domain prior information response encoding latent space matrix is ​​constructed between the vehicle condition fusion feature vector and each vehicle domain core prior information encoding vector in the set of vehicle domain core prior information encoding vectors to obtain a set of vehicle domain prior information response encoding latent space matrices. The process can be expressed by the formula: ;in, represents the vehicle situation fusion feature vector, Express After linear transformation, the feature vector has the same feature scale as the corresponding vehicle domain core prior information encoding vector. Indicates A vehicle domain core prior information encoding vector, represents matrix multiplication, represents the length of the core prior information encoding vector in the vehicle domain, Indicates The vehicle domain prior information response encoding latent space matrix; The vehicle domain prior information correlation measurement factor of each vehicle domain prior information response encoding latent space matrix in the set of vehicle domain prior information response encoding latent space matrices is calculated to obtain a set of vehicle domain prior information correlation measurement factors. The process can be expressed by the formula: ;in, Represents the matrix norm, Indicates A measurement factor for the relevance of prior information in the vehicle domain; Based on the set of vehicle domain prior information correlation measurement factors, the set of vehicle domain prior information response encoding latent space matrices is sparsely dynamically fused to obtain the vehicle domain prior information response projection encoding matrix. The process can be expressed by the formula: ;in, represents the normalized exponential function, Represents the vehicle domain prior information response projection encoding matrix; The vehicle condition fusion feature vector is mapped to the feature space of the vehicle domain prior information response projection coding matrix to obtain the optimized vehicle condition fusion feature vector. The process can be expressed by the formula: ;in, Represents the optimized vehicle situation fusion feature vector.

[0037] In view of the above technical problems, in the technical solution of the present application, the vehicle condition fusion feature vector is subjected to latent space mapping based on low-rank feature extraction to obtain an optimized vehicle condition fusion feature vector. First, it is necessary to extract the vehicle domain prior knowledge constraint matrix. It should be understood that by extracting the vehicle domain prior knowledge constraint matrix, the original scattered and fuzzy vehicle domain prior knowledge can be organized in the form of a matrix, which can make the vehicle domain knowledge orderly from disorder and easier to analyze later.

[0038] Next, the vehicle domain prior knowledge constraint matrix is ​​subjected to low-rank feature extraction to obtain a set of vehicle domain core prior information encoding vectors. It should be understood that in the original prior knowledge constraint matrix, there are some non-critical information, which may interfere with the optimization process of the model and affect the model's learning and judgment of the core features. Through low-rank feature extraction, these information are filtered to remove non-critical information that has little impact on tasks such as vehicle accident level classification, and only the truly important core prior information is retained. Moreover, the original vehicle domain prior knowledge constraint matrix has a high dimension because it contains a large amount of information. After low-rank feature extraction, it can be converted into a relatively low-dimensional set of vehicle domain core prior information encoding vectors. This greatly reduces the amount of calculation when performing subsequent calculations and processing based on this information, improves the operating efficiency of the algorithm, and helps save computing resources and time costs.

[0039] Then, a vehicle domain prior information response encoding latent space matrix is ​​constructed between the vehicle condition fusion feature vector and each vehicle domain core prior information encoding vector in the set of vehicle domain core prior information encoding vectors to obtain a set of vehicle domain prior information response encoding latent space matrices. It should be understood that the vehicle condition fusion feature vector contains various types of original feature information during the vehicle operation process, while the vehicle domain core prior information encoding vector set is the refined prior knowledge. Constructing the latent space matrix between the two can deeply explore the complex nonlinear relationship between the original features and the prior knowledge. For example, when analyzing the vehicle collision situation, through the latent space mapping, it can be found that the deep connection between the vehicle acceleration characteristics and the prior collision severity judgment knowledge is not just a simple linear association, but a complex response mode based on knowledge interpretation. In essence, the vehicle domain prior information response encoding latent space matrix is ​​obtained by re-encoding the vehicle condition fusion feature vector from the perspective of vehicle domain prior knowledge. This process integrates the interpretation and processing of prior knowledge into the feature representation, which can achieve directional enhancement of feature representation.

[0040] Next, the vehicle field prior information relevance measurement factor of each vehicle field prior information response encoding latent space matrix in the set of the vehicle field prior information response encoding latent space matrix is ​​calculated to obtain a set of vehicle field prior information relevance measurement factors. It should be understood that in the vehicle field, the prior information response encoding latent space matrix set contains rich and diverse information, which involves multiple aspects of vehicle operation and multiple dimensions of prior knowledge. By calculating the vehicle field prior information relevance measurement factor, key information can be accurately extracted from each latent space matrix. For example, in the latent space matrix of a vehicle collision accident, there may be a lot of information such as vehicle speed, acceleration change, and posture at the moment of collision. The calculation of the relevance measurement factor can help the model focus on the most critical information for judging the severity of the collision and the type of accident, and discard those redundant information that has little impact on the accident judgment, so as to make the subsequent analysis more efficient and accurate. And because the relevance measurement factor is a deep compression and refinement of the vehicle field prior information response encoding latent space matrix, a substantial reduction in data dimension is achieved, which can significantly reduce the complexity of subsequent calculations.

