AI pavement event interaction and processing device and method

Abstract

The invention discloses an AI pavement event interaction and processing device and method, and the device comprises an interface unit which is used for being connected with cloud equipment, a data source and traffic safety equipment; the knowledge base is generated by learning and training of the cloud equipment, is downloaded to the current device through the interface unit, and is used for performing fitting prejudgment on the current traffic event according to the real-time data provided by the field data source and sending out early warning in advance for the event conforming to the traffic accident characteristics; and the processing unit controls and / or drives the traffic safety equipment to complete corresponding instructions through the interface unit according to the early warning sent by the knowledge base. According to the invention, the multi-dimensional data of the traffic site can be collected in real time, and the real-time data is analyzed based on the requirement of active accident prevention of the knowledge base, so that one or more traffic safety equipment is driven to implement accident early warning, and the effects of early prevention, early intervention and reduction of the probability of occurrence of traffic accidents can be realized.

Description

Technical Field

[0001] The present invention relates to an AI road surface event interaction and processing device and method, belonging to the technical field of road traffic. Background Art

[0002] Studies have shown that road traffic accidents are generally caused by the combined effects of multiple internal and external factors. Theoretically, some of the main factors that lead to traffic accidents can be detected by existing technologies or sensors before the accident. If these identified factors that may cause accidents are effectively utilized, and individual main factors that lead to accidents are changed or eliminated through technical means, traffic accidents may be avoided.

[0003] The widely used equipment in the field of road traffic, such as signboards, variable signal display devices, pavement sensors, ice and snow melting devices, roadside and road surface induction devices, etc., are all based on single-task settings and can only perform the product functions defined by themselves. These traffic safety equipment and sensors installed in the transportation field generate massive amounts of traffic event data from different dimensions every day. Many of these data are discarded, and only a very small amount is stored. However, these stored data are usually shelved before they are discovered. If these data are clearly directed and these devices and data are effectively used, the goal of active accident prevention can be gradually achieved in the transportation field.

[0004] At present, these data still exist in the form of data fragments and have not been effectively mined and utilized. By mining and refining these data (or conducting machine learning and training on these data), it is possible to profile the formation of traffic accidents in different dimensions and form three-dimensional accident components. The accident elements composed of these data summarize the preconditions for accidents from different dimensions and form a group of accident symptom characteristics that can be identified in advance. If this knowledge can be effectively mined and used, the existing intelligent traffic safety equipment can play a more effective disaster reduction function, but no such devices or equipment have been found to be put into field application.

[0005] At present, the vast majority of motor vehicles on the road are still in the manual driving stage. These vehicles basically do not have networking functions, and it is difficult for the roadside to conduct fixed-point interaction with key vehicles (also known as characteristic vehicles with obvious accident symptoms). To achieve information delivery to non-networked vehicles, the edge device needs to have the ability to deliver information to dynamic or static vehicles; if information is delivered to a specific dynamic vehicle, it is also necessary to have the ability to identify and locate vehicle characteristics. These independently formed technologies and equipment are not yet popular, but roadside, road surface guidance equipment or perception equipment embedded with these technologies have been widely used. Mapping the data obtained by certain display functions or perception devices in these devices to AI road surface event interaction and processing devices can meet the information delivery requirements in road traffic safety assurance. However, information delivery based on knowledge base is still rare.

[0006] At present, most of the main basic conditions for realizing active accident prevention have been met, but these conditions are still in a discrete state. There is a lack of technical means to integrate these discrete data, technologies, and equipment. There is also a lack of a device to combine these discrete data and technologies from different dimensions into a comprehensive management device for active accident prevention mission services. Summary of the invention

[0007] The purpose of the present invention is to provide an apparatus and method for AI road surface event interaction and processing. The present invention can collect multi-dimensional data of traffic scenes in real time, analyze these real-time data based on the requirements of active accident prevention in the knowledge base, and then drive one or more traffic safety equipment to implement accident warning, so as to achieve the effect of early prevention, early intervention and reduction of the probability of traffic accidents.

[0008] The technical solution of the present invention: an AI road surface event interaction and processing device, comprising:

[0009] An interface unit for connecting with cloud devices, data sources, and traffic safety equipment;

[0010] The knowledge base is generated by learning and training of cloud devices and downloaded to the current device through the interface unit. It is used to fit and predict current traffic events based on real-time data provided by on-site data sources, and to issue early warnings for events that meet the characteristics of traffic accidents;

[0011] The processing unit controls and / or drives the traffic safety equipment to complete corresponding instructions through the interface unit according to the warning issued by the knowledge base.

[0012] The above-mentioned AI road event interaction and processing device, the data source includes sensors related to traffic events; the device reads data from different sensors based on the same time mark and different coordinate values ​​and uploads it to a cloud device, and the cloud device learns the data and trains to form a knowledge base.

[0013] The aforementioned AI pavement event interaction and processing device, the sensor data related to the traffic event includes pavement dryness and wetness, snow accumulation and snow thickness, ice and ice thickness, temperature drop slope, pavement dryness and wetness change efficiency, visibility value, rainfall, vehicle speed, vehicle model, lane occupancy, headway, vehicle speed variable, congestion, slow driving, collision with guardrail, pavement illumination, vehicle driving direction, road surface waterlogging, time speed variable, emergency stop event, parking event and the current status of pavement closure equipment.

