An intelligent driving safety detection system and method based on artificial intelligence

Abstract

The present application relates to the technical field of image detection, in particular to a driving safety intelligent detection system and method based on artificial intelligence, comprising the following steps: S100: collecting vehicle information and corresponding vehicle trajectory information of illegal driving, collecting vehicle information and collecting image data of a certain time period obtained by a wide-angle camera, forming an image data set; S200: using a digital signature algorithm to encrypt and store the collected data; S300: analyzing the change rule of each image feature point at different times from each image in the image data set, and judging whether the state of each image feature point is static or dynamic; S400: matching the dynamic feature data with the historical trajectory set, and further analyzing the risk degree of the current state of the concerned vehicle; S500: reminding when the risk degree of the concerned vehicle exceeds the threshold value; the present application reduces illegal driving and detects the safety status of people during driving in a timely manner.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, specifically to an intelligent vehicle safety detection system and method based on artificial intelligence. Background Technology

[0002] Transportation is an indispensable part of modern life. In a narrow sense, it refers to all man-made devices used for human travel or transportation. In today's era of efficiency and the belief that time is money, the benefits of convenient transportation are enormous. It not only changes people's travel, lifestyles, values, and quality of life, but also alters their social relationships, communication methods, activity pace, knowledge structure, and cultural customs.

[0003] With the advancement of technology and economic development, most families are able to purchase vehicles. However, with the increase in vehicles, many drawbacks have also emerged: people often neglect safety issues when using vehicles, and traffic violations have become commonplace. This leads to frequent vehicle collisions, causing property damage and even endangering lives. Therefore, how to address traffic violations and monitor safety during driving has become a major research issue.

[0004] Therefore, people need an intelligent vehicle safety detection system and method based on artificial intelligence to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent vehicle safety detection system and method based on artificial intelligence to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent vehicle safety detection method based on artificial intelligence, comprising the following steps:

[0007] S100: Collect information on all vehicles that have committed traffic violations and their corresponding trajectory information from the big data network to form a vehicle information set and a historical trajectory set. Collect information on vehicles that are currently logged into the intelligent detection system, set the vehicles logged into the intelligent detection system as vehicles to be monitored, and collect image data of the vehicles to be monitored within a certain time period using a wide-angle camera to form an image dataset.

[0008] S200: Acquire vehicle information set and historical trajectory set, and obtain vehicle information and corresponding image datasets for vehicles logged into the cloud platform, and use digital signature algorithm to encrypt and store the data;

[0009] S300: Extract the stored image dataset, analyze the change pattern of each image feature point at different times from each image in the image dataset, and further determine whether the state of each image feature point is static or dynamic;

[0010] S400: Fit the trajectory of the feature points identified as dynamic feature points to construct a two-dimensional plane coordinate system, extract feature points aggregated towards the y-axis, match the dynamic feature data with the historical trajectory set based on the relationship between the vehicle of interest and the vehicle information set, analyze whether the vehicle information belongs to the vehicle information set, if it does, issue a warning for data with a high risk level according to rough calculation, otherwise continue matching; use the matching results to analyze the risk level of each dynamic feature point, and further analyze the risk level of the vehicle of interest in its current state; first, perform autonomous screening to filter out those with a low risk level according to rough calculation, such as dynamic feature points that are far away from the vehicle of interest; then, match the data with higher risks to further accurately analyze the risk level.

[0011] S500: Issue a warning when the danger level of the vehicle under concern exceeds a threshold.

[0012] Furthermore, step S100 includes:

[0013] S110: Collect information on vehicles that have committed traffic violations and their corresponding trajectory information from the big data network, forming a vehicle information set A = {a1, a2, ..., an} and a historical trajectory set B = {b1, b2, ..., bn}, where a1, a2, ..., an represent the information of the 1st, 2nd, ..., nth vehicles, and b1, b2, ..., bn represent all historical trajectory information of the 1st, 2nd, ..., nth vehicles; the vehicle information for violations includes information on vehicles that violated traffic rules but did not cause an accident and information on vehicles that violated traffic rules and caused an accident; the vehicle information set includes vehicle type, speed standard, etc.; the historical trajectory set includes vehicle speed and safety status when traveling on each trajectory.

[0014] S120: When the intelligent detection system identifies a vehicle of interest (aν) logging in, it uses the wide-angle camera of the vehicle of interest (aν) to collect image data captured at a certain time series, forming an image dataset: C={c1,c2,…,cm}, where c1,c2,…,cm represent the image data captured at the 1st, 2nd,…,mth time points.

[0015] Furthermore, step S200 includes:

[0016] The system acquires vehicle information sets and historical trajectory sets, and obtains vehicle information and corresponding image datasets for logging into the cloud platform. The data is then encrypted and stored using a digital signature algorithm. The digital signature algorithm is a conventional technique used by those skilled in the art, and therefore will not be described in detail in this application.

[0017] Furthermore, step S300 includes:

[0018] S310: Extract the stored image dataset C of the vehicle of interest aν, and use the LBP feature algorithm to feature m images in the image dataset C; obtain the feature values ​​of the m feature images to form a pixel feature set D = {d1, d2, ..., dh}, where d1, d2, ..., dh represent the feature values ​​of the 1st, 2nd, ..., hth pixels in the m images; based on any pixel di in the pixel feature set D, obtain the pixel consistency formula: α = 1 - 1 / (d(i+k) - di), where k = 1, 2, ..., hi. If α = 0, it means that the pixel feature values ​​are consistent, and they are classified into one category and identified with the same color. Conversely, if α = 1, it means that the pixel feature values ​​are different and are identified with different colors. The LBP feature algorithm is a conventional technique for those skilled in the art, so it will not be described in detail in this application.

[0019] S320: For each color-marked image feature data, construct a two-dimensional planar coordinate system, forming m image coordinate sets: L(j) = {(x1,y1),(x2,y2),…,(xz,yz)}, where j = 1,2,…,m, and (x1,y1),(x2,y2),…,(xz,yz) represent the coordinates of the 1st, 2nd,…,zth pixels of any j-th image; traverse the m image coordinate sets L(j) and determine the distance between any two feature points (xγ,yγ) and (xδ,yδ) of the same color: if Where β represents the distance threshold, the feature points of the same color are merged and displayed as a new feature point of the same color with the center of the region.