[0041] Then, based on the set of vehicle domain prior information correlation measurement factors, the set of vehicle domain prior information response coding latent space matrices is sparsely dynamically fused to obtain the vehicle domain prior information response projection coding matrix. It should be understood that the vehicle domain prior information response coding latent space matrix set contains rich but complex information, which reflects the relationship between the vehicle's operating state and prior knowledge from different angles. With the help of the vehicle domain prior information correlation measurement factor set, sparse dynamic fusion can accurately determine which information is the most critical. For example, when analyzing a vehicle collision accident, some matrix information related to the sharp change in acceleration at the moment of collision, if its correlation measurement factor is high, will be focused on the fusion process, while those matrix information that has little to do with the collision and low measurement factors, such as some small posture change information when the vehicle is driving normally, will be weakened or ignored. In this way, when constructing the vehicle domain prior information response projection coding matrix, we can focus on the features that are truly valuable for accident analysis, greatly improving the accuracy of feature selection.

[0042] Finally, the vehicle situation fusion feature vector is mapped to the feature space of the vehicle domain prior information response projection coding matrix to obtain the optimized vehicle situation fusion feature vector. It should be understood that the vehicle domain prior information response projection coding matrix combines multi-angle prior knowledge and is a highly refined and integrated knowledge of the vehicle domain. When the vehicle situation fusion feature vector is mapped to the feature space of this matrix, the deep fusion of the vehicle actual operation data features and prior knowledge is achieved. The optimized vehicle situation fusion feature vector better fits the constraints and guidance of prior knowledge, which means that the model can analyze more stably and accurately when processing different vehicle operation scenarios and data. Specifically, since prior knowledge is integrated into the feature vector, the model no longer relies solely on the patterns in the training data, but can make reasonable inferences and judgments on new and unseen data based on prior knowledge. For example, when the vehicle encounters some rare accident scenes, the model can accurately analyze and classify the accident with the help of the prior knowledge contained in the optimized feature vector, instead of misjudgment or inability to judge due to lack of relevant training data, thereby significantly improving the generalization ability of the model.

[0043] In an embodiment of the present application, the vehicle condition fusion feature parsing subunit 1242 is used to input the optimized vehicle condition fusion feature vector into the accident level classifier to obtain the accident level classification result. It should be understood that the optimized vehicle condition fusion feature vector integrates key information from many aspects such as vehicle collision and rollover, but this information itself is only a representation of data and cannot directly judge the severity of the accident. The accident level classifier is a model specially designed to process such feature vectors and is trained with a large amount of data. By inputting the optimized vehicle condition fusion feature vector into it, the ability of the classifier can be used to quantify and grade the accident condition of the vehicle, thereby obtaining an accurate accident level classification result. This process converts complex data into practical and easy-to-understand accident levels, providing a key basis for subsequent decision-making and actions.

[0044] In an embodiment of the present application, the accident alarm information reporting module 130 is used for the intelligent vehicle terminal to report the accident level classification results and the alarm information such as the positioning data. It should be understood that after a traffic accident occurs, the alarm information such as the accident level classification results and the positioning data is crucial for subsequent rescue, traffic management and other work. The intelligent vehicle terminal reports this information to ensure that relevant departments and personnel obtain key data as soon as possible. For example, rescue personnel need to know the severity of the accident (through the accident level classification results) to prepare corresponding rescue equipment and resources, and also need accurate positioning data (positioning data) to quickly find the accident scene. Moreover, automatically reporting information after an accident is detected meets the requirements of intelligent traffic management, reduces the links of manual intervention, and avoids problems caused by human negligence or delays. This automated reporting mechanism can respond to traffic accidents more quickly and accurately, and improve the safety and reliability of the entire transportation system.

[0045] In summary, the traffic accident detection alarm system 100 based on the intelligent vehicle terminal according to the embodiment of the present application is explained, which first obtains the vehicle acceleration data, posture data and positioning data during the operation of the vehicle through the intelligent vehicle terminal, and then uses the data analysis technology based on artificial intelligence to comprehensively analyze the vehicle acceleration data and posture data during the operation of the vehicle to obtain the accident level of the vehicle, and finally the intelligent vehicle terminal reports the accident level of the vehicle and the alarm information such as the positioning data. In this way, the judgment standard can be dynamically adjusted according to the actual data characteristics of different vehicle models and driving conditions, which can improve the accuracy of the final vehicle accident level, thereby improving the rescue response speed and efficiency.