[0014] The aforementioned AI pavement event interaction and processing device, the traffic safety equipment includes a roadside or pavement induction device, a roadside or gantry variable information board, a roadside or pavement ice and snow melting device, an automatic pavement cleaning device, a warning light, a signal light, a tunnel induction device, a road opening and closing device and a tunnel locking device.

[0015] The aforementioned AI pavement event interaction and processing device is embedded with installation point coordinates, which are regionally unique and are used to receive a knowledge base with corresponding coordinates and management section coordinates downloaded from a cloud device; the learning and training of the knowledge base are based on the same time identifier and adjacent and continuous coordinates, and the trained knowledge base is downloaded from the cloud device to the corresponding device based on the installation point coordinates.

[0016] The aforementioned AI road surface event interaction and processing device compares the data collected in different dimensions with various factors that may cause accidents in the knowledge base within a continuous coordinate range based on the same time. When data of certain dimensions in the real-time data generated at the road traffic scene are compared with the knowledge base and potential accident hazards are found, the device will drive the traffic safety equipment connected to the device to make an effective response based on the current sensor data and the solution corresponding to the knowledge base, so as to warn the target to avoid possible accident hazards.

[0017] The aforementioned AI road event interaction and processing device, wherein all sensor data collected by the device are identified based on a reference time source and continuous adjacent coordinates and serve as a reference for identifying multi-dimensional data of traffic events; the processing unit controls and / or drives traffic safety equipment based on the same time source and adjacent coordinate segments.

[0018] The aforementioned AI road event interaction and processing device, the historical data received or collected by the device are directly uploaded to the cloud device after identification processing, and are not stored locally; the real-time data received or collected by the device is uploaded to the cloud device and a copy is retained locally for identifying and comparing the current traffic event anomalies. This value is rolled over and stored for a certain period of time and is overwritten by new data after exceeding the preset time.

[0019] The aforementioned AI road surface event interaction and processing device has the ability to transmit current comprehensive information and current control information to be driven to the manual management platform. The control information will be classified. General information will directly control and / or drive the traffic safety equipment after being broadcast to the manual management platform. Key information will be sent to the manual management platform before issuing control and / or drive instructions. If no intervention instruction is received within the set time period, the control and / or drive instruction will be issued directly. If an intervention instruction is received, it will be executed according to the intervention instruction of the manual management platform.

[0020] The method of the aforementioned AI road surface event interaction and processing device, when installed, connects the device to the cloud device, data source and traffic safety equipment via the interface unit; uses the device to receive or collect historical data of the data source, and then transmits the historical data to the cloud device, and the cloud device learns and trains to generate a knowledge base, and the knowledge base is downloaded to the local device via the interface unit;

[0021] When in use, the device obtains real-time data of traffic events based on on-site data sources, and then compares and fits the probability of traffic accidents in the current traffic environment based on the local knowledge base and real-time data, and issues early warnings for events that meet the characteristics of traffic accidents. Finally, the device controls and / or drives traffic safety equipment to complete corresponding instructions based on warning learning, so as to warn the target to avoid possible accident hazards.

[0022] Compared with the prior art, the present invention is a comprehensive management and control device capable of identifying and warning traffic events. The present invention can configure the induction devices and various sensor devices installed on the road surface and roadside at the current stage, and can respond to traffic events based on the knowledge base data, thereby achieving the effect of early prevention, early intervention and reducing the probability of traffic accidents. The present invention can realize data collection and operation management of intelligent traffic safety equipment installed on the road surface, and can also sense the current road surface state and upload it to the cloud device. The cloud device forms a knowledge base after machine learning and downloads it to the local device, so that a corresponding local knowledge base can be formed according to different road traffic environments, so that fitting and prediction can be carried out according to the actual situation of a certain place, which greatly improves the actual application ability of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1is a schematic diagram of an interface unit, a knowledge base, and a processing unit of the present invention;

[0024] Figure 2 It is a structural schematic diagram of a device in a specific example of the present invention;

[0025] Figure 3 The figure is a schematic diagram of the application process of the device in a specific example of the present invention. DETAILED DESCRIPTION

[0026] The present invention is further described below in conjunction with the accompanying drawings and embodiments, but they are not intended to limit the present invention.

[0027] Embodiment 1: An AI road surface event interaction and processing device 100, such as Figure 1 As shown, including:

[0028] An interface unit 101 is used to connect with cloud devices, data sources and traffic safety equipment;

[0029] The knowledge base 102 is generated by learning and training of the cloud device and downloaded to the current device through the interface unit, and is used to fit and predict the current traffic events based on the real-time data provided by the on-site data source, and issue early warnings for events that meet the characteristics of traffic accidents;

[0030] The processing unit 103 controls and / or drives the traffic safety equipment to complete corresponding instructions via the interface unit according to the warning issued by the knowledge base.

