Road seven-in-one intelligent system
Through the seven-in-one intelligent system of roads, rapid response and accurate judgment of highway traffic safety issues are achieved, the level of safety and management intelligence is improved, and the problems of frequent traffic accidents and lagging emergency response are solved.
Patent Information
- Application Number
- CN202510837767.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-23
- Publication Date
- 2025-07-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Highway traffic safety issues include frequent traffic accidents, lagging emergency response, and poor information transmission, resulting in low safety and operational efficiency, affecting the safety of drivers and passengers and road management.
The road seven-in-one intelligent system is adopted to obtain automatic trigger and manual trigger data through the data acquisition module, and feature vectorization and abnormal event detection are carried out in combination with the data intelligent processing and analysis module, and an auxiliary optimization of static target classification model is built to achieve rapid response and accurate judgment.
It greatly reduces alarm time, improves road safety guarantee efficiency and intelligent management, and provides a more reliable basis for safety decision-making.
Smart Images
Figure CN120356339A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of road safety, and in particular to a seven-in-one road intelligent system. Background Art
[0002] With the rapid development of the economy, the highway network has been rapidly expanded and has become a key infrastructure connecting cities and promoting the process of regional economic integration. While highways play a vital role in transportation, their rapid development has also caused a series of traffic safety issues that cannot be underestimated. Among them, traffic accidents not only cause casualties and property losses, but also seriously affect the normal traffic order of roads; delayed emergency response makes it impossible to carry out rescue and treatment in a timely and effective manner after the accident, delaying the best rescue time and further aggravating the degree of harm caused by the accident; poor information transmission makes it impossible for drivers and passengers to obtain accurate road conditions information and emergency instructions in time, increasing the uncertainty and danger of driving. The existence of these problems not only greatly reduces the safety and operation efficiency of highways, but also poses a potential threat to the life safety of drivers and passengers, becoming a bottleneck restricting the sustainable development of highways.
[0003] In order to provide drivers and passengers with a safer, more convenient and more efficient travel environment and promote the intelligent development of the highway industry and the improvement of safety management level, a seven-in-one road intelligent system is now provided. Summary of the invention
[0004] In order to achieve the above object, the present invention provides the following technical solution: a seven-in-one road intelligent system, comprising: Data acquisition module, used to collect automatic trigger data, manual trigger data and equipment status data; The data intelligent processing module is used to pre-process the automatic trigger data and the manual trigger data to obtain the pre-processed automatic trigger data and the pre-processed manual trigger data; perform feature vectorization processing on the pre-processed automatic trigger data to obtain the automatic trigger feature vector; perform feature extraction processing on the pre-processed manual trigger data to obtain the manual trigger feature vector; The data intelligent analysis module is used to detect abnormal events on the automatic trigger feature vectors to obtain automatic trigger event instructions; perform correlation analysis on the manual trigger feature vectors to obtain manual trigger event instructions; and execute corresponding abnormal event processing solutions according to the automatic trigger event instructions and manual trigger event instructions; The intelligent optimization module is used to construct an auxiliary optimization stationary target classification model based on the automatic trigger feature vector, the manual trigger feature vector and the equipment status data, obtain the stationary target classification factor, and classify and assist in optimizing the status of the stationary target based on the stationary target classification factor.
[0005] According to one preferred embodiment of the present invention, the process of collecting automatic trigger data, manual trigger data and device status data includes: A data acquisition device is provided, which consists of a plurality of automatically triggered data acquisition units, a manually triggered data acquisition unit and an equipment status acquisition unit; the automatically triggered data acquisition unit is used to collect horizontal angles, vertical angles, reflection intensity, timestamps, target speeds, target classifications and clustering results, and is recorded as automatically triggered data; the manually triggered data acquisition unit is used to collect button duration, terminal number, user input data and reporting time, and is recorded as manually triggered data; the equipment status acquisition unit is used to collect equipment status data of the terminal.
[0006] According to one of the preferred embodiments of the present invention, the process of preprocessing the automatic trigger data and the manual trigger data includes: Obtain automatically triggered data and manually triggered data; The process of preprocessing the automatic trigger data includes: For horizontal angle, vertical angle, and reflection intensity, outlier removal operations are performed, and the distance or deviation between each data point and its adjacent points of the corresponding data is calculated. Based on statistical methods, outliers are identified and removed; outlier processing is performed on target speed and target classification. Based on the outlier processing algorithm, data points with target speed exceeding the normal range and target classification errors are removed; for data from multiple automatically triggered data acquisition units, based on the time synchronization protocol , calibrating the timestamp of the automatic triggering data acquisition unit to ensure the time consistency of the data; for the objects in the clustering results, associating the objects with different timestamps based on the target tracking algorithm to ensure the consistency of the objects in different frames; The preprocessed horizontal angle, vertical angle, reflection intensity, timestamp, target speed, target classification and clustering results are recorded as preprocessing automatic trigger data; The process of preprocessing artificial trigger data includes: Check the integrity of the data fields of button duration, terminal number, location coordinates, user input data, and reporting time; convert the reporting time into a unified time standard; if the user input data is in text form, perform text cleaning operations, including removing special characters, punctuation marks, and stop words, and perform word segmentation on the text to split it into individual words or phrases; if the user input data is in voice form, convert it into text form; The preprocessed button duration, terminal number, user input data and reporting time are recorded as preprocessed manual trigger data.