[0020] S330: Based on the m image coordinate sets after feature point fusion, the image feature data at the m-th time point is extracted. For the color identifier of the image feature points, feature points in the other m-1 images that do not have the same color identifier are removed, which helps reduce subsequent system calculations and improves computational efficiency. Furthermore, an overlap algorithm is used to sequentially overlay the image data, ultimately forming a new image coordinate set Ls. The overlap algorithm is a conventional technique used by those skilled in the art, and therefore will not be described in detail in this application.

[0021] S340: The process of obtaining image overlay involves using a fitting algorithm to fit the coordinates of feature points with the same color markings in different images, forming U feature point trajectory sets: Ls = {l1, l2, ..., lU}, where l1, l2, ..., lU represent the trajectories of the 1st, 2nd, ..., Uth feature points; based on any feature point trajectory Lp in the feature point trajectory set Ls, the similarity between Lp and other feature point trajectories is obtained: λp = |Lp∩Lp+e| / |Lp∪Lp+e|, e = 1, 2, ..., Up; wherein, the fitting algorithm can calculate the trajectory route, preparing for subsequent dynamic and static analysis of objects;

[0022] S350: Analyze trajectory similarity to determine whether the state of each image feature point is static or dynamic: If there is no similarity λp>η, where η represents the similarity threshold, it means that the feature point trajectory Lp is not similar to other feature point trajectories, indicating that the movement trajectory of the feature point has regularity, then the state of the feature point is determined to be dynamic and marked as a dynamic feature point; conversely, if there is a similarity λp>η, it means that the feature point trajectory Lp is similar to other feature point trajectories, then the state of the feature point is determined to be static and marked as a static feature point.

[0023] Furthermore, step S400 includes:

[0024] S410: Construct a two-dimensional plane coordinate system with the center of the superimposed image as the origin, extract the data of the identified dynamic feature points in the image, and obtain u feature point trajectories; based on the fitted u feature point trajectories, determine whether each trajectory direction converges towards the y-axis. If there are no feature point trajectories that converge towards the y-axis, confirm that the current danger level of the vehicle of concern is 0; otherwise, if there are feature point trajectories that converge towards the y-axis, proceed to step S420 in order to further analyze the danger level of the vehicle of concern.

[0025] S420: Acquire the feature point trajectories that aggregate towards the y-axis to obtain s feature point trajectories, confirm the vehicle information corresponding to each feature point trajectory, set them as target vehicles, and use the trajectory length / time series m to confirm the running speed of each target vehicle;

[0026] S430: Analyze the relationship between the vehicle of interest aν and the information set A of all vehicles that have committed traffic violations in the big data network: if aν∈A, it means that the vehicle of interest has committed a traffic violation, then confirm that the current danger level of the vehicle of interest aν is s / u;

[0027] S440: If aν∉A, it means that the vehicle under interest has not committed any traffic violations. Then, obtain any feature point trajectory lg that aggregates towards the y-axis from the s feature point trajectories and the corresponding target vehicle information. At the same time, extract the information set A={a1,a2,…,an} of each vehicle that has committed a traffic violation in the big data network and the historical trajectory set B={b1,b2,…,bn}. Match the feature point trajectory lg with the historical trajectory set B, and use the matching results to analyze the danger of the target vehicle corresponding to each feature point trajectory, and further analyze the current danger level of the vehicle under interest.

[0028] Furthermore, step S440 includes:

[0029] S441: Extract any feature point trajectory lg aggregated towards the y-axis and the historical trajectory set B in the big data network. Based on any historical trajectory bf in the historical trajectory set B, where bf∈B, compare the similarity between the two trajectories: ρg=|lg∩bf| / |lg∪bf|. Traverse the historical trajectory set B. When the similarity ρg>θ, where θ is the data similarity threshold, it indicates that the feature point trajectory lg and the historical trajectory bf are similar, and record the number of similarities with the historical trajectory bf in the historical trajectory set B as εg. Based on any feature point trajectory lg and the number of similarities εg of the historical trajectories, extract the corresponding target vehicle information, set as ag. If εg / n>φ, where φ represents the quantity ratio threshold, it indicates that the target vehicle corresponding to the feature point trajectory lg in the big data network has a high number of traffic violations and is highly dangerous; conversely, if εg / n<φ, it indicates that the target vehicle corresponding to the feature point trajectory lg in the big data network has a low number of traffic violations and is less dangerous.

[0030] S442: Based on any target vehicle ag, filter εg similar historical trajectories according to the vehicle information set A: Obtain the vehicle information set Ag = {a1, a2, ..., aεg} corresponding to the εg historical trajectories, traverse Ag, and compare the similarity μg between the target vehicle ag and any vehicle information aq in the corresponding vehicle information set Ag: μg = |ag∩aq| / |ag∪aq| where aq∈Ag, filter the vehicle information with similarity μg>σ, and record the number of similar violating vehicles as ωg, where ωg≤εg; at this time, the danger of the current target vehicle is obtained as ωg / n;

[0031] S443: Based on the risk of any target vehicle ag as ωg / n, traverse the target vehicles corresponding to the trajectories of s feature points aggregated towards the y-axis. The risk analysis is the same as in steps S441 and S442, resulting in the risk set: {ω1 / n, ω2 / n, ..., ωs / n}, where ω1 / n, ω2 / n, ..., ωs / n represent the risk values ​​of the 1st, 2nd, ..., sth target vehicles;

[0032] S444: Extract the hazard values ​​of the 1st, 2nd, ..., sth target vehicles respectively, and confirm that the hazard level of the vehicle of interest aν in the current state is (s / u)*[Σ]. s g=1 ωg / (s*n)]; If the degree of danger is >(s / u) / 8, it indicates that the safety of the vehicle of concern is threatened.

[0033] Furthermore, step S500 includes:

[0034] When the analysis shows that the danger level of the vehicle of interest is greater than (s / u) / 8, an alert is issued, and the information of the target vehicle with the highest danger is displayed using intelligent voice.

[0035] A vehicle safety intelligent detection system, comprising: a data acquisition module, a database, an image analysis module, a self-learning module, and an intelligent reminder module;

[0036] The data acquisition module collects information on all vehicles that have committed traffic violations and their corresponding trajectory information from the big data network, forming a vehicle information set and a historical trajectory set. It also collects information on vehicles currently logged into the intelligent detection system, designates these logged-in vehicles as vehicles of interest, and collects image data of these vehicles over a certain period of time using a wide-angle camera, forming an image dataset.