[0031] The AI ​​road surface event interaction and processing device of the present invention (hereinafter referred to as the device) is an intelligent integrated control device installed on the roadside with an independent operation configuration. The hardware of the device can be composed of an independent component or can be compatible with other devices. Figure 2 The modules in the schematic diagram can be in an independent device or dispersed to other devices and combined and applied through software; this structural diagram is only used to better understand the content of the present invention, and it is not the final and only configuration mode. At the same time, in other embodiments, the device does not have to be installed on the roadside, and can also be associated with each section of the road through a link, and its effective area has a precise section of the road. The power supply requirements of this device allow the use of a DC power supply in the range of 10 to 16V. The device has multiple built-in external interfaces, and a gateway or a direct external data source can be configured for each interface. It can also connect and drive other traffic safety-related devices through interfaces or gateways.

[0032] like Figure 2As shown, the interface unit 101 includes a V2I interface, a wired interface and a wireless interface, wherein the V2I interface is a direct connection interface and can also use an Internet access method. When using Internet access, the V2I direct connection module may not be used; the wired interface includes a bus interface and an Internet interface, the bus interface is used to directly connect to field equipment or sensors, and the wired interface is used to connect to access devices such as interactive machines and access the Internet through the device; the wireless interface refers to a built-in zigbee module or / and a 4G / 5G module, wherein the zigbee module is used to interconnect with field equipment or / and third-party equipment, and the 4G / 5G module is used for mobile access to the Internet (usually using an Internet of Things card).

[0033] The cloud devices mentioned above refer to cloud servers that can be commercially used on the market, including but not limited to Alibaba Cloud, Tencent Cloud, Huawei Cloud, etc.

[0034] The data sources come from various sensors that detect traffic events and meteorological data, including but not limited to road surface dryness and humidity, snow accumulation and snow thickness, ice and ice thickness, temperature drop slope (temperature change efficiency in integer unit time, measurement starts when the temperature is below 6 degrees Celsius), road surface dryness and humidity change efficiency (detection starts when the road surface dryness and humidity begins to change, and the road surface dryness and humidity change efficiency is in minutes), visibility value, rainfall, vehicle speed, vehicle type, lane occupancy, headway, vehicle speed variable (vehicle speed change per unit time), congestion, slow driving, collision with guardrails, road surface illumination, vehicle driving direction, road surface waterlogging, time vehicle speed variable (vehicle speed change in a specific section within 24 hours with the average speed per hour as the unit), emergency stop events, parking events, and the current status of road surface closure equipment. The above detection items are a list of different dimensional data that can be obtained by this device at the same time. This device allows a maximum of 36 different data sources to be collected and labeled at the same time (no less than 128 different types of data collected in non-real time). The various data provided to this device by the data source are uploaded to the cloud device after being confirmed to be compliant. Compliance refers to whether the elements such as data expression format, timestamp, coordinates, etc. are complete. If incomplete, they should be supplemented or discarded. The data collected by this device needs to comply with the definition mode and requirements of the data format of this device and cloud devices. Since the data sources collected by this device from different data sources are relatively complex and the data formats are also varied, all collected data formats must be tested for compliance. Non-compliant data formats need to be re-processed for compliance. Only after the compliance processing is completed can the data be used for comparison, fitting or uploading. Some data cannot be processed for compliance due to various factors, so these data can only be discarded. This device sets different built-in or external gateways for receiving and parsing these data, so as to standardize the use of these third-party data formats that may be obtained. For example, wired and wireless interfaces based on TCP / IP protocol, wireless interfaces based on zigebb protocol, built-in 4G / 5G wireless communication cards based on mobile communications, and local interconnection using 485 bus; this device also has a power monitoring module that can remotely monitor and manage the power consumption of this device and the continuous power supply capability of the power supply;

[0035] The traffic safety equipment is connected to the device through an interface or a gateway, and includes but is not limited to a roadside or road surface induction device, a roadside or gantry variable information board, a roadside or road surface ice and snow melting device, an automatic road surface cleaning device, a warning light, a signal light, a tunnel induction device, a road opening and closing device, and a tunnel locking device. The actual connection is determined by the traffic safety equipment at the installation point of the device. The connected traffic safety equipment can accept the driving instructions of the device and can display warning information on the interactive device.

[0036] The processing unit 103 uses a 16-bit MSP430F449IPZR LQFP-10060KB+256B Flash2KB RAM processor. Since the hardware is related to the production cost of the device, and there is no process that requires high computing power in the device (the processes that require high computing power have been placed in the cloud device), the device does not have high requirements for the processor, thereby reducing costs.

[0037] The operation of this device is based on the "Traffic Safety Incident Handling Countermeasure Library" downloaded from the cloud, also known as the knowledge database or simply the knowledge base; the knowledge base is a model generated by the cloud device after learning and training. It is based on the multi-dimensional historical data and traffic events of this road section plus the publicly known traffic event knowledge data to form a traffic accident judgment model. The model analyzes the occurrence of various types of traffic accidents from multiple dimensions to form a dynamic judgment model for traffic accident warning. It can fit and predict the current traffic event based on the real-time data provided by the on-site sensors, and issue warnings in advance for events that meet the characteristics of traffic accidents. The knowledge base can be understood as a data combination based on multiple factors to form an accident. It contains the necessary conditions and accidental conditions that must be met to cause traffic accidents, and is also a prediction of the probability of traffic accidents under certain specific conditions. The accuracy of the algorithm module of the knowledge base depends on the time length of the original data and the number of dimensions contained in these data. The more complete these basic data are, the higher the prediction accuracy will be based on the comparison or fitting of the measured data with the algorithm module. In this embodiment, the data time length is 3 and a half years, the sampling step is 1 second, the data dimension is 12, and the road length is 21 kilometers. These data lack official confirmed accident details (the data are confidential), so the accident data in this knowledge base is the accident data of this road section obtained in the cloud using a five-dimensional comprehensive data inference model. The knowledge base running in this device has multi-dimensional traffic information data and traffic accident models that can be compared in real time to achieve early warning of traffic accidents. At the same time, since the geographical environment of different coordinates may be different, the knowledge base is downloaded and deployed in the device of the corresponding coordinates. The knowledge base of the present invention can also be generated by manual editing (special environments or environments without historical data can be manually edited by the early warning model), and the information in the knowledge base or some of them can be manually edited later during the application process. In the knowledge base downloaded from the cloud, no less than 8 real-time variables can be derived at the same time and at least one or two dependent variables can be given, and anthropomorphic professional warning information can be automatically generated in a targeted manner according to the dependent variables.