[0007] According to one preferred embodiment of the present invention, the process of performing feature vectorization on preprocessed automatic trigger data includes: Obtain the preprocessed automatic trigger data; Record the horizontal angle of the object within the acquisition range as and the vertical angle as ; According to the horizontal angle , vertical angle , generate the three-dimensional point cloud coordinates of the corresponding object ; The three-dimensional point cloud coordinates are: ; Among them, is the target distance; the target distance ; Among them, is the speed of light; is the time difference between the transmission and reception of the automatic trigger data acquisition unit; Equidistantly divide the normalization range of the reflection intensity of the object within the acquisition range into intervals, and mark the intervals according to the target classification to obtain the marked reflection intensity intervals, and each interval corresponds to the reflection intensity of a different object; According to the target speed and the clustering result, obtain the radial speed and the transverse speed ; Concatenate the three-dimensional point cloud coordinates , radial speed and transverse speed in sequence into a feature vector, denoted as the automatic trigger feature vector.
[0008] According to one preferred embodiment of the present invention, the process of obtaining the radial speed and the transverse speed according to the target speed and the clustering result includes: Obtain the three-dimensional point cloud coordinates of the object within the acquisition range for each frame, denoted as ; Among them, represents the corresponding frame, represents the number of objects corresponding to the th frame; Establish a cost matrix, which is used to associate the association cost of the objects in adjacent frames; Denote the cost matrix as , and the cost matrix represents the th object in the The association cost between the th object in the frame; The cost matrix is denoted as : ; Wherein, , , correspond to the three-dimensional point cloud coordinates of the th object in the th frame; According to the cost matrix , and based on the bipartite graph optimal matching algorithm, obtain the optimal matching result of each object in the t-th frame with the corresponding object in the th frame, denoted as , wherein, is the target index in the th frame, is the target index in the th frame, , g is the number of matched objects; According to the optimal matching result , obtain the centroid displacement ; The centroid displacement is: ; According to the centroid displacement and the time interval , obtain the velocity components in three directions of the target velocity, which are respectively: , , ; Wherein, , , respectively correspond to the velocities in the direction; The time interval is determined according to the actual sampling frequency; The unit vector corresponding to the line of sight direction of the preset automatic trigger data acquisition unit is ; The radial velocity ; The transverse velocity ; Wherein, is the target velocity.
[0009] According to one preferred embodiment of the present invention, the process of feature extraction processing for preprocessed manually triggered data includes: Perform a threshold judgment on the button duration. If the button duration does not exceed the preset duration threshold, mark it as a mis-touch and do not respond; if the button duration exceeds the preset duration threshold, obtain the corresponding terminal number, user input data, and reporting time; Hash-encode the terminal number to obtain a terminal encoding vector with a length of ; Based on recognition technology, extract features from the user input data to obtain the corresponding value, so as to obtain the user input data feature vector of the user input data; Extract time features from the reporting time, obtain the numerical features corresponding to the time, obtain the corresponding season and week according to the reporting time, and then obtain the time feature vector; Concatenate the terminal encoding vector, the user input data feature vector, and the time feature vector in sequence into a feature vector, denoted as the artificial trigger feature vector.
[0010] According to one preferred embodiment of the present invention, the process of detecting abnormal events for the automatic trigger feature vector includes: Preset the point cloud feature displacement threshold ; Collect the three-dimensional point cloud coordinates in real time to obtain the corresponding centroid displacement , if the centroid displacement is less than or equal to the cloud feature displacement threshold , mark it as a normal event; if the centroid displacement is greater than the cloud feature displacement threshold , mark it as an abnormal event; Preset the radial velocity threshold and the transverse velocity threshold ; When the radial velocity collected in real time is less than or equal to the radial velocity threshold , mark it as a normal event; if the radial velocity is greater than the radial velocity threshold , mark it as an abnormal event; When the transverse velocity collected in real time is less than or equal to the transverse velocity threshold , mark it as a normal event; if the transverse velocity is greater than the transverse velocity threshold , mark it as an abnormal event; Generate a cloud feature displacement abnormal instruction, a radial velocity abnormal instruction, and a transverse velocity abnormal instruction for the three-dimensional point cloud coordinates, radial velocity and transverse velocity corresponding to the marked abnormal event, and uniformly denote the cloud feature displacement abnormal instruction, the radial velocity abnormal instruction, and the transverse velocity abnormal instruction as the automatic trigger event instruction.