[0037] The system obtains vehicle information sets and historical trajectory sets from the database, and acquires vehicle information and corresponding image datasets from vehicles logged into the cloud platform. The data is then encrypted and stored using a digital signature algorithm.

[0038] The image analysis module extracts the stored image dataset, analyzes the changing patterns of each image feature point at different times from each image in the image dataset, and further determines whether the state of each image feature point is static or dynamic.

[0039] The autonomous learning module fits the trajectory of the feature points identified as dynamic feature points, constructs a two-dimensional plane coordinate system, extracts feature points that aggregate towards the y-axis, matches the dynamic feature data with the historical trajectory set based on the relationship between the vehicle of interest and the vehicle information set, analyzes the danger of each dynamic feature point using the matching results, and further analyzes the degree of danger of the vehicle of interest in its current state.

[0040] The intelligent reminder module provides alerts when the vehicle under surveillance is in a high-risk situation.

[0041] Furthermore, the data acquisition module includes a historical vehicle acquisition unit and a current vehicle acquisition unit;

[0042] The historical vehicle acquisition unit is used to collect information on all vehicles that have logged into the cloud platform in the past and their trajectory information; the current vehicle acquisition unit is used to collect information on vehicles using the detection system in the current state and image data within a certain time period.

[0043] Furthermore, the image analysis module includes a feature point acquisition unit, a feature point analysis unit, and a state discrimination unit;

[0044] The feature point acquisition unit is used to extract feature information of each image acquired at different times using a feature extraction algorithm; the feature point analysis unit is used to analyze the changing patterns of feature points in each image at different times; and the state discrimination unit is used to determine whether the state of the image feature points is static or dynamic based on the image changing patterns.

[0045] Furthermore, the autonomous learning module includes a dynamic feature extraction unit, a data matching unit, and a risk assessment unit;

[0046] The dynamic feature extraction unit is used to extract data that is confirmed as dynamic feature points to obtain dynamic feature data; the data matching unit is used to match the dynamic feature data with the vehicle trajectory information of all illegal vehicles in the big data network to obtain a set of similar trajectories; the hazard assessment unit is used to combine the set of similar trajectories with the vehicle information set under historical conditions to conduct a hazard assessment and analyze the degree of danger of the vehicle of concern in the current state.

[0047] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0048] This invention utilizes the LBP feature algorithm to characterize images, analyzes the consistency of image feature points, and uses different colors to color-label different image feature points, which is beneficial for subsequent analysis and differentiation of image overlap. By constructing a two-dimensional plane coordinate system, feature points with a distance less than a threshold are fused pixel by pixel according to the distance formula. At the same time, feature points that do not have the same color label as the last image are removed. Furthermore, an overlap algorithm is used to overlay the captured images, which helps reduce subsequent system calculations and improves computational efficiency. By using similarity analysis to determine the regularity of the trajectory of feature points formed by each color, feature points without regular trajectories are identified as dynamic feature points, which serves as a way to distinguish between static and dynamic objects, and is beneficial for subsequent extraction and analysis of dynamic feature points to confirm hazards. By analyzing the relationship between the vehicle of interest and the vehicle information set, the hazard level of the vehicle of interest is roughly analyzed and screened, which helps improve the computational efficiency of the detection system. By matching dynamic feature data with historical trajectory sets, the hazard level of each dynamic feature point is analyzed using the matching results, further obtaining the hazard level of the vehicle of interest in its current state, which is beneficial for timely detection and early warning of vehicle travel safety. Attached Figure Description

[0049] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0050] Figure 1 This is a structural diagram of an intelligent vehicle safety detection system based on artificial intelligence according to the present invention;

[0051] Figure 2 This is a flowchart of an intelligent vehicle safety detection method based on artificial intelligence according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figures 1-2 This invention provides a technical solution: an intelligent vehicle safety detection method based on artificial intelligence, comprising the following steps:

[0054] S100: Collect information on all vehicles that have committed traffic violations and their corresponding trajectory information from the big data network to form a vehicle information set and a historical trajectory set. Collect information on vehicles that are currently logged into the intelligent detection system, set the vehicles logged into the intelligent detection system as vehicles to be monitored, and collect image data of the vehicles to be monitored within a certain time period using a wide-angle camera to form an image dataset.

[0055] Step S100 includes:

[0056] S110: Collect information on vehicles that have committed traffic violations and their corresponding trajectory information from the big data network, forming a vehicle information set A = {a1, a2, ..., an} and a historical trajectory set B = {b1, b2, ..., bn}, where a1, a2, ..., an represent the information of the 1st, 2nd, ..., nth vehicles, and b1, b2, ..., bn represent all historical trajectory information of the 1st, 2nd, ..., nth vehicles; the vehicle information for violations includes information on vehicles that violated traffic rules but did not cause an accident and information on vehicles that violated traffic rules and caused an accident; the vehicle information set includes vehicle type, speed standard, etc.; the historical trajectory set includes vehicle speed and safety status when traveling on each trajectory.

[0057] S120: When the intelligent detection system identifies a vehicle of interest (aν) logging in, it uses the wide-angle camera of the vehicle of interest (aν) to collect image data captured at a certain time series, forming an image dataset: C={c1,c2,…,cm}, where c1,c2,…,cm represent the image data captured at the 1st, 2nd,…,mth time points.

[0058] S200: Acquire vehicle information set and historical trajectory set, and obtain vehicle information and corresponding image datasets for vehicles logged into the cloud platform, and use digital signature algorithm to encrypt and store the data;

[0059] Step S200 includes:

[0060] The system acquires vehicle information sets and historical trajectory sets, and obtains vehicle information and corresponding image datasets for logging into the cloud platform. The data is then encrypted and stored using a digital signature algorithm. The digital signature algorithm is a conventional technique used by those skilled in the art, and therefore will not be described in detail in this application.

