Collision early warning method, vehicle-mounted terminal and related system

Abstract

The invention discloses a collision early warning method, a vehicle-mounted terminal and a related system. According to the scheme, firstly, driving data of a first vehicle at the current moment are obtained; calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle by using the driving data at the current moment and the driving data of the first vehicle at the historical moment; the track prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is located at the current moment are sent to the cloud computing service system; and finally, collision early warning information returned by the cloud computing service system is received. Different from the existing technical scheme depending on V2X, the technical scheme of the invention can realize collision early warning without depending on the V2X technology, thereby saving the collision early warning cost, getting rid of the protocol constraint, enhancing the flexibility and universality of technology implementation, and effectively improving the road traffic safety level and the driving safety performance.

Description

Technical Field

[0001] The present application relates to the field of vehicle driving technology, and in particular to a collision warning method, a vehicle-mounted terminal and related systems. Background Art

[0002] With the advancement of science and technology and the development of society, the continuous maturity and popularization of motor vehicle manufacturing technology, the number of vehicles on the road has increased day by day, the urban traffic density has increased significantly, and vehicles are everywhere in urban and rural areas. The popularity of vehicles and the increase in the number of vehicles have increased the pressure on road traffic, and the risk of vehicle collisions has also increased. In order to reduce vehicle collisions, the industry usually uses vehicle to everything (V2X) technology to monitor vehicle driving conditions for coordination and early warning of vehicle driving.

[0003] The realization of V2X technology depends on the construction of V2X equipment and infrastructure, which means that a lot of manpower and material resources are needed. In addition, the vehicle collision warning realized by V2X technology is also subject to the constraints of V2X fixed format protocol in terms of communication. Therefore, the current solution of using V2X technology to realize vehicle collision warning is not easy to popularize and promote, and the cost is relatively high. Summary of the invention

[0004] Based on the above problems, this application provides a collision warning method, a vehicle-mounted terminal and related systems, with the aim of saving the cost of collision warning, breaking through the protocol constraints in the implementation of the collision warning scheme, improving the feasibility of collision warning, and improving driving safety performance.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the present application provides a collision warning method, which is applied to an on-board terminal of a first vehicle, wherein the on-board terminals of multiple vehicles including the first vehicle are all communicatively connected to a cloud computing service system; the method comprises:

[0007] Acquiring driving data of the first vehicle at the current moment; the driving data at the current moment includes driving parameters, location information, and a navigation path within a preset time period in the future from the current moment;

[0008] Calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments;

[0009] Sending the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system;

[0010] Receive collision warning information returned by the cloud computing service system; the collision warning information is sent by the cloud computing service system under the condition that the first vehicle and the other vehicles in the electronic fence will have a temporal and spatial intersection within the electronic fence based on the trajectory prediction sequence of the first vehicle and the trajectory prediction sequence of other vehicles in the electronic fence.

[0011] In an optional implementation, the calling of the conditional random field CRF model uses the driving data at the current moment and the driving data of the first vehicle at historical moments to generate a trajectory prediction sequence of the first vehicle, including:

[0012] Using the driving data at the historical moment to learn the parameters of the CRF model;

[0013] Preprocessing is performed on the basis of the driving data at the current moment to construct an observation sequence at the current moment;

[0014] Using the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and constructing the state sequence at the current moment;

[0015] The CRF model is used to output a state transfer matrix based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment; the state space includes a plurality of states to be predicted in trajectory prediction; the state transfer matrix describes the prediction probability of trajectory prediction transferring from one state to another state;

[0016] The trajectory prediction sequence of the first vehicle is obtained by decoding the state transfer matrix.

[0017] In an optional implementation, the obtaining the trajectory prediction sequence of the first vehicle by decoding the state transfer matrix includes:

[0018] If a plurality of trajectory prediction sequences of the first vehicle are obtained by decoding the state transfer matrix, a recommended trajectory prediction sequence is determined from the plurality of trajectory prediction sequences according to the navigation path of the first vehicle within a preset time period in the future from the current moment and / or the road condition information within a preset time period in the future from the current moment.

[0019] In an optional implementation, determining a recommended trajectory prediction sequence from the plurality of trajectory prediction sequences according to the navigation path of the first vehicle within a preset time period in the future from the current moment and / or the road condition information within a preset time period in the future from the current moment includes:

[0020] A trajectory prediction sequence that maximizes the value of a reward function is determined from the multiple trajectory prediction sequences through a reinforcement learning model as a recommended trajectory prediction sequence; the reward function is a function set according to the navigation path of the first vehicle within a preset time period in the future from the current moment, and / or the road condition information within a preset time period in the future from the current moment.

[0021] In an optional implementation, when sending the trajectory prediction sequence of the first vehicle to the cloud computing service system, the method further includes:

[0022] The vehicle identification code of the first vehicle is sent to the cloud computing service system; the vehicle identification code is used by the cloud computing service system to extract the size information and / or model information of the first vehicle, and to construct the spatial prediction trajectory of the first vehicle using the trajectory prediction sequence of the first vehicle; the collision warning information is specifically sent by the cloud computing service system under the condition that the spatial prediction trajectory of the first vehicle and the spatial prediction trajectory of other vehicles in the electronic fence have a temporal and spatial intersection.