[0011] According to one preferred embodiment of the present invention, the process of performing correlation analysis on the manually triggered feature vectors includes: Mark the values corresponding to a number of user input data feature vectors as , , ……, ; where is a natural number; Obtain the correlation coefficient based on the values corresponding to the marked user input data feature vectors and the time feature vector. The correlation coefficient is: where is the mean value of the values, and represents the time feature vector; a preset correlation coefficient threshold ; If the correlation coefficient , mark it as a normal event; if the correlation coefficient , mark it as an abnormal event;
[0012] According to one preferred embodiment of the present invention, the process of executing the corresponding abnormal event handling scheme according to the automatically triggered event instruction and the manually triggered event instruction includes: The abnormal event handling scheme includes an automatic alarm handling scheme and a manual alarm handling scheme; The automatic alarm handling scheme includes a primary abnormal handling scheme and a secondary handling scheme; if the number of cloud feature displacement abnormal instructions, radial velocity abnormal instructions, and lateral velocity abnormal instructions is greater than or equal to , trigger the primary abnormal handling scheme, otherwise trigger the secondary handling scheme; The automatic alarm handling scheme includes: setting the induction lamp control logic, which includes the primary abnormal induction lamp control logic and the secondary abnormal induction lamp control logic; setting the broadcast processing logic, including the primary abnormal broadcast logic and the secondary abnormal broadcast logic; presetting the linkage mechanism, the primary abnormal induction lamp control logic and the primary abnormal broadcast logic are started synchronously, and the secondary abnormal induction lamp control logic and the secondary abnormal broadcast logic are started synchronously, forming a "visual auditory" dual warning; The manual alarm handling scheme includes: presetting the manual induction lamp control logic and the manual broadcast processing logic.
[0013] According to one preferred embodiment of the present invention, the process of constructing an auxiliary optimization static target classification model includes: Obtain several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data; Group and label several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data, denoted as where is a natural number; Take groups of several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data as sample data, and is a natural number less than . Using the sample data, based on a machine learning algorithm, obtain the sample data mean, denoted as the sample set; take the remaining several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data as the test set; According to the sample set and the test set, form a training sample set; based on a convolutional neural network, construct a standard auxiliary optimization model; and input the training sample set into the standard auxiliary optimization model to train the standard auxiliary optimization model, and denote the standard auxiliary optimization model after completion of training as the auxiliary optimization static target classification model; Generate a static target classification factor under the current conditions according to the auxiliary optimization static target classification model ; where is a weight vector, is a bias term, is a comprehensive feature vector; the comprehensive feature vector is: ; where is an automatically triggered feature vector; is a historical manually triggered feature vector; is the vector representation of the device status data; , , are the weights of the historical automatically triggered feature vector, historical manually triggered feature, and vector representation of the device status data; Preset a standard static target classification factor ; If the static target classification factor , then the status of the static target under the current conditions is normal; If the static target classification factor , then the status of the static target under the current conditions is abnormal, and an automatic alarm processing scheme is executed, greatly reducing the alarm time.
[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. The data acquisition module collects automatic trigger data, manual trigger data, and device status data, and performs preprocessing, feature vectorization, extraction, etc. operations. It analyzes and processes the automatic trigger feature vector and the manual trigger feature vector respectively, quickly obtains the corresponding event instructions, and executes the abnormal event handling plan, greatly reducing the alarm time, being able to respond to road abnormal events in a timely manner, and improving the road safety guarantee efficiency.
[0015] 2. Build an auxiliary optimization static target classification model to obtain static target classification factors, which can classify and assist in optimizing the status of static targets, helping to more accurately judge the status of static targets, providing a more reliable basis for road management and safety decision-making, and further improving the intelligent level and safety of road management. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings.
[0017] Figure 1 It is a step schematic diagram of a seven-in-one intelligent road system.
[0018] Figure 2 It is a process schematic diagram of a seven-in-one intelligent road system.
[0019] Figure 3 It is a module schematic diagram of a seven-in-one intelligent road system. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will describe the technical solutions of the present invention in detail. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other implementation manners obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present invention.
[0021] As Figure 1 shown, a seven-in-one intelligent road system includes: a data acquisition module, a data intelligent processing module, a data intelligent analysis module, and an intelligent management module; The data acquisition module is used to collect automatic trigger data, manual trigger data, and device status data; The data intelligent processing module is used to preprocess the automatically triggered data and the manually triggered data to obtain preprocessed automatically triggered data and preprocessed manually triggered data; perform feature vectorization processing on the preprocessed automatically triggered data to obtain automatically triggered feature vectors; perform feature extraction processing on the preprocessed manually triggered data to obtain manually triggered feature vectors; The data intelligent analysis module is used to detect abnormal events for the automatically triggered feature vectors to obtain automatically triggered event instructions; perform correlation analysis on the manually triggered feature vectors to obtain manually triggered event instructions; execute corresponding abnormal event handling solutions according to the automatically triggered event instructions and the manually triggered event instructions; The intelligent optimization module is used to construct an auxiliary optimization static target classification model based on the automatically triggered feature vectors and the manually triggered feature vectors to obtain static target classification factors, and classify and assist in optimizing the states of static targets according to the static target classification factors; It should be further noted that in the specific implementation process, the specific processes of collecting automatically triggered data, manually triggered data, and device status data include: Set up a data collection device, which is composed of several automatically triggered data collection units, manually triggered data collection units, and device status collection units, and is set at the corresponding data collection positions of the terminal; The automatically triggered data collection unit is used to collect the horizontal angle, vertical angle, reflection intensity, timestamp, target speed, target classification, and clustering results of the objects within the range, and record them as automatically triggered data; The manually triggered data collection unit is used to collect the button duration, terminal number, user input data, and reporting time, and record them as manually triggered data; The device status collection unit is used to collect the battery power, charging status, lidar temperature, lens pollution index, and network signal strength of the terminal, and record them as device status data.