[0061] S300: Extract the stored image dataset, analyze the change pattern of each image feature point at different times from each image in the image dataset, and further determine whether the state of each image feature point is static or dynamic;

[0062] Step S300 includes:

[0063] S310: Extract the stored image dataset C of the vehicle of interest aν, and use the LBP feature algorithm to feature m images in the image dataset C; obtain the feature values ​​of the m feature images to form a pixel feature set D = {d1, d2, ..., dh}, where d1, d2, ..., dh represent the feature values ​​of the 1st, 2nd, ..., hth pixels in the m images; based on any pixel di in the pixel feature set D, obtain the pixel consistency formula: α = 1 - 1 / (d(i+k) - di), where k = 1, 2, ..., hi. If α = 0, it means that the pixel feature values ​​are consistent, and they are classified into one category and identified by the same color, such as "green, blue, red, etc."; conversely, if α = 1, it means that the pixel feature values ​​are different and are identified by different colors. The LBP feature algorithm is a conventional technique for those skilled in the art, so it will not be described in detail in this application.

[0064] By using the LBP feature algorithm to characterize images, analyzing the consistency of image feature points, and using different colors to color-label different image feature points, it is beneficial for subsequent analysis and differentiation of image overlap.

[0065] S320: For each color-marked image feature data, construct a two-dimensional planar coordinate system, forming m image coordinate sets: L(j) = {(x1,y1),(x2,y2),…,(xz,yz)}, where j = 1,2,…,m, and (x1,y1),(x2,y2),…,(xz,yz) represent the coordinates of the 1st, 2nd,…,zth pixels of any j-th image; traverse the m image coordinate sets L(j) and determine the distance between any two feature points (xγ,yγ) and (xδ,yδ) of the same color: if Where β represents the distance threshold, the feature points of the same color are merged and displayed as a new feature point of the same color with the center of the region.

[0066] S330: Based on the m image coordinate sets after feature point fusion, the image feature data at the m-th time point is extracted. Feature points in the previous m-1 images that do not have the same color identifier are removed based on the color identifier of the image feature points. This helps reduce subsequent system computation and improves computational efficiency. Furthermore, an overlap algorithm is used to sequentially overlay the image data to finally form a new image coordinate set Ls. The color identifier is selected based on a color library of "blue, green, red, pink…". The overlap algorithm is a conventional technique used by those skilled in the art, and therefore will not be elaborated upon further in this application.

[0067] By constructing a two-dimensional plane coordinate system, feature points with a distance less than a threshold are fused pixel by pixel according to the distance formula. At the same time, feature points that do not have the same color identifier as the last image are removed. Furthermore, the images are overlapped using an overlap algorithm, which helps to reduce the computation of the subsequent system and improve computational efficiency.

[0068] S340: The process of obtaining image overlay involves using a fitting algorithm to fit the coordinates of feature points with the same color markings in different images, forming U feature point trajectory sets: Ls = {l1, l2, ..., lU}, where l1, l2, ..., lU represent the trajectories of the 1st, 2nd, ..., Uth feature points; based on any feature point trajectory Lp in the feature point trajectory set Ls, the similarity between Lp and other feature point trajectories is obtained: λp = |Lp∩Lp+e| / |Lp∪Lp+e|, e = 1, 2, ..., Up; wherein, the fitting algorithm can calculate the trajectory route, preparing for subsequent dynamic and static analysis of objects;

[0069] S350: Analyze trajectory similarity to determine whether the state of each image feature point is static or dynamic: If there is no similarity λp>η, where η represents the similarity threshold, it means that the feature point trajectory Lp is not similar to other feature point trajectories, indicating that the movement trajectory of the feature point has regularity, then the state of the feature point is determined to be dynamic and marked as a dynamic feature point; conversely, if there is a similarity λp>η, it means that the feature point trajectory Lp is similar to other feature point trajectories, then the state of the feature point is determined to be static and marked as a static feature point.

[0070] By using similarity analysis to identify the regularity of the trajectory of feature points formed by various colors, feature points without regular trajectories are identified as dynamic feature points. This serves as a way to distinguish between static and dynamic objects, which is beneficial for the subsequent extraction and analysis of dynamic feature points and the confirmation of hazards.

[0071] S400: Fit the trajectory of the feature points identified as dynamic feature points to construct a two-dimensional plane coordinate system, extract feature points aggregated towards the y-axis, match the dynamic feature data with the historical trajectory set based on the relationship between the vehicle of interest and the vehicle information set, analyze whether the vehicle information belongs to the vehicle information set, if it does, issue a warning for data with a high risk level according to rough calculation, otherwise continue matching; use the matching results to analyze the risk level of each dynamic feature point, and further analyze the risk level of the vehicle of interest in its current state; first, perform autonomous screening to filter out those with a low risk level according to rough calculation, such as dynamic feature points that are far away from the vehicle of interest; then, match the data with higher risks to further accurately analyze the risk level.

[0072] Step S400 includes:

[0073] S410: Construct a two-dimensional plane coordinate system with the center of the superimposed image as the origin, extract the data of the identified dynamic feature points in the image, and obtain u feature point trajectories; based on the fitted u feature point trajectories, determine whether each trajectory direction converges towards the y-axis. If there are no feature point trajectories that converge towards the y-axis, confirm that the current danger level of the vehicle of concern is 0; otherwise, if there are feature point trajectories that converge towards the y-axis, proceed to step S420 in order to further analyze the danger level of the vehicle of concern.

[0074] S420: Acquire the feature point trajectories that aggregate towards the y-axis to obtain s feature point trajectories, confirm the vehicle information corresponding to each feature point trajectory, set them as target vehicles, and use the trajectory length / time series m to confirm the running speed of each target vehicle;

[0075] S430: Analyze the relationship between the vehicle of interest aν and the information set A of all vehicles that have committed traffic violations in the big data network: if aν∈A, it means that the vehicle of interest has committed a traffic violation, then confirm that the current danger level of the vehicle of interest aν is s / u;

[0076] S440: If aν∉A, it means that the vehicle under interest has not committed any traffic violations. Then, obtain any feature point trajectory lg that aggregates towards the y-axis from the s feature point trajectories and the corresponding target vehicle information. At the same time, extract the information set A={a1,a2,…,an} of each vehicle that has committed a traffic violation in the big data network and the historical trajectory set B={b1,b2,…,bn}. Match the feature point trajectory lg with the historical trajectory set B, and use the matching results to analyze the danger of the target vehicle corresponding to each feature point trajectory, and further analyze the current danger level of the vehicle under interest.