[0038] The comparison and fitting of the knowledge base is to put the real-time data of different dimensions on the substrate for comparison and fitting. Successful comparison means that the real-time data and the data of one of the multi-dimensional accident causes in the current traffic environment meet the accident definition of the knowledge base; fitting means that the accident causes of multiple main dimensions together constitute a traffic accident that is likely to occur. Since the knowledge base is dynamic, the more dimensions and the larger the amount of data, the higher the accuracy of the knowledge base for traffic accident warning will be, so the probability of traffic accident warning accuracy will also increase synchronously with the increase of time and data volume.

[0039] The real-time data required for the operation of this device refers to the data obtained by this device from the data source, and is temporarily stored in this device in a rolling manner within a specified time, mainly used for real-time comparison or fitting. When the temporary storage time is exceeded, it is uploaded to the cloud device (this is a parameterized setting. When the uplink bandwidth is wide, it can be directly uploaded when the data is collected. When the uplink bandwidth is narrow, delayed idle upload can be used). The temporary storage time is approximately equal to the warning advance time. For example, the warning mode is the warning time mode, and the warning advance time is 15 seconds. The cumulative local storage time of real-time data is no more than 15 seconds after the first call for data comparison or fitting, and then the previous data is covered by a rolling method; in this way, multiple vehicles that are connected to the comparison line one after another can be continuously compared or fitted (based on the dynamic main parameters in the warning time mode are vehicles with different speeds, so the comparison line is also dynamic, so the comparison is also continuously and dynamically performed. The comparison line mentioned here refers to the physical or logical starting point for comparison at this time or place).

[0040] Furthermore, the device is connected to existing roadside and road surface intelligent traffic safety equipment and sensors related to traffic events or / and traffic meteorological events through an interface or / and a gateway, and then reads the real-time data of different sensors based on the same time identifier and different coordinate values ​​and uploads them to the cloud device, which learns and trains these data and forms a "Traffic Safety Event Processing Countermeasure Library". The device is embedded with installation point coordinates and has regional uniqueness. It can receive a knowledge database with corresponding coordinates and management section coordinates downloaded by the cloud device. The knowledge database is based on the multi-dimensional comprehensive data collected by the sensors of this section and / or traffic accidents or / and traffic events, which are converted into a local and unique knowledge base after being processed by the cloud. The learning and training of events in this knowledge base are based on the same time source and adjacent and continuous coordinates. The trained knowledge base is downloaded from the cloud to the device installed at the corresponding coordinates according to the coordinates.

[0041] Furthermore, the device will collect data from different dimensions (different dimensions refer to data with different directions, such as road surface dryness and wetness, whether it is frozen, visibility value, traffic flow, headway, lane occupancy, vehicle type, vehicle speed and other traffic event data) within a continuous coordinate range based on the same time, and after excluding non-factor data (non-factor data refers to currently useless data, for example, when the temperature is above the freezing point, there is no need to consider the road surface icing factor, and so on), compare it with various factors that may cause accidents in the knowledge base. When data of certain dimensions in the real-time data generated by different dimensions at the road traffic scene are compared with the knowledge base and potential accident hazards are found, the AI ​​road surface event interaction and processing device will drive the intelligent traffic safety equipment connected to the AI ​​road surface event interaction and processing device to make an effective response (such as sending sound, light, electrical signals, characters, voice, etc. to the target vehicle) according to the solution corresponding to the knowledge base based on the current sensor data, so as to warn the target vehicle to avoid possible accident hazards.

[0042] Furthermore, the installation point of the device is different due to the regional differences in road traffic environment (for example, there are differences in roads in the Northeast, Northwest, and Central China), road type differences (for example, mountainous areas and plains, uphill and downhill sections of varying lengths, roads of different widths, and road surfaces composed of different materials), and climate differences (fog, rain, snow, ice, etc.). At the same time, traffic accidents have causes of accidents in different dimensions, and the proportions of different dimensions are generated after statistics of historical data in the cloud. In the knowledge base, the weights of various real-time data in the comparison and fitting process of traffic accidents (the difference between key data and auxiliary data) are generated based on the statistics of historical data; since the distinction between key data and auxiliary data of the causes of accidents is also closely related to the region, season, climate, and traffic environment, the puzzle of the causes of traffic accidents is dynamic and changes with time, season, environment, and region. Therefore, some knowledge in the content of the knowledge base is unique, and the application of the knowledge base needs to correspond to the coordinate segment. That is to say, the hardware of the device can be consistent during the production process, but after the knowledge base is installed, it has the uniqueness of the application.