[0022] It should be further noted that in the specific implementation process, the specific processes of preprocessing the automatically triggered data and the manually triggered data include: Obtain the automatically triggered data and the manually triggered data; The specific process of preprocessing the automatically triggered data includes: Perform outlier removal operations on the horizontal angle, vertical angle, and reflection intensity, calculate the distance or deviation between each data point of the corresponding data and its adjacent points, and identify and remove outliers based on statistical methods to ensure the accuracy of the data.
[0023] For example, there is a data sample set of reflection intensity , where is the number of samples, represents the reflection intensity of the th sample; Calculate the mean of the reflection intensity ; Calculate the standard deviation of the reflection intensity ; Set the lower limit of the reflection intensity to: ; The upper limit of the reflection intensity to: ; Exclude the reflection intensity below the lower limit and above the upper limit of the reflection intensity.
[0024] Normalize the reflection intensity, based on the normalization algorithm to normalize the reflection intensity to for subsequent data analysis and comparison.
[0025] Perform outlier processing on the target speed and target classification. Based on the outlier processing algorithm, exclude data points where the target speed exceeds the normal range, such as a vehicle speed exceeding several times the highway speed limit, and where the target classification is incorrect, such as misclassifying a non-vehicle target as a vehicle.
[0026] For the data of multiple automatically triggered data acquisition units, based on the time synchronization protocol , calibrate the timestamps of the automatically triggered data acquisition units to ensure the time consistency of the data.
[0027] For the objects in the clustering results, based on the target tracking algorithm, associate the objects with different timestamps to ensure the consistency of the objects in different frames, which helps to analyze the movement trajectories and behaviors of the objects.
[0028] If there are missing values in the collected data, different methods are used for filling according to the characteristics and distribution of the corresponding data. For example, for numerical data, filling is based on the mean, median, or interpolation method; for categorical data, filling is based on the mode.
[0029] Record the preprocessed horizontal angle, vertical angle, reflection intensity, timestamp, target speed, target classification, and clustering results as preprocessed automatically triggered data.
[0030] The specific process of preprocessing the manually triggered data includes: Check whether data fields such as button duration, terminal number, position coordinates, user input data, and reporting time are complete to ensure there are no missing values. If there are missing values, decide whether to supplement or delete relevant records according to the specific situation; It should be further noted that if all are lost, locate according to other terminals and remind the maintenance personnel to repair.
[0031] For the reporting time, ensure that its format meets the actual requirements and convert it to a unified time standard.
[0032] If the user input data is in text form, perform text cleaning operations, including removing special characters, punctuation marks, and stop words, tokenize the text, and split it into individual words or phrases for operations such as text classification, sentiment analysis, or keyword extraction; if the user input data is in voice form, convert it to text form and process it based on the above operations.
[0033] Record the preprocessed button duration, terminal number, user input data, and reporting time as preprocessed manually triggered data.
[0034] It should be further noted that in the specific implementation process, the specific process of feature vectorization of preprocessed automatically triggered data includes: Obtain the preprocessed automatically triggered data; Record the horizontal angle of the object within the acquisition range as and the vertical angle as ; According to the horizontal angle , vertical angle , generate the three-dimensional point cloud coordinates of the corresponding object ; the three-dimensional point cloud coordinates are: ; where is the target distance; It should be further noted that the target distance ; where is the speed of light; is the time difference between the transmission and reception of the automatically triggered data acquisition unit.
[0035] Equidistantly divide the normalization range of the reflection intensity of the object within the acquisition range into intervals, and mark the intervals according to the target classification to obtain the marked reflection intensity intervals, and each interval corresponds to the reflection intensity of a different object.
[0036] For example, if the marked interval is the reflection intensity interval of the car surface, and the collected reflection intensity is 200 at this time, it means that a car has passed by at this time.
[0037] The specific process of obtaining the radial velocity and transverse velocity according to the target velocity and clustering result includes: Obtain the three-dimensional point cloud coordinates of the object within the acquisition range for each frame, and record it as ; Among them, represents the corresponding frame, represents the number of objects corresponding to the frame; Build a cost matrix, which is used to associate the association cost of objects in connected frames; Denote the cost matrix as , and the cost matrix represents the association cost between the th object in the th frame and the th object in the th frame; is: ; Among them, , , correspond to the 3D point cloud coordinates of the th object in the According to the cost matrix , and based on the bipartite graph optimal matching algorithm, obtain the optimal matching result of each object in the t-th frame and the corresponding object in the th frame, denoted as , where is the target index in the th frame, is the target index in the th frame, , and g is the number of matched objects; According to the optimal matching result , obtain the centroid displacement ; The centroid displacement is: ; According to the centroid displacement and the time interval , obtain the velocity components in three directions of the target velocity, which are respectively: , , ; Among them, , , respectively correspond to the direction velocity; It should be further noted that the time interval is determined according to the actual sampling frequency; The unit vector corresponding to the line of sight direction of the preset automatic trigger data acquisition unit is ; Denote the radial velocity and the transverse velocity as and ; The radial velocity ; The transverse velocity ; where is the target velocity.
[0038] Concatenate the three-dimensional point cloud coordinates , the radial velocity and the transverse velocity in sequence into a feature vector, denoted as the automatic trigger feature vector.