[0077] By analyzing the relationship between the vehicles of interest and the vehicle information set, a rough analysis of the danger level of the vehicles of interest can be performed and a screening can be conducted, which helps to improve the computational efficiency of the detection system.

[0078] Step S440 includes:

[0079] S441: Extract any feature point trajectory lg aggregated towards the y-axis and the historical trajectory set B in the big data network. Based on any historical trajectory bf in the historical trajectory set B, where bf∈B, compare the similarity between the two trajectories: ρg=|lg∩bf| / |lg∪bf|. Traverse the historical trajectory set B. When the similarity ρg>θ, where θ is the data similarity threshold, it indicates that the feature point trajectory lg and the historical trajectory bf are similar, and record the number of similarities with the historical trajectory bf in the historical trajectory set B as εg. Based on any feature point trajectory lg and the number of similarities εg of the historical trajectories, extract the corresponding target vehicle information, set as ag. If εg / n>φ, where φ represents the quantity ratio threshold, it indicates that the target vehicle corresponding to the feature point trajectory lg in the big data network has a high number of traffic violations and is highly dangerous; conversely, if εg / n<φ, it indicates that the target vehicle corresponding to the feature point trajectory lg in the big data network has a low number of traffic violations and is less dangerous.

[0080] S442: Based on any target vehicle ag, filter εg similar historical trajectories according to the vehicle information set A: Obtain the vehicle information set Ag = {a1, a2, ..., aεg} corresponding to the εg historical trajectories, traverse Ag, and compare the similarity μg between the target vehicle ag and any vehicle information aq in the corresponding vehicle information set Ag: μg = |ag∩aq| / |ag∪aq| where aq∈Ag, filter the vehicle information with similarity μg>σ, and record the number of similar violating vehicles as ωg, where ωg≤εg; at this time, the danger of the current target vehicle is obtained as ωg / n;

[0081] S443: Based on the risk of any target vehicle ag as ωg / n, traverse the target vehicles corresponding to the trajectories of s feature points aggregated towards the y-axis. The risk analysis is the same as in steps S441 and S442, resulting in the risk set: {ω1 / n, ω2 / n, ..., ωs / n}, where ω1 / n, ω2 / n, ..., ωs / n represent the risk values ​​of the 1st, 2nd, ..., sth target vehicles;

[0082] S444: Extract the hazard values ​​of the 1st, 2nd, ..., sth target vehicles respectively, and confirm that the hazard level of the vehicle of interest aν in the current state is (s / u)*[Σ]. s g=1 ωg / (s*n)]; If the degree of danger is >(s / u) / 8, it indicates that the safety of the vehicle of concern is threatened.

[0083] S500: Issue a warning when the danger level of the vehicle under concern exceeds a threshold;

[0084] Step S500 includes:

[0085] When the analysis shows that the danger level of the vehicle of interest is greater than (s / u) / 8, an alert is issued, and the information of the target vehicle with the highest danger is displayed using intelligent voice.

[0086] By matching dynamic feature data with historical trajectory sets, and using the matching results to analyze the danger of each dynamic feature point, the degree of danger of the vehicle of interest in its current state can be obtained, which is conducive to timely detection and early warning of vehicle travel safety.

[0087] A vehicle safety intelligent detection system, comprising: a data acquisition module, a database, an image analysis module, a self-learning module, and an intelligent reminder module;

[0088] The data acquisition module collects information on all vehicles that have committed traffic violations and their corresponding trajectory information from the big data network, forming a vehicle information set and a historical trajectory set. It also collects information on vehicles currently logged into the intelligent detection system, designates these logged-in vehicles as vehicles of interest, and collects image data of these vehicles over a certain period of time using a wide-angle camera, forming an image dataset.

[0089] The system obtains vehicle information sets and historical trajectory sets from the database, and acquires vehicle information and corresponding image datasets from vehicles logged into the cloud platform. The data is then encrypted and stored using a digital signature algorithm.

[0090] The image analysis module extracts the stored image dataset, analyzes the changing patterns of each image feature point at different times from each image in the image dataset, and further determines whether the state of each image feature point is static or dynamic.

[0091] The autonomous learning module fits the trajectory of the feature points identified as dynamic feature points, constructs a two-dimensional plane coordinate system, extracts feature points that aggregate towards the y-axis, matches the dynamic feature data with the historical trajectory set based on the relationship between the vehicle of interest and the vehicle information set, analyzes the danger of each dynamic feature point using the matching results, and further analyzes the degree of danger of the vehicle of interest in its current state.

[0092] The intelligent reminder module provides alerts when the danger level of the vehicle under surveillance exceeds a threshold.

[0093] The data acquisition module includes a historical vehicle acquisition unit and a current vehicle acquisition unit;

[0094] The historical vehicle acquisition unit is used to collect information on all vehicles that have logged into the cloud platform in the past and their trajectory information; the current vehicle acquisition unit is used to collect information on vehicles using the detection system in the current state and image data within a certain time period.

[0095] The image analysis module includes a feature point acquisition unit, a feature point analysis unit, and a state discrimination unit;

[0096] The feature point acquisition unit is used to extract feature information of each image acquired at different times using a feature extraction algorithm; the feature point analysis unit is used to analyze the changing patterns of feature points in each image at different times; and the state discrimination unit is used to determine whether the state of the image feature points is static or dynamic based on the image changing patterns.

[0097] The autonomous learning module includes a dynamic feature extraction unit, a data matching unit, and a risk assessment unit;

[0098] The dynamic feature extraction unit is used to extract data that is confirmed as dynamic feature points to obtain dynamic feature data; the data matching unit is used to match the dynamic feature data with the vehicle trajectory information of all illegal vehicles in the big data network to obtain a set of similar trajectories; the hazard assessment unit is used to combine the set of similar trajectories with the vehicle information set under historical conditions to conduct a hazard assessment and analyze the degree of danger of the vehicle of concern in the current state.

[0099] Example 1:

[0100] In step S100:

[0101] S110: Collect information on vehicles that have committed traffic violations and their corresponding trajectory information from the big data network, forming a vehicle information set A = {a1, a2, ..., a5000} and a historical trajectory set B = {b1, b2, ..., b5000}, where a1, a2, ..., a5000 represent the information of the 1st, 2nd, ..., 5000th vehicles, and b1, b2, ..., b5000 represent all historical trajectory information of the 1st, 2nd, ..., 5000th vehicles; wherein, the information on vehicles that committed traffic violations includes information on vehicles that committed violations but did not have accidents and information on vehicles that committed violations and had accidents; the vehicle information set includes vehicle type, speed standard, etc.; the historical trajectory set includes vehicle speed and safety status when traveling on each trajectory.