[0046] Figure 6 FIG. 1 is a flow chart of a traffic accident detection and alarm method based on an intelligent vehicle terminal according to an embodiment of the present application. Figure 6 As shown, in the traffic accident detection and alarm method based on the intelligent vehicle terminal, it includes: S1, obtaining the vehicle acceleration data, posture data and positioning data of the vehicle during operation through the intelligent vehicle terminal; S2, comprehensively analyzing the vehicle acceleration data and posture data of the vehicle during operation to obtain the accident level classification result; S3, the intelligent vehicle terminal reports the accident level classification result and the positioning data and other alarm information.

[0047] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned traffic accident detection and alarm method based on the intelligent vehicle terminal have been referred to above. Figures 1 to 5 The invention has been introduced in detail in the description of the traffic accident detection and alarm system based on the intelligent vehicle terminal, and therefore, its repeated description will be omitted.

[0048] In summary, the traffic accident detection and alarm method based on the intelligent vehicle terminal according to the embodiment of the present application is explained. It first obtains the vehicle acceleration data, posture data and positioning data during the operation of the vehicle through the intelligent vehicle terminal, and then uses the data analysis technology based on artificial intelligence to comprehensively analyze the vehicle acceleration data and posture data during the operation of the vehicle to obtain the accident level of the vehicle. Finally, the intelligent vehicle terminal reports the accident level of the vehicle and the alarm information such as the positioning data. In this way, the judgment standard can be dynamically adjusted according to the actual data characteristics of different vehicle models and driving conditions, which can improve the accuracy of the final vehicle accident level, thereby improving the rescue response speed and efficiency.

Claims

1. A traffic accident detection and alarm system based on an intelligent vehicle-mounted terminal, characterized in that: include: The vehicle operation data acquisition module is used to obtain the vehicle acceleration data, posture data and positioning data of the vehicle during operation through the intelligent vehicle terminal; A vehicle accident grade result analysis module is used to comprehensively analyze the vehicle acceleration data and posture data of the vehicle during operation to obtain an accident grade classification result; The accident alarm information reporting module is used for the intelligent vehicle terminal to report the accident level classification result and the positioning data and other alarm information.

2. The traffic accident detection and alarm system based on the intelligent vehicle terminal according to claim 1 is characterized in that: The vehicle accident level result analysis module includes: A vehicle collision situation analysis unit, used for performing acceleration analysis on the vehicle acceleration data of the vehicle during operation to obtain a vehicle collision situation reference data fusion feature vector; A vehicle rollover analysis unit, used for performing posture analysis on the posture data of the vehicle during operation to obtain a vehicle rollover reference data feature vector; A vehicle comprehensive situation fusion generation unit, used for fusing the vehicle collision situation reference data fusion feature vector and the vehicle rollover situation reference data feature vector to obtain a vehicle situation fusion feature vector; The vehicle accident level result generating unit is used to fuse the feature vector based on the vehicle situation to obtain the accident level classification result, and the accident level classification result is used to indicate the accident level of the vehicle accident.

3. The traffic accident detection and alarm system based on the intelligent vehicle terminal according to claim 2 is characterized in that: The vehicle collision situation analysis unit comprises: A vehicle acceleration data sorting subunit, used for sorting the vehicle acceleration data during the operation of the vehicle to obtain a vehicle acceleration time series input vector; A vehicle acceleration time series feature generating subunit, used for inputting the vehicle acceleration time series input vector into a time series feature capturer to obtain a vehicle acceleration time series feature vector; A vehicle acceleration difference calculation subunit, used for calculating the acceleration difference between two adjacent positions in the vehicle acceleration time series input vector to obtain a vehicle acceleration difference time series input vector; A vehicle acceleration difference timing feature generating subunit, used for inputting the vehicle acceleration difference timing input vector into the timing feature capturer to obtain a vehicle acceleration difference timing feature vector; The vehicle collision situation related feature fusion subunit is used to fuse the vehicle acceleration time series feature vector and the vehicle acceleration difference time series feature vector to obtain the vehicle collision situation reference data fusion feature vector.