[0039] It should be further noted that in the specific implementation process, the specific process of feature extraction processing for preprocessed artificial trigger data includes: Perform a threshold judgment on the button duration. If the button duration does not exceed the preset duration threshold, mark it as a mis-touch and do not respond; if the button duration exceeds the preset duration threshold, obtain the corresponding terminal number, user input data, and reporting time; Perform a hash encoding on the terminal number to obtain a terminal encoding vector with a length of ; Based on recognition technology, perform feature extraction on the user input data, considering the occurrence frequency of words in the user input data and the rarity of words in the corresponding corpus; by calculating values to represent the importance of words, thereby obtaining the user input data feature vector of the user input data.
[0040] For example, the user input data is "There was a car accident here. The danger level of the car accident is relatively low. Relevant personnel are needed to handle the car accident!", where the word "car accident" appears 3 times; the total number of words in the user input data is 11, and there is a corpus containing 100 documents, and 10 of them contain the word "car accident"; According to the word frequency The calculation formula is: ; where is the number of occurrences of the corresponding word; is the sum of the number of all words; obtain the word frequency of the corresponding word; According to the inverse document frequency The calculation formula is: ; where is the total number of documents in the corpus; is the number of documents containing the corresponding word; obtain the inverse document frequency of the corresponding word; According to the word frequency of the corresponding words and the inverse document frequency , calculate to obtain a value, the ; and so on, obtain the values of other words; according to the value size order, form a corresponding user input data feature vector.
[0041] Perform time feature extraction on the reported time, obtain the numerical features corresponding to the time, obtain the corresponding season and week according to the reported time, and further obtain a time feature vector; It should be further noted that the time feature vector is related to the occurrence probability or type of the event.
[0042] For example, for the reported time "2024-10-05 14:30:15", extract the year as 2024, the month as 10, the date as 5, the hour as 14, the minute as 30, and the second as 15; and obtain the corresponding autumn and Saturday according to the reported time.
[0043] Concatenate the terminal coding vector, the user input data feature vector, and the time feature vector in sequence into a feature vector, denoted as the artificial trigger feature vector.
[0044] It should be further noted that in the specific implementation process, the specific process of detecting abnormal events for the automatic trigger feature vector includes: Preset the point cloud feature displacement threshold ; It should be further noted that if an object undergoes a large displacement in a short period of time, it means that the state of the object has changed.
[0045] Real-time collect the three-dimensional point cloud coordinates to obtain the corresponding centroid displacement , if the centroid displacement is less than or equal to the cloud feature displacement threshold , mark it as a normal event; if the centroid displacement is greater than the cloud feature displacement threshold , mark it as an abnormal event.
[0046] Preset the radial velocity threshold and the lateral velocity threshold ; For example, in a highway scenario, if the radial velocity of a vehicle exceeds the radial velocity threshold (such as 10 km / h) and lasts for a certain period of time, it means that the vehicle has a malfunction or an accident; if the lateral velocity exceeds the lateral velocity threshold (such as 5 m / s), it means that the vehicle has skidded or lost control.
[0047] Radial velocity collected in real time Less than or equal to the radial velocity threshold When it is, it is marked as a normal event; if the radial velocity Is greater than the radial velocity threshold When it is, it is marked as an abnormal event.
[0048] When the lateral velocity collected in real time Less than or equal to the lateral velocity threshold When it is, it is marked as a normal event; if the lateral velocity Is greater than the lateral velocity threshold When it is, it is marked as an abnormal event.
[0049] It should be further noted that the abnormal events are classified into different categories according to the characteristics and patterns of the abnormal events, including traffic accidents, vehicle failures, object drops, vehicle out of control, and vehicle skidding.
[0050] The three-dimensional point cloud coordinates, radial velocity corresponding to the marked abnormal event And the lateral velocity Generate a cloud feature displacement abnormal instruction, a radial velocity abnormal instruction, and a lateral velocity abnormal instruction, and uniformly record the cloud feature displacement abnormal instruction, the radial velocity abnormal instruction, and the lateral velocity abnormal instruction as an automatic trigger event instruction.
[0051] It should be further noted that in the specific implementation process, the specific process of performing correlation analysis on the manually triggered feature vector includes: For the corresponding Values of several user input data feature vectors are marked and recorded as , , ……, ; Is a natural number; According to the corresponding Values of several user input data feature vectors after marking and the time feature vector, the correlation coefficient is obtained, and the correlation coefficient is: ; where is the mean value of values, represents the time feature vector; The preset correlation coefficient threshold ; If the correlation coefficient , it is marked as a normal event; if the correlation coefficient , it is marked as an abnormal event.
[0052] Generate an artificial trigger event instruction according to the terminal encoding vector marked as an abnormal event, the user input data feature vector, and the time feature vector.
[0053] It should be further explained that, in the specific implementation process, according to the automatic event triggering instruction and the manual event triggering instruction, the specific process of executing the corresponding abnormal event handling solution includes: The abnormal event processing scheme includes an automatic alarm processing scheme and a manual alarm processing scheme; The automatic alarm processing scheme includes: The automatic alarm processing scheme includes a primary abnormality processing scheme and a secondary abnormality processing scheme; if the number of cloud characteristic displacement abnormality instructions, radial velocity abnormality instructions and lateral velocity abnormality instructions is greater than or equal to If the error is 0, the first-level exception handling solution is triggered, otherwise the second-level handling solution is triggered; Setting an induction light control logic, wherein the induction light control logic includes a first-level abnormal induction light control logic and a second-level abnormal induction light control logic; The first-level abnormal induction light control logic includes: 1. Color: red, brightness , automatically drops to , avoid light pollution; 2. Lighting mode: flashing every second times, and continue until the abnormality is resolved or manual intervention occurs; 3. Linkage range: trigger the surrounding area of the terminal All induction lights within meters enter this mode synchronously and send it to the monitoring center.