[0102] S120: When the intelligent detection system identifies a vehicle of interest (aν) logging in, it uses the wide-angle camera of the vehicle of interest (aν) to collect image data captured at a certain time series, forming an image dataset: C={c1,c2,…,c10}, where c1,c2,…,c10 represent the image data captured at the 1st, 2nd,…, 10th time points.

[0103] In step S200:

[0104] The system acquires vehicle information sets and historical trajectory sets, as well as vehicle information and corresponding image datasets for vehicles logged into the cloud platform. It then uses a digital signature algorithm to encrypt and store the data.

[0105] In step S300:

[0106] S310: Extract the stored image dataset C of the vehicle of interest aν, and use the LBP feature algorithm to feature 10 images in the image dataset C; obtain the feature values ​​of the 10 featured images to form a pixel feature set D = {d1, d2, ..., d1000}, where d1, d2, ..., d1000 represent the feature values ​​of the 1st, 2nd, ..., 1000th pixels in the 10 images; based on any pixel di in the pixel feature set D, obtain the pixel consistency formula: α = 1 - 1 / (d(i+k) - di), where k = 1, 2, ..., 1000 - i. If α = 0, it means that the pixel feature values ​​are consistent, and they are classified into one category and identified with the same color; conversely, if α = 1, it means that the pixel feature values ​​are different and are identified with different colors; wherein, the color identification is selected according to the color library "blue, green, red, pink...".

[0107] S320: Construct a two-dimensional plane coordinate system for each color-marked image feature data, forming 10 image coordinate sets: L(j) = {(x1,y1),(x2,y2),…,(x100,y100)}, where j = 1,2,…,10, and (x1,y1),(x2,y2),…,(x100,y100) represent the coordinates of the 1st, 2nd,…,100th pixels of any j-th image; traverse the 10 image coordinate sets L(j) and determine the distance between any two feature points (xγ,yγ) and (xδ,yδ) of the same color: If If the distance is less than 0.5cm, then all feature points of the same color will be merged and displayed as a new feature point of the same color with the center of the region.

[0108] S330: Based on the 10 image coordinate sets after feature point fusion, the image feature data at the 10th time point is extracted. Feature points that do not have the same color identifier in the other 9 images are removed based on the color identifier of the image feature points. Then, the image data are superimposed sequentially using the overlap algorithm to finally form a new image coordinate set Ls.

[0109] S340: The process of obtaining image overlay involves using a fitting algorithm to fit the coordinates of feature points with the same color in different images, forming five feature point trajectory sets: Ls = {l1, l2, ..., l5}, where l1, l2, ..., l5 represent the trajectories of the 1st, 2nd, ..., 5th feature points; based on any feature point trajectory Lp in the feature point trajectory set Ls, the similarity between Lp and other feature point trajectories is obtained: λp = |Lp∩Lp+e| / |Lp∪Lp+e|, e = 1, 2, ..., 5-p;

[0110] S350: Analyze trajectory similarity to determine whether the state of each image feature point is static or dynamic: Traverse the trajectories of 5 feature points. If there is no similarity λp>0.85, it means that the feature point trajectory Lp is not similar to the other feature point trajectories, indicating that the movement trajectory of the feature point has regularity. In this case, the state of the feature point is determined to be dynamic and it is marked as a dynamic feature point. Conversely, if there is a similarity λp>0.85, it means that the feature point trajectory Lp is similar to the other feature point trajectories. In this case, the state of the feature point is determined to be static and it is marked as a static feature point.

[0111] In step S400:

[0112] S410: Using the center of the superimposed image as the origin, construct a two-dimensional plane coordinate system, extract the data of the identified dynamic feature points in the image, and obtain the trajectory of 3 feature points; based on the fitted trajectory of 3 feature points, determine whether each trajectory direction converges towards the y-axis. If there is a feature point trajectory that converges towards the y-axis, proceed to step S420 in order to further analyze the degree of danger of the vehicle of concern.

[0113] S420: Acquire the feature point trajectory of the aggregation towards the y-axis to obtain one feature point trajectory l2. Confirm the vehicle information corresponding to each feature point trajectory, set it as the target vehicle, and use 5 / 10=0.5cm / ms to confirm that the running speed of the target vehicle is 0.5cm / ms, and further convert it into the actual vehicle speed.

[0114] S430: Analyze the relationship between the vehicle of interest aν and the information set A of all vehicles that have committed traffic violations in the big data network: aν∉A, indicating that the vehicle of interest has not committed any traffic violations. Then, obtain the feature point trajectory l2 aggregated along the y-axis and the corresponding target vehicle information. At the same time, extract the information sets A={a1,a2,…,a5000} and the historical trajectory set B={b1,b2,…,b5000} of each vehicle that has committed traffic violations in the big data network. Match the feature point trajectory l2 and the historical trajectory set B. Use the matching results to analyze the danger of the target vehicle corresponding to each feature point trajectory, and further analyze the current danger level of the vehicle of interest.

[0115] In step S440:

[0116] S441: Extract the feature point trajectory l2 and the historical trajectory set B in the big data network. Based on any historical trajectory bf in the historical trajectory set B, where bf∈B, compare the similarity between the two trajectories: ρg=|l2∩bf| / |l2∪bf|. Traverse the historical trajectory set B. When the similarity ρg>0.7, it means that the feature point trajectory lg and the historical trajectory bf are similar, and record the number of similarities with the historical trajectory bf in the historical trajectory set B as 40. Based on the number of similarities between any feature point trajectory l2 and the historical trajectory 40, extract the corresponding target vehicle information and set it as ag. If 40 / 5000<0.02, it means that the target vehicle corresponding to the feature point trajectory l2 in the big data network has a low number of traffic violation events and low risk.