[0043] Furthermore, the device uses satellite timing as the reference time source. All data collected by the device are identified based on the reference time source and continuous adjacent coordinates and used as the reference basis for identifying multi-dimensional data of traffic events, which is one of the key bases for analyzing and identifying the causes of a certain type of traffic events. The early warning information triggered by real-time data and the precise control of traffic safety equipment are also based on the same time source and adjacent coordinate segments. For example, when a high level of danger appears on the local road surface and it is necessary to induce vehicles on the road to leave the main lane, it is necessary to turn on the induction lights in a time sequence and form a water induction mode to provide highly directional induction to vehicles on the road. At the same time, these inductions are limited to local sections, so adjacent coordinate segments are required to implement active accident prevention functions.

[0044] Furthermore, the device is a host device similar to the roadside intelligent device. It is not managed and driven by the roadside intelligent device and the sensing device, but only accepts manual control and the control of the host management device of the device. The historical data received or collected by the device (the data before the current X hours are all historical data, X is a custom time, and the unit is hours) are directly uploaded to the cloud device after identification processing (identification processing refers to adding coordinates, data types and other tags to data of different dimensions, and different dimensions refer to different traffic data sensing devices, such as vehicle speed, vehicle type, vehicle distance, etc.), and are not stored locally. The real-time data received by the device (the time within Y hours from the current time is the current time, and the value can be customized) will be kept locally after being uploaded to the cloud device. A copy is used to identify and compare the abnormal points of the current traffic event. The value is rolled over and stored within Y hours until it exceeds the preset time and is overwritten by new data. The device is an application-type intelligent device. It does not need to have the ability to learn and train the original data, but has the ability to directly apply the model downloaded to the local after the training is completed (the trained model is called a knowledge base), and can continuously upgrade and continuously improve the effectiveness of the active accident prevention function based on the continuously updated knowledge base.

[0045] Furthermore, this device has the ability to transmit current comprehensive information and current control information to be driven to the management platform (this management platform is the upper computer of this device relative to this device). The control information will be classified. General information will directly drive the lower device after being broadcast to the upper device. Key information (such as road closure) will be sent to the manual management platform before issuing the driving command. If no intervention command is received within the set time period, the driving control command will be issued directly. If an intervention command is received, it will be executed according to the upper intervention command.

[0046] Furthermore, in order to optimize the equipment and reduce costs, the machine learning and training interactive facade dimensions of the data are completed in the cloud (the main reason is that these operations require a lot of computing power). Since the response time required for traffic events is very short, in order to ensure the response time of on-site equipment and devices, this device will filter out comparison factors that do not meet the current conditions during actual use (for example, when the temperature is higher than 4 degrees Celsius, ice and snow factors will not be compared, and when the road surface is dry, factors of slippery road surface will not be extracted for event identification, etc.), and only the key dimensions of the knowledge base that have been trained and meet the current environmental conditions are extracted for real-time comparison with key data, so as to ensure rapid response in the field while reducing the computing power required by this device.

[0047] The above describes the basic structure and principle of the device, so the device can collect multi-dimensional data of the traffic scene in real time, and analyze these real-time data based on the requirements of active accident prevention of the knowledge base, and then drive one or more traffic safety equipment to implement accident warning, so as to achieve the effect of early prevention, early intervention and reduction of the probability of traffic accidents. At the same time, the devices distributed in different regions will continuously provide various data to the cloud device, and the human-computer interaction port will also provide various newly discovered basic knowledge and test data to the cloud device, which can continuously improve the accuracy of the knowledge base over time.

[0048] Embodiment 2: This embodiment provides a specific application mode based on the device in Embodiment 1, such as Figure 3 The specific steps are as follows:

[0049] Step 1: Enable early warning search;

[0050] Step 2: Review current environmental factors and establish exclusions (exclusions refer to comparison data items that do not currently have this type of risk);

[0051] Step 3. Review the current environmental factors and determine the content and quantity of effective comparison or fitting items (mainly based on the number of data source dimensions and the warning mode, so as to call the currently available and most important data or data group as the main comparison or fitting data source). The number of main comparison items is parameterized, and one or more items can be selected for comparison or fitting at the same time. When using time warning, due to the large number of variables, it is not advisable to compare and fit more than 8 main data at the same time. There is no limit when using distance warning mode.

[0052] Step 4: Use the actual main comparison real-time data to compare with the conditions for the formation of such accidents in the knowledge base. If more than 65% of the total amount is successfully matched, it is in the discovery warning state (if more than or equal to 65% of the main warning items are successfully matched, it will enter the auxiliary item comparison). If it is insufficient or no warning items are found, return to step 3;

[0053] Step 5: call auxiliary parameters for comparison or fitting. Auxiliary parameters or data refer to non-main items in the accident cause set. 50% of the auxiliary parameters are successfully compared and fitted. When the proportions of the main and secondary warning items are met at the same time, enter the warning analysis. If the auxiliary warning item does not meet the warning conditions, return to step 3.