[0054] The secondary abnormal induction light control logic includes: 1. Color: yellow, brightness , automatically drops to , avoid light pollution; 2. Lighting mode: flashing every second times, continued After 15 minutes, it will automatically switch to steady light until the abnormality is cleared.
[0055] 3. Area marking: trigger the surrounding area of the terminal All corresponding induction lights within meters form a "yellow warning belt" and send it to the monitoring center.
[0056] Set up broadcast processing logic, including first-level abnormal broadcast logic and second-level abnormal broadcast logic; The first-level abnormal broadcast logic includes: 1. Preset audio: Chinese: "An emergency has occurred ahead! Please slow down immediately, turn on the hazard lights, and detour slowly from the right lane!", and play it in a loop with a 5-second interval; English: "Emergency ahead! Slow down immediately, turn on hazard lights, and detour from the right lane!".
[0057] 2. Playback strategy: Play immediately after being triggered, and adjust the volume to , and reduce it to at night, covering the speakers of other terminals within a radius of meters.
[0058] 3. Dynamically adjust the content in combination with the corresponding abnormal instructions.
[0059] The secondary abnormal broadcast logic includes: 1. Preset audio: "Abnormal parking or obstacle ahead! Please slow down to or below and pay attention to the road conditions!", and play it once every 10 seconds.
[0060] 2. Dynamically adjust the content in combination with the corresponding abnormal instructions.
[0061] Preset linkage mechanism, induction lights and broadcasts cooperate: The control logic of the primary abnormal induction lights starts synchronously with the primary abnormal broadcast logic, and the control logic of the secondary abnormal induction lights starts synchronously with the secondary abnormal broadcast logic, forming a "visual auditory" double warning.
[0062] The manual alarm handling plan includes:
[0063] Preset manual induction light control logic;
[0064] 1. Color: white, brightness , and last for 3 minutes to avoid dazzling the driver.
[0065] 2. Lighting mode: Constant on, used to mark the exact location of the event, synchronize the induction lights of other terminals within 50 meters near the terminal installation location into this mode, form a "position beacon", and send it to the monitoring center.
[0066] Preset manual broadcast handling logic; 1. Button-triggered broadcast content: "You have triggered an emergency alarm, and the on-site photos have been uploaded. The rescue team is on the way! Please wait in a safe area.".
[0067] 2. Supplementary reminder: "Please turn on the vehicle's hazard lights and place a triangular warning sign 150 meters in the oncoming direction", for vehicle users.
[0068] 3. Broadcast content triggered by scanning code: Dynamically generated according to user input: If the user inputs "oil leakage", play "Attention! Oil pollution is detected ahead. Do not brake suddenly!"; if the input is "fire", play "Emergency! A fire has occurred ahead. Please stop immediately and evacuate from the safety exit!"
[0069] 4. Multilingual support: Automatically switch languages according to the terminal's location area.
[0070] It should be further noted that in the specific implementation process, the specific process of constructing the auxiliary optimization static target classification model includes: Obtain several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data; Group and label several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data, denoted as where is a natural number; Take groups of several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data as sample data, and is a natural number less than and use the sample data to obtain the mean of the sample data based on the machine learning algorithm, denoted as the sample set; Take the remaining several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data as the test set; According to the sample set and the test set, form a training sample set;
[0071] Input the training sample set into the standard auxiliary optimization model, train the standard auxiliary optimization model, and denote the trained standard auxiliary optimization model as the auxiliary optimization static target classification model.
[0072] Generate the static target classification factor under the current conditions according to the auxiliary optimization static target classification model ; where is the weight vector, is the bias term, is the comprehensive feature vector; It should be further noted that the comprehensive feature vector is: ; where is the automatically triggered feature vector; is the historical manually triggered feature vector; is the vector representation of the device status data; , , The weights represented by the historical automatically triggered feature vector, historical manually triggered feature, and device status data vector.
[0073] Preset standard stationary target classification factor ; If the stationary target classification factor , then the status of the stationary target under the current conditions is normal; If the stationary target classification factor , then the status of the stationary target under the current conditions is abnormal, and an automatic alarm processing scheme is executed, greatly reducing the alarm time.
[0074] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.
Claims
1. A seven-in-one intelligent road system, characterized in that, Including: A data acquisition module, which is used to acquire automatically triggered data, manually triggered data, and device status data; A data intelligent processing module, which is used to preprocess the automatically triggered data and the manually triggered data to obtain preprocessed automatically triggered data and preprocessed manually triggered data; Perform feature vectorization processing on the preprocessed automatically triggered data to obtain an automatically triggered feature vector; Perform feature extraction processing on the preprocessed manually triggered data to obtain a manually triggered feature vector; A data intelligent analysis module, which is used to detect abnormal events for the automatically triggered feature vector to obtain an automatically triggered event instruction; perform correlation analysis on the manually triggered feature vector to obtain a manually triggered event instruction; execute corresponding abnormal event handling solutions according to the automatically triggered event instruction and the manually triggered event instruction; An intelligent optimization module, which is used to construct an auxiliary optimization static target classification model based on the automatically triggered feature vector, the manually triggered feature vector, and the device status data to obtain a static target classification factor, and classify and assist in optimizing the status of the static target according to the static target classification factor.