4. The traffic accident detection and alarm system based on the intelligent vehicle terminal according to claim 3 is characterized in that: The vehicle rollover situation analysis unit comprises: A vehicle attitude data sorting subunit, used for sorting the attitude data of the vehicle during operation to obtain a pitch angle timing input vector, a heading angle timing input vector and a roll angle timing input vector; A vehicle attitude timing feature encoding subunit, used for inputting the pitch angle timing input vector, the heading angle timing input vector and the roll angle timing input vector into an attitude timing feature extractor to obtain a pitch angle timing feature vector, a heading angle timing feature vector and a roll angle timing feature vector; A vehicle attitude timing feature integration subunit, used for performing two-dimensional arrangement on the pitch angle timing feature vector, the heading angle timing feature vector and the roll angle timing feature vector to obtain an attitude timing input matrix; A vehicle posture key feature capture subunit, used for inputting the posture time series input matrix into a posture key feature capturer to obtain a vehicle rollover condition reference data feature matrix; The vehicle posture key feature dimensionality reduction subunit is used to expand the vehicle rollover condition reference data feature matrix to obtain the vehicle rollover condition reference data feature vector.

5. The traffic accident detection and alarm system based on the intelligent vehicle terminal according to claim 4 is characterized in that: The posture temporal feature extractor is a convolutional neural network model including a fully connected layer and a one-dimensional convolutional layer, the posture key feature capturer is a convolutional neural network model using a spatial attention mechanism, and the temporal feature capturer is a convolutional neural network model including a first convolutional layer and a second convolutional layer, wherein the first convolutional layer uses a one-dimensional convolutional kernel of a first scale, and the second convolutional layer uses a one-dimensional convolutional kernel of a second scale, and the first scale is different from the second scale.

6. The traffic accident detection and alarm system based on the intelligent vehicle terminal according to claim 5 is characterized in that: The vehicle accident level result generating unit comprises: A vehicle condition fusion feature optimization subunit, used for performing latent space mapping based on low-rank feature extraction on the vehicle condition fusion feature vector to obtain an optimized vehicle condition fusion feature vector; The vehicle condition fusion feature analysis subunit is used to input the optimized vehicle condition fusion feature vector into the accident level classifier to obtain the accident level classification result.

7. The traffic accident detection and alarm system based on the intelligent vehicle terminal according to claim 6 is characterized in that: The vehicle situation fusion feature optimization subunit is used to: Extract vehicle domain prior knowledge constraint matrix; Performing low-rank feature extraction on the vehicle domain prior knowledge constraint matrix to obtain a set of vehicle domain core prior information encoding vectors; Constructing a vehicle domain prior information response encoding latent space matrix between the vehicle condition fusion feature vector and each vehicle domain core prior information encoding vector in the set of vehicle domain core prior information encoding vectors to obtain a set of vehicle domain prior information response encoding latent space matrices; Calculating the vehicle domain prior information correlation measurement factor of each vehicle domain prior information response encoding latent space matrix in the set of vehicle domain prior information response encoding latent space matrices to obtain a set of vehicle domain prior information correlation measurement factors; Based on the set of vehicle domain prior information correlation measurement factors, sparsely dynamically fuse the set of vehicle domain prior information response encoding latent space matrices to obtain a vehicle domain prior information response projection encoding matrix; The vehicle situation fusion feature vector is mapped to the feature space of the vehicle domain prior information response projection coding matrix to obtain the optimized vehicle situation fusion feature vector.

8. A traffic accident detection and alarm method based on an intelligent vehicle-mounted terminal, characterized in that: include: Obtain vehicle acceleration data, posture data, and positioning data during vehicle operation through an intelligent vehicle terminal; Comprehensively analyzing the vehicle acceleration data and posture data of the vehicle during operation to obtain an accident level classification result; The intelligent vehicle-mounted terminal reports the accident level classification result and the positioning data and other alarm information.

9. The traffic accident detection and alarm method based on the intelligent vehicle terminal according to claim 8 is characterized in that: Comprehensively analyzing the vehicle acceleration data and posture data of the vehicle during operation to obtain accident level classification results, including: Performing acceleration analysis on the vehicle acceleration data of the vehicle during operation to obtain a vehicle collision situation reference data fusion feature vector; Performing posture analysis on the posture data of the vehicle during operation to obtain a reference data feature vector of the vehicle rollover condition; Fusing the vehicle collision condition reference data fusion feature vector and the vehicle rollover condition reference data feature vector to obtain a vehicle condition fusion feature vector; Based on the fusion feature vector of the vehicle condition, the accident level classification result is obtained, and the accident level classification result is used to indicate the accident level of the accident occurring to the vehicle.

10. The traffic accident detection and alarm method based on the intelligent vehicle terminal according to claim 9 is characterized in that: Based on the vehicle condition fusion feature vector, the accident level classification result is obtained, and the accident level classification result is used to indicate the accident level of the vehicle accident, including: Performing latent space mapping based on low-rank feature extraction on the vehicle condition fusion feature vector to obtain an optimized vehicle condition fusion feature vector; The optimized vehicle condition fusion feature vector is input into the accident level classifier to obtain the accident level classification result.

Images