[0054] Step 6: According to the real-time data and system configuration, a handshake warning information release device, such as a variable information display screen or an induction device or a V21 device, generates corresponding warning information based on the coordinates and form of the information delivery device that has been handshaked after the handshake is successful; at the same time, according to the warning scanning interval, continue to judge whether the accident warning basis still exists and compare and fit again to determine whether the next warning will continue to be released. If all warning conditions or refresh conditions are met, go to step 7, otherwise return to step 4;

[0055] Step 7: Generate warning information based on the acquired target data and issue accurate warning information through delivery or interaction devices; here, it refers to generating warning information according to the corresponding format for different information delivery devices, such as text for variable information display screens, control instruction suggestions for V2I, color blocks or continuous text for induction devices, etc.

[0056] Step 8: Continue to display the delivered warning information or refresh the subsequent warning information until the information is cancelled (the expected element disappears). After the warning is cancelled, enter a new cycle again (step 2).

[0057] This embodiment tested the main functions of the device in the "Smart Forest System - Smart Highway Safety" project and achieved satisfactory results.

[0058] Example 3: This example provides a specific application case based on the device in Example 1. This example installs the device on the roadside of the highway. The device uses an ARM processor to form a low-power control device, and is embedded with a 4G wireless module for accessing the Internet and interconnecting with cloud devices. The test point for testing the device was selected on a highway in Henan. The road section has been installed with an accident prevention device for low visibility induction, and the cumulative installation length of the prevention device is 21 kilometers. The terminal devices of the induction system (referred to as edge marks) are installed in pairs on the guardrails on both sides of the road. A pair of terminals (installed on the left and right guardrail columns) are installed at intervals of 20 meters along the road. The terminal display surface points to the direction of the oncoming vehicle. The warning lights are composed of two groups of red and yellow dot matrix color blocks. The dot matrix color block of each warning light can be switched to a 16×32 Chinese character through software; the induction terminal contains an infrared vehicle detector, a speed measurement and lane occupancy perception module installed at staggered intervals, an interconnected data link, an illuminance meter, an energy management module, a satellite positioning and timing module; a remote sensing road condition detector with an icing warning function is installed at the entrance of the starting point of the equipment installation. It can detect the wetness of the road surface, whether it is frozen, and the thickness of the ice layer. It also has an icing warning function and can display warning information about 1 hour in advance. The above two sets of traffic safety equipment are connected to the device of this embodiment through an invalid data link. This embodiment was used in the "Induction System-Smart Highway Safety" project in the test, and the effect was very good.

[0059] The road traffic related information that can be directly obtained by the device in this embodiment includes:

[0060] 1. Vehicles occupying the lane (the number of currently occupied lanes, collected by the speed measurement module of the induction terminal); 2. Vehicle speed (provided by the built-in speed measurement module of the induction device); 3. Vehicle type (estimated by the induction terminal vehicle detector); 4. Vehicle speed variable (vehicle speed change per unit time, provided by the induction terminal vehicle detector + speed measurement module); 5. Traffic flow (provided by the induction terminal vehicle detector); 6. Headway (estimated by the induction terminal vehicle detector and speed measurement module); 7. Congestion (estimated by the induction terminal vehicle detector); 8. Slow movement (estimated by the induction terminal Vehicle detector + speed measurement module estimate); 9. Collision guardrail (measured by the induction terminal vibration sensor); 10. Emergency stop event (provided by the induction terminal vehicle detector + speed measurement module); 11. Parking event (provided by the induction terminal vehicle detector); 12. Day and night speed variables (the speed change in a specific road section within 24 hours with the average speed per hour as the unit, and the basic data is jointly provided by the induction terminal vehicle detector + speed measurement module + satellite timing module); 13. Road surface dryness and humidity (provided by the remote sensing road condition detector); 14. Road area Water thickness (provided by remote sensing road condition detector); 15. Rainfall and change trend (provided by remote sensing road condition detector); 16. Road surface temperature: provided by remote sensing road condition detector); 17. Temperature change trend (temperature change slope in the past time, provided by remote sensing road condition detector and induction terminal); 18. Ambient temperature (provided by the built-in temperature sensor of the induction terminal); 18. Snow thickness and change trend (provided by remote sensing road condition detector); 20. Road surface illumination (measured by the illumination detection module of the induction terminal) ); 21. Visibility value (provided by the visibility meter configured in the induction terminal); 22. Standard time (provided by the satellite module); 23. Coordinates (the coordinates of each sensor are provided by the satellite module built into the induction terminal); 24. On-site data link (completed by the chain wireless interconnection data link built into the induction terminal); 25. Cloud interface (the induction device includes a gateway to the Internet interface and a VPN connected to the cloud device); 26. Interconnection data link (provided by the built-in data link of the induction terminal, including remote point operation).