[0117] S442: Based on any target vehicle ag, filter 40 similar historical trajectories according to the vehicle information set A: Obtain the vehicle information set Ag = {a1, a2, ..., a40} corresponding to the 40 historical trajectories, traverse Ag, and compare the similarity μg between the target vehicle ag and any vehicle information aq in the corresponding vehicle information set Ag: μg = |ag∩aq| / |ag∪aq| where aq∈Ag, filter the vehicle information with similarity μg>σ, and record the number of similar violating vehicles as 10. At this time, the danger of the current target vehicle is 10 / 5000 = 0.002;

[0118] S444: Extract the danger value of the target vehicle and confirm that the danger level of the vehicle of interest aν in the current state is (1 / 3)*0.002; if the danger level is >1 / 24, it means that the safety of the vehicle of interest is threatened.

[0119] In step S500:

[0120] When the analysis indicates that the danger level of the vehicle of interest is greater than 1 / 24, an alert will be issued, and the information of the target vehicle with the highest danger level will be displayed using intelligent voice.

[0121] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0122] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent vehicle safety detection method based on artificial intelligence, characterized in that: Includes the following steps: S100: Collect information on all vehicles that have committed traffic violations and their corresponding trajectory information from the big data network to form a vehicle information set and a historical trajectory set. Collect information on vehicles that are currently logged into the intelligent detection system, set the vehicles logged into the intelligent detection system as vehicles to be monitored, and collect image data of the vehicles to be monitored within a certain time period using a wide-angle camera to form an image dataset. S200: Acquire vehicle information set and historical trajectory set, and obtain vehicle information and corresponding image datasets for vehicles logged into the cloud platform, and use digital signature algorithm to encrypt and store the data; S300: Extract the stored image dataset, analyze the change pattern of each image feature point at different times from each image in the image dataset, and further determine whether the state of each image feature point is static or dynamic; S400: Fit the trajectory of the feature points identified as dynamic feature points, construct a two-dimensional plane coordinate system, extract the feature points aggregated towards the y-axis, match the dynamic feature data with the historical trajectory set according to the relationship between the vehicle of interest and the vehicle information set, analyze the danger of each dynamic feature point using the matching results, and further analyze the degree of danger of the vehicle of interest in its current state. S420: Acquire the feature point trajectories that aggregate towards the y-axis to obtain s feature point trajectories, confirm the vehicle information corresponding to each feature point trajectory, set them as target vehicles, and use the trajectory length / time series m to confirm the running speed of each target vehicle; Step S400 includes: S410: Construct a two-dimensional plane coordinate system with the center of the superimposed image as the origin, extract the data of the identified dynamic feature points in the image, and obtain u feature point trajectories; based on the fitted u feature point trajectories, determine whether each trajectory direction converges towards the y-axis. If there are no feature point trajectories that converge towards the y-axis, confirm that the current danger level of the vehicle of concern is 0; otherwise, if there are feature point trajectories that converge towards the y-axis, proceed to step S420 in order to further analyze the danger level of the vehicle of concern. S500: Issue a warning when the danger level of the vehicle under concern exceeds a threshold.

2. The intelligent vehicle safety detection method based on artificial intelligence according to claim 1, characterized in that: Step S100 includes: S110: Collect information on each vehicle that has committed a traffic violation and its corresponding trajectory information from the big data network, forming a vehicle information set A = {a1, a2, ..., an} and a historical trajectory set B = {b1, b2, ..., bn}, where a1, a2, ..., an represent the information of the 1st, 2nd, ..., nth vehicles, and b1, b2, ..., bn represent all the historical trajectory information of the 1st, 2nd, ..., nth vehicles; S120: When the intelligent detection system identifies a vehicle of interest (aν) logging in, it uses the wide-angle camera of the vehicle of interest (aν) to collect image data captured at a certain time series, forming an image dataset: C={c1,c2,…,cm}, where c1,c2,…,cm represent the image data captured at the 1st, 2nd,…,mth time points.

3. The intelligent vehicle safety detection method based on artificial intelligence according to claim 2, characterized in that: Step S300 includes: S310: Extract the stored image dataset C of the vehicle of interest aν, and use the LBP feature algorithm to feature the m images in the image dataset C; obtain the feature values ​​of the m feature images to form a pixel feature set D={d1,d2,…,dh}, where d1,d2,…,dh represent the feature values ​​of the 1st, 2nd,…,hth pixels in the m images; based on any pixel di in the pixel feature set D, obtain the pixel consistency formula: α=1-! (d(i+k)-di), where k=1,2,…,hi. If α=0, it means that the pixel feature values ​​are consistent, and they are classified into one category and identified with the same color. Conversely, if α=1, it means that the pixel feature values ​​are different and are identified with different colors. S320: For each color-marked image feature data, construct a two-dimensional planar coordinate system, forming m image coordinate sets: L(j) = {(x1,y1),(x2,y2),…,(xz,yz)}, where j = 1,2,…,m, and (x1,y1),(x2,y2),…,(xz,yz) represent the coordinates of the 1st, 2nd,…,zth pixels of any j-th image; traverse the m image coordinate sets L(j) and determine the distance between any two feature points (xγ,yγ) and (xδ,yδ) of the same color: if Where β represents the distance threshold, the feature points of the same color are merged and displayed as a new feature point of the same color with the center of the region. S330: Based on the m image coordinate sets after feature point fusion, the image feature data at the m-th time point is extracted. Based on the color identifier of the image feature points, feature points that do not have the same color identifier in the other m-1 images are removed. Furthermore, the image data are superimposed sequentially using an overlap algorithm to finally form a new image coordinate set Ls. S340: The process of obtaining image overlay involves using a fitting algorithm to fit the coordinates of feature points with the same color in different images, forming U feature point trajectory sets: Ls = {l1, l2, ..., lU}, where l1, l2, ..., lU represent the trajectories of the 1st, 2nd, ..., Uth feature points; based on any feature point trajectory Lp in the feature point trajectory set Ls, the similarity between Lp and other feature point trajectories is obtained: λp = |Lp∩Lp+e| / |Lp∪Lp+e|, e = 1, 2, ..., Up; S350: Analyze trajectory similarity to determine whether the state of each image feature point is static or dynamic: If there is no similarity λp>η, where η represents the similarity threshold, it means that the feature point trajectory Lp is not similar to other feature point trajectories, then the state of the feature point is determined to be dynamic and marked as a dynamic feature point; otherwise, if there is a similarity λp>η, then it is marked as a static feature point.