[0023] A second aspect of the present application provides a collision warning method, which is applied to a cloud computing service system, wherein the cloud computing service system is communicatively connected to vehicle terminals of multiple vehicles including a first vehicle; the method comprises:

[0024] Receive driving data of the first vehicle at the current moment and an identifier of the electronic fence where the first vehicle is located at the current moment; the driving data at the current moment includes driving parameters, location information at the current moment, and a navigation path within a preset time period in the future from the current moment;

[0025] Calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments;

[0026] Predicting whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence;

[0027] If the prediction result is yes, collision warning information is sent to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminal of the other vehicle respectively.

[0028] In an optional implementation, the calling of the conditional random field CRF model uses the driving data at the current moment and the driving data of the first vehicle at historical moments to generate a trajectory prediction sequence of the first vehicle, including:

[0029] Using the driving data at the historical moment to learn the parameters of the CRF model;

[0030] Preprocessing is performed on the basis of the driving data at the current moment to construct an observation sequence at the current moment;

[0031] Using the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and constructing the state sequence at the current moment;

[0032] The CRF model is used to output a state transfer matrix based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment; the state space includes a plurality of states to be predicted in trajectory prediction; the state transfer matrix describes the prediction probability of trajectory prediction transferring from one state to another state;

[0033] The trajectory prediction sequence of the first vehicle is obtained by decoding the state transfer matrix.

[0034] In an optional implementation, the obtaining the trajectory prediction sequence of the first vehicle by decoding the state transfer matrix includes:

[0035] If a plurality of trajectory prediction sequences of the first vehicle are obtained by decoding the state transfer matrix, a recommended trajectory prediction sequence is determined from the plurality of trajectory prediction sequences according to the navigation path of the first vehicle within a preset time period in the future from the current moment and / or the road condition information within a preset time period in the future from the current moment.

[0036] In an optional implementation, determining a recommended trajectory prediction sequence from the plurality of trajectory prediction sequences according to the navigation path of the first vehicle within a preset time period in the future from the current moment and / or the road condition information within a preset time period in the future from the current moment includes:

[0037] A trajectory prediction sequence that maximizes the value of a reward function is determined from the multiple trajectory prediction sequences through a reinforcement learning model as a recommended trajectory prediction sequence; the reward function is a function set according to the navigation path of the first vehicle within a preset time period in the future from the current moment, and / or the road condition information within a preset time period in the future from the current moment.

[0038] In an optional implementation, when receiving the driving data of the first vehicle at the current moment, the method further includes:

[0039] Receiving a vehicle identification code of the first vehicle sent by the vehicle-mounted terminal of the first vehicle;

[0040] Extracting the size information and / or model information of the first vehicle according to the mapping relationship between the vehicle identification code, the vehicle size information and the vehicle model information;

[0041] constructing a spatial prediction trajectory of the first vehicle using the trajectory prediction sequence of the first vehicle and the extracted size information and / or model information;

[0042] The step of predicting whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence is specifically as follows:

[0043] Analyze whether the spatial prediction trajectory of the first vehicle and the spatial prediction trajectory of other vehicles in the electronic fence have a temporal and spatial intersection.

[0044] In an optional implementation, the cloud computing service system is deployed with multiple computing clusters and multiple electronic fence computing proxy units; the computing clusters are connected one-to-one with the electronic fence computing proxy units;

[0045] The method further comprises:

[0046] According to the traffic busyness of different electronic fences, a corresponding relationship between the electronic fence logo and the electronic fence calculation agent unit is constructed;

[0047] The receiving the driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is located at the current moment includes:

[0048] The driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is currently located are received through a first electronic fence calculation agent unit; the first electronic fence calculation agent unit is an electronic fence calculation agent unit among the multiple electronic fence calculation agents that has a corresponding relationship with the identifier of the electronic fence where the first vehicle is currently located;

[0049] The first electronic fence calculation agent unit reports the driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is currently located to the first computing cluster connected to the first electronic fence calculation agent unit.

[0050] A third aspect of the present application provides a vehicle-mounted terminal, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the collision warning method introduced in any implementation method of the first aspect.

[0051] The fourth aspect of the present application provides a cloud computing service system, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the collision warning method introduced in any implementation method of the second aspect.

[0052] A fifth aspect of the present application provides a collision warning system, the collision warning system comprising:

[0053] The vehicle-mounted terminal provided by the third aspect and the cloud computing service system provided by the fourth aspect.

[0054] Compared with the prior art, this application has the following beneficial effects:

[0055] In the technical solution of the present application, the driving data of the first vehicle at the current moment is first obtained; then the Conditional Random Field (CRF) model is called to use the driving data at the current moment and the driving data of the first vehicle at historical moments to generate the trajectory prediction sequence of the first vehicle; then the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located are sent to the cloud computing service system; finally, the collision warning information returned by the cloud computing service system is received. It can be seen that the technical solution of the present application can realize collision warning without relying on V2X technology, which is different from the collision warning method that relies on V2X technology in the existing technology, significantly saving the cost of collision warning, and getting rid of the protocol constraints, enhancing the flexibility and popularity of technology implementation, and effectively improving the road traffic safety level and driving safety performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0057] Figure 1A A schematic diagram of a collision warning scenario provided in an embodiment of the present application;

[0058] Figure 1B A schematic diagram of another collision warning scenario provided in an embodiment of the present application;

[0059] Figure 1C A flow chart of a collision warning method provided in an embodiment of the present application;

[0060] Figure 2A flowchart of another collision warning method provided in an embodiment of the present application;

[0061] Figure 3 A flowchart of another collision warning method provided in an embodiment of