[0061] Among the above 26 dimensional traffic data, the traffic meteorological data (provided by the remote sensing road condition detector) can generally cover about 5 to 10 kilometers (the data related to traffic meteorology can be regarded as unchanged within a radius of 5 to 10 kilometers per unit time), and the rest of the data are measured according to the coordinates of the collection point and the collection time. For example, when a vehicle passes through 10 sets of induction terminals, 10 sets of multi-dimensional traffic data will be generated. Each set of data corresponds to the coordinate address of a set of induction terminals (also known as edge marks). If a motor vehicle passes through these 10 sets of edge marks at a uniform speed in a lane, the 10 sets of data generated are basically the same. The difference is the time of passing each edge mark and the coordinate address of the 10 sets of edge marks. The above description only explains how the induction terminal and the road condition detector generate data and the similarities and differences of these data. Therefore, the information contained in these data is generated at a specific place and time and is continuously accumulated over time. Its data characteristics are continuous, so each data contains the timestamp and coordinate stamp when it is generated. The mark is identified by the data generating device or sensor when outputting the data, and the third-party data without identification is supplemented and embedded by this device when uploading.

[0062] Data features and applications in this embodiment: From the above data, a simple multi-dimensional profile of the situation at the traffic scene can be made, or the event can be described from a single dimension. In the profile composed of the above data, there is a lack of official specific data on traffic accidents (the data has not yet been opened to the public), so it is only possible to discover possible traffic accidents and road traffic data and traffic meteorological data at the time of the accident from the multi-dimensional data that has been collected. These data are handed over to the cloud for machine learning and training, and the final result is formed into a knowledge base downloaded to the device in this embodiment. Based on the active accident prevention algorithm, the device compares the current real-time collected data, finds possible accident signs and issues warnings.

[0063] The extraction and formation basis of accident features in this embodiment: the factors that may cause accidents are multidimensional, and basic data can be obtained from existing data, such as the braking distance of different models at different speeds, the impact of the wetness of the road surface on vehicle driving, the difference in static visibility and dynamic visibility on the driver's sight distance, the relationship between road type and safe braking (such as uphill, downhill, long downhill, curve), etc.; after combining the basic data and then based on the multi-dimensional traffic data obtained by actual measurement, it is possible to more effectively judge and identify possible or already occurred traffic accidents; for example, if it is detected that the vehicle speed variable data exceeds the sudden stop distance in the basic data, that is, the vehicle's emergency stop distance is less than the shortest braking distance in the basic data, it can be determined as a collision. When judging this distance, it is also necessary to add traffic meteorological data and road surface data to judge the accident more quickly and effectively. These judgment bases are integrated into a set of simple independent variable databases after machine learning and training in the cloud, and the dependent variable generated based on this is the required prediction result. In other words, if there is a slow-moving vehicle or an obstacle or a line of sight fault (fog) ahead on the current road, if there is a sensitive point about to enter the section (a sensitive point refers to a possible collision point with clear coordinates), based on the independent variables and measured data provided by the knowledge base, it can be determined whether the vehicle about to enter will have an accident and an effective warning can be issued when the vehicle enters the final control point (the coordinates of this point refer to the accident that cannot be avoided if the vehicle continues at this speed, and the accident can be avoided if the vehicle takes measures before entering the coordinates). The warning information can be displayed from the induction terminal by displaying the warning color (red) or displaying characters (continuously displaying characters on multiple induction terminals to form a warning word) or through a variable information board. If the vehicle has V2I, it can directly broadcast warnings to vehicles in the coordinate segment; in short, the function realized by this device is similar to the God's perspective, detecting and identifying possible traffic accidents from the top level and giving advance warnings to the relevant parties before the accident occurs.

[0064] The warning time in this embodiment: The setting of the warning starting point can be selected from two modes: time and distance, which is a parameterized setting. This embodiment adopts a countdown warning, starting with a countdown of 30 seconds, and the warning information changes in steps of 5 seconds (it can also be said that the shortest display time of the warning information is 5 seconds). During this period, the device will monitor in real time at a step length of less than 1 second whether the vehicle on the target section is within the warning range (during the display of the warning information, the dynamic variables of the monitored target are under high-speed dynamic monitoring internally). If the target has an accident risk, a warning message will be sent to the target section and vehicle in real time. It should be noted that when using the time warning mode, the warning starting point is related to the speed of the vehicle in the monitored section, and has little correlation with the distance; that is, the device measures the speed of the vehicle entering the monitored range, and sets the countdown starting point according to the time required between the speed of the entering vehicle and the danger point (the danger point refers to the physical location where the system detects obstacles, curves, line of sight faults, slippery sections, or other sensitive points that may cause traffic accidents). For example, if the distance to the danger point is 15 seconds and the speed is 60 kilometers, the distance to the warning starting point is the distance of 30 seconds at a speed of 60 kilometers per hour. Based on real-time data, it is possible to identify whether the monitored vehicle can pass safely through the knowledge base through fitting or comparison.

[0065] Distance warning in this embodiment: within a specified distance, the speed of the vehicle entering the "starting line" is measured and the time when the vehicle reaches the danger point is counted down. The degree of danger of the vehicle is identified according to the knowledge base. When it is within the safe range, no warning is required. When it is at a high degree of danger, a warning is issued in real time. Specifically, the length of the warning section is generally set according to the distance of the 10-second countdown at the highest speed of the section, wherein the countdown of 10 seconds is parameterized. The shorter the definition time, the higher the actual measurement accuracy (relatively fewer variables during the period), but the warning effect will be worse (lack of sufficient reaction time for the driver), otherwise the warning accuracy is lower (the longer the countdown time, the greater the variables generated during the period), but the warning effect is better (giving the driver enough time to take measures). If a distance warning is selected, it is more appropriate to select a 10-second countdown.