4. The intelligent vehicle safety detection method based on artificial intelligence according to claim 3, characterized in that: Step S400 includes: S430: Analyze the relationship between the vehicle of interest aν and the information set A of all vehicles that have committed traffic violations in the big data network: if aν∈A, it means that the vehicle of interest has committed a traffic violation, then confirm that the current danger level of the vehicle of interest aν is s / u; S440: If aν∉A, it means that the vehicle of interest has not committed any traffic violations. Then, obtain any feature point trajectory lg that aggregates towards the y-axis from the s feature point trajectories and the corresponding target vehicle information. At the same time, extract the information set A={a1,a2,…,an} of each vehicle that has committed a traffic violation in the big data network and the historical trajectory set B={b1,b2,…,bn}. Match the feature point trajectory lg with the historical trajectory set B, and use the matching results to analyze the danger of the target vehicle corresponding to each feature point trajectory, and further analyze the current danger level of the vehicle of interest.

5. The intelligent vehicle safety detection method based on artificial intelligence according to claim 4, characterized in that: Step S440 includes: S441: Extract any feature point trajectory lg aggregated towards the y-axis and the historical trajectory set B in the big data network. Based on any historical trajectory bf in the historical trajectory set B, where bf∈B, compare the similarity between the two trajectories: ρg=|lg∩bf| / |lg∪bf|. Traverse the historical trajectory set B. When the similarity ρg>θ, where θ is the data similarity threshold, it indicates that the feature point trajectory lg and the historical trajectory bf are similar, and record the number of similarities with the historical trajectory bf in the historical trajectory set B as εg. Based on any feature point trajectory lg and the number of similarities εg of the historical trajectories, extract the corresponding target vehicle information, set as ag. If εg / n>φ, where φ represents the quantity ratio threshold, it indicates that the target vehicle corresponding to the feature point trajectory lg in the big data network has a high number of traffic violations and is highly dangerous; conversely, if εg / n<φ, it indicates that the target vehicle corresponding to the feature point trajectory lg in the big data network has a low number of traffic violations and is less dangerous. S442: Based on any target vehicle ag, filter εg similar historical trajectories according to the vehicle information set A: Obtain the vehicle information set Ag = {a1, a2, ..., aεg} corresponding to the εg historical trajectories, traverse Ag, and compare the similarity μg between the target vehicle ag and any vehicle information aq in the corresponding vehicle information set Ag: μg = |ag∩aq| / |ag∪aq| where aq∈Ag, filter the vehicle information with similarity μg>σ, and record the number of similar violating vehicles as ωg, where ωg≤εg; at this time, the danger of the current target vehicle is obtained as ωg / n; S443: Based on the risk of any target vehicle ag as ωg / n, traverse the target vehicles corresponding to the trajectories of s feature points aggregated towards the y-axis. The risk analysis is the same as in steps S441 and S442, resulting in the risk set: {ω1 / n, ω2 / n, ..., ωs / n}, where ω1 / n, ω2 / n, ..., ωs / n represent the risk values ​​of the 1st, 2nd, ..., sth target vehicles; S444: Extract the hazard values ​​of the 1st, 2nd, ..., sth target vehicles respectively, and confirm that the hazard level of the vehicle of interest aν in the current state is (s / u)*[Σ]. s g=1 ωg / (s*n)]; If the degree of danger is >(s / u) / 8, it indicates that the safety of the vehicle of concern is threatened.

6. A driving safety intelligent detection system that implements the driving safety intelligent detection method based on artificial intelligence according to any one of claims 1-5, characterized in that: The system includes: a data acquisition module, a database, an image analysis module, a self-learning module, and an intelligent reminder module; The data acquisition module collects information on all vehicles that have committed traffic violations and their corresponding trajectory information from the big data network, forming a vehicle information set and a historical trajectory set. It also collects information on vehicles currently logged into the intelligent detection system, designates these logged-in vehicles as vehicles of interest, and collects image data of these vehicles over a certain period of time using a wide-angle camera, forming an image dataset. The system obtains vehicle information sets and historical trajectory sets from the database, and acquires vehicle information and corresponding image datasets from vehicles logged into the cloud platform. The data is then encrypted and stored using a digital signature algorithm. The image analysis module extracts the stored image dataset, analyzes the changing patterns of each image feature point at different times from each image in the image dataset, and further determines whether the state of each image feature point is static or dynamic. The autonomous learning module fits the trajectory of the feature points identified as dynamic feature points, constructs a two-dimensional plane coordinate system, extracts feature points that aggregate towards the y-axis, matches the dynamic feature data with the historical trajectory set based on the relationship between the vehicle of interest and the vehicle information set, analyzes the danger of each dynamic feature point using the matching results, and further analyzes the degree of danger of the vehicle of interest in its current state. The intelligent reminder module provides alerts when the danger level of the vehicle under surveillance exceeds a threshold.

7. The intelligent vehicle safety detection system according to claim 6, characterized in that: The data acquisition module includes a historical vehicle acquisition unit and a current vehicle acquisition unit; The historical vehicle acquisition unit is used to collect information on all vehicles that have logged into the cloud platform in the past and their trajectory information; the current vehicle acquisition unit is used to collect information on vehicles using the detection system in the current state and image data within a certain time period.

8. The intelligent vehicle safety detection system according to claim 6, characterized in that: The image analysis module includes a feature point acquisition unit, a feature point analysis unit, and a state discrimination unit; The feature point acquisition unit is used to extract feature information of each image acquired at different times using a feature extraction algorithm; the feature point analysis unit is used to analyze the changing patterns of feature points of each image at different times; and the state discrimination unit is used to determine whether the state of the image feature points is static or dynamic.

9. The intelligent vehicle safety detection system according to claim 6, characterized in that: The autonomous learning module includes a dynamic feature extraction unit, a data matching unit, and a risk assessment unit; The dynamic feature extraction unit is used to extract data that is confirmed as dynamic feature points to obtain dynamic feature data; the data matching unit is used to match the dynamic feature data with the vehicle trajectory information of all illegal vehicles in the big data network to obtain a set of similar trajectories; the hazard assessment unit is used to combine the set of similar trajectories with the vehicle information set under historical conditions to conduct a hazard assessment and analyze the degree of danger of the vehicle of concern in the current state.

Images