2. The seven-in-one intelligent road system according to claim 1, characterized in that The process of acquiring automatically triggered data, manually triggered data, and device status data includes: Set up a data acquisition device, which is composed of several automatically triggered data acquisition units, manually triggered data acquisition units, and device status acquisition units; the automatically triggered data acquisition unit is used to acquire horizontal angle, vertical angle, reflection intensity, timestamp, target speed, target classification, and clustering results, and record them as automatically triggered data; the manually triggered data acquisition unit is used to acquire button duration, terminal number, user input data, and reporting time, and record them as manually triggered data; the device status acquisition unit is used to acquire the device status data of the terminal.
3. The seven-in-one intelligent road system according to claim 2, characterized in that The process of preprocessing the automatically triggered data and the manually triggered data includes: Obtain the automatically triggered data and the manually triggered data; The process of preprocessing the automatically triggered data includes: For horizontal angle, vertical angle, and reflection intensity, outlier removal operations are performed, and the distance or deviation between each data point and its adjacent points of the corresponding data is calculated. Based on statistical methods, outliers are identified and removed; outlier processing is performed on target speed and target classification. Based on the outlier processing algorithm, data points with target speed exceeding the normal range and target classification errors are removed; for data from multiple automatically triggered data acquisition units, based on the time synchronization protocol , calibrating the timestamp of the automatic triggering data acquisition unit to ensure the time consistency of the data; for the objects in the clustering results, associating the objects with different timestamps based on the target tracking algorithm to ensure the consistency of the objects in different frames; Record the preprocessed horizontal angle, vertical angle, reflection intensity, timestamp, target speed, target classification, and clustering results as preprocessed automatically triggered data; The process of preprocessing the manually triggered data includes: Check the integrity of the data fields of the button duration, terminal number, position coordinates, user input data, and reporting time; for the reporting time, convert it to a unified time standard; if the user input data is in text form, perform text cleaning operations, including removing special characters, punctuation marks, and stop words, perform word segmentation on the text, and split it into individual words or phrases; if the user input data is in voice form, convert it to text form; Record the preprocessed button duration, terminal number, user input data, and reporting time as preprocessed manually triggered data.
4. The one - road seven - in - one intelligent system according to claim 3, characterized in that, The process of performing feature vectorization processing on the preprocessed automatically triggered data includes: Obtain the preprocessed automatically triggered data; Record the horizontal angle of the object within the acquisition range as , and record the vertical angle as ; According to the horizontal angle , vertical angle , generate the three-dimensional point cloud coordinates of the corresponding object ; The three-dimensional point cloud coordinates are as follows: ; wherein, is the target distance; the said target distance ; wherein, is the speed of light; is the time difference between the transmission and reception of the automatic trigger data acquisition unit; The normalized range of the reflection intensity of the objects within the acquisition range is equally divided into intervals, and according to the target classification, the intervals are marked to obtain marked reflection intensity intervals, and each interval corresponds to the reflection intensity of different objects; Obtain the radial velocity according to the target velocity and the clustering result and the transverse velocity ; Concatenate the three-dimensional point cloud coordinates , the radial velocity and the transverse velocity in sequence into a feature vector, denoted as the automatic trigger feature vector.
5. The seven-in-one intelligent road system according to claim 4, characterized in that Obtain the radial velocity based on the target velocity and the clustering result and the transverse velocity The process includes: Obtain the three-dimensional point cloud coordinates of the objects within the acquisition range of each frame, denoted as ; where represents the corresponding frame, represents the number of objects corresponding to the frame; Establish a cost matrix, which is used to associate the association cost of objects in adjacent frames; Denote the cost matrix as , where the cost matrix represents the association cost between the -th object in the -th frame and the -th object in the -th frame; The cost matrix is denoted as as follows: ； Among them, , , correspond to the three-dimensional point cloud coordinates of the th object in the th frame; According to the cost matrix , and based on the bipartite graph optimal matching algorithm, obtain the optimal matching result of each object in the t-th frame with the corresponding object in the -th frame, denoted as , where is the target index in the -th frame, is the target index in the -th frame, , and g is the number of matched objects; According to the optimal matching result , the centroid displacement is obtained ; the centroid displacement is as follows: ； According to the centroid displacement and the time interval , velocity components in three directions of the target velocity are obtained, which are respectively: , , ; where , , correspond to the velocities in the directions respectively; the time interval is determined according to the actual sampling frequency; The unit vector corresponding to the line of sight direction of the preset automatic trigger data acquisition unit is ; The radial velocity ; The transverse velocity ; Wherein is the target velocity.
6. The seven-in-one intelligent road system according to claim 5, characterized in that, The process of performing feature extraction processing on the preprocessed manually triggered data includes: Perform a threshold judgment on the button duration. If the button duration does not exceed the preset duration threshold, mark it as a mis-touch and do not respond; if the button duration exceeds the preset duration threshold, obtain the corresponding terminal number, user input data, and reporting time. Perform a hash encoding on the terminal number to obtain a terminal encoding vector with a length of ; Based on the recognition technology, extract features from the user input data to obtain the corresponding value, thereby obtaining the user input data feature vector of the user input data; Extract time features from the reporting time, obtain the numerical features corresponding to the time, obtain the corresponding season and week according to the reporting time, and then obtain a time feature vector. Concatenate the terminal coding vector, user input data feature vector, and time feature vector in sequence to form a feature vector, denoted as the artificial trigger feature vector.
7. The one-way seven-in-one intelligent system according to claim 6, characterized in that, The process of detecting abnormal events for the automatic trigger feature vector includes: Preset point cloud feature displacement threshold ; Three-dimensional point cloud coordinates collected in real time to obtain the corresponding centroid displacement , if the centroid displacement is less than or equal to the cloud feature displacement threshold , it is marked as a normal event; if the centroid displacement is greater than the cloud feature displacement threshold , it is marked as an abnormal event; Preset radial velocity threshold and transverse velocity threshold ; When the radially acquired velocity is less than or equal to the radial velocity threshold , it is marked as a normal event; if the radial velocity is greater than the radial velocity threshold , it is marked as an abnormal event; When the horizontally velocity collected in real time is less than or equal to the horizontal velocity threshold it is marked as a normal event; if the horizontal velocity is greater than the horizontal velocity threshold it is marked as an abnormal event; The three-dimensional point cloud coordinates and radial velocity corresponding to the marked abnormal events and the lateral velocity Generate a cloud feature displacement abnormal instruction, a radial velocity abnormal instruction, and a lateral velocity abnormal instruction, and uniformly record the cloud feature displacement abnormal instruction, the radial velocity abnormal instruction, and the lateral velocity abnormal instruction as an automatic trigger event instruction.
8. The one kind of road seven-in-one intelligent system according to claim 7, characterized in that The process of performing correlation analysis on the artificial trigger feature vector includes: Mark the values corresponding to a number of user input data feature vectors as and denote them as , , ……, ; where is a natural number; According to the values corresponding to a number of user input data feature vectors after marking and the time feature vector, a correlation coefficient is obtained. The correlation coefficient is as follows: ; wherein, is the value mean, indicating the time feature vector; Preset correlation coefficient threshold ; If the correlation coefficient , it is marked as a normal event; if the correlation coefficient , it is marked as an abnormal event; Generate an artificial trigger event instruction according to the terminal coding vector, user input data feature vector, and time feature vector marked as abnormal events.
9. The seven-in-one intelligent road system according to claim 8, wherein The process of executing the corresponding abnormal event handling scheme according to the automatic trigger event instruction and the artificial trigger event instruction includes: The abnormal event handling scheme includes an automatic alarm handling scheme and an artificial alarm handling scheme; The automatic alarm handling solution includes a primary exception handling solution and a secondary handling solution; if the number of cloud feature displacement exception instructions, radial velocity exception instructions, and lateral velocity exception instructions is greater than or equal to then the primary exception handling solution is triggered; otherwise, the secondary handling solution is triggered. The automatic alarm handling solution includes: setting induction lamp control logic, which includes primary abnormal induction lamp control logic and secondary abnormal induction lamp control logic; setting broadcast handling logic, including primary abnormal broadcast logic and secondary abnormal broadcast logic; presetting a linkage mechanism, where the primary abnormal induction lamp control logic and the primary abnormal broadcast logic are started synchronously, and the secondary abnormal induction lamp control logic and the secondary abnormal broadcast logic are started synchronously, forming a "visual auditory" dual warning; The artificial alarm handling scheme includes: preset artificial induction lamp control logic and artificial broadcast handling logic.
10. The seven-in-one intelligent road system according to claim 9, characterized in that, The process of constructing an auxiliary optimized static target classification model includes: Obtain several groups of historical automatic trigger feature vectors, historical artificial trigger feature vectors, and device status data; Group numbers are assigned to several sets of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data, denoted as which is a natural number; Put A number of groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data are used as sample data, and is a natural number less than . Using the sample data, based on a machine learning algorithm, the mean of the sample data is obtained and denoted as the sample set; the remaining several groups of historical automatically triggered feature vectors, historical manually triggered feature vectors, and device status data are used as the test set; According to the sample set and the test set, form a training sample set; based on the convolutional neural network, construct a standard auxiliary optimization model; and input the training sample set into the standard auxiliary optimization model to train the standard auxiliary optimization model, and denote the trained standard auxiliary optimization model as the auxiliary optimized static target classification model. Generate the stationary target classification factor under the current conditions according to the auxiliary optimization of the stationary target classification model ; where is the weight vector, is the bias term, is the comprehensive feature vector; the comprehensive feature vector is: ; wherein, is the automatically triggered feature vector; is the historical manually triggered feature vector; is the vector representation of the device status data; , , are the weights of the historical automatically triggered feature vector, the historical manually triggered feature, and the vector representation of the device status data; Preset standard stationary target classification factor ; If the static target classification factor , then the state of the static target under the current condition is normal; If the stationary target classification factor , then the state of the stationary target under the current conditions is abnormal, and an automatic alarm processing scheme is executed, greatly reducing the alarm time.