[0066] In summary, the device in this embodiment can receive or collect historical data from a data source when installed, and then transmit the historical data to a cloud device, which learns and trains to generate a knowledge base, and the knowledge base is downloaded to the local device via an interface unit; when in use, the device obtains real-time data of traffic events based on the on-site data source, and then compares and fits the probability of traffic accidents in the current traffic environment based on the local knowledge base and real-time data, and issues early warnings for events that meet the characteristics of traffic accidents. Finally, the device controls and / or drives the traffic safety equipment to complete corresponding instructions based on the warning learning, so as to warn the target to avoid possible accident hazards.

[0067] The above embodiments only express the implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An AI road surface event interaction and processing device, characterized by: include: An interface unit for connecting with cloud devices, data sources, and traffic safety equipment; The knowledge base is generated by learning and training of cloud devices and downloaded to the current device through the interface unit. It is used to fit and predict current traffic events based on real-time data provided by on-site data sources, and to issue early warnings for events that meet the characteristics of traffic accidents; The processing unit controls and / or drives the traffic safety equipment to complete corresponding instructions through the interface unit according to the warning issued by the knowledge base.

2. The device for AI road surface event interaction and processing according to claim 1, characterized in that: The data source includes sensors related to traffic events; the device reads data from different sensors based on the same time mark and different coordinate values ​​and uploads it to a cloud device, which learns the data and trains to form a knowledge base.

3. The device for AI pavement event interaction and processing according to claim 1, characterized in that: The sensor data related to the traffic events include road surface dryness and wetness, snow accumulation and snow thickness, ice and ice thickness, temperature drop slope, road surface dryness and wetness change efficiency, visibility value, rainfall, vehicle speed, vehicle type, lane occupancy, headway, vehicle speed variable, congestion, slow driving, collision with guardrail, road surface illumination, vehicle travel direction, road surface waterlogging, time vehicle speed variable, emergency stop event, parking event and the current status of road surface closure equipment.

4. The device for AI road surface event interaction and processing according to claim 1, characterized in that: The traffic safety equipment includes roadside or road surface induction devices, roadside or gantry variable information boards, roadside or road surface ice and snow melting devices, road surface automatic cleaning devices, warning lights, signal lights, tunnel induction devices, road opening and closing devices and tunnel locking devices.

5. The device for AI road surface event interaction and processing according to claim 2, characterized in that: The device is embedded with installation point coordinates, which are regionally unique and are used to receive a knowledge base with corresponding coordinates and management section coordinates downloaded from a cloud device; the learning and training of the knowledge base are based on the same time identifier and adjacent and continuous coordinates, and the trained knowledge base is downloaded from the cloud device to the corresponding device based on the installation point coordinates.

6. The device for AI pavement event interaction and processing according to claim 1, characterized in that: The device compares the data collected in different dimensions within a continuous coordinate range based on the same time with various factors that may cause accidents in the knowledge base. When data of certain dimensions in the real-time data generated at the road traffic scene are compared with the knowledge base and potential accident hazards are found, the device will drive the traffic safety equipment connected to the device to make an effective response based on the current sensor data and the solution corresponding to the knowledge base, so as to warn the target to avoid possible accident hazards.

7. The device for AI road surface event interaction and processing according to claim 2, characterized in that: All sensor data collected by the device are identified based on a reference time source and continuous adjacent coordinates and used as a reference for identifying multi-dimensional data of traffic events; the processing unit controls and / or drives traffic safety equipment based on the same time source and adjacent coordinate segments.

8. The device for AI road surface event interaction and processing according to claim 5, characterized in that: The historical data received or collected by the device are directly uploaded to the cloud device after identification processing and are not stored locally; the real-time data received or collected by the device is retained locally after being uploaded to the cloud device for identifying and comparing the current traffic event anomalies. This value is rolled over and stored for a certain period of time and is overwritten by new data after exceeding the preset time.

9. The device for AI road surface event interaction and processing according to claim 1, characterized in that: The device is capable of transmitting current comprehensive information and current control information to be driven to the manual management platform. The control information will be classified. General information will directly control and / or drive the traffic safety equipment after being broadcast to the manual management platform. Key information will be sent to the manual management platform as a prompt before issuing control and / or drive instructions. If no intervention instruction is received within a set time period, the control and / or drive instruction will be issued directly. If an intervention instruction is received, it will be executed according to the intervention instruction of the manual management platform.

10. The method of the device for AI pavement event interaction and processing according to any one of claims 1 to 9, characterized in that: During installation, the device is connected to the cloud device, the data source and the traffic safety device via the interface unit; the device is used to receive or collect historical data from the data source, and then the historical data is transmitted to the cloud device, and the cloud device learns and trains to generate a knowledge base, which is downloaded to the local device via the interface unit; When in use, the device obtains real-time data of traffic events based on on-site data sources, and then compares and fits the probability of traffic accidents in the current traffic environment based on the local knowledge base and real-time data, and issues early warnings for events that meet the characteristics of traffic accidents. Finally, the device controls and / or drives traffic safety equipment to complete corresponding instructions based on warning learning, so as to warn the target to avoid possible accident hazards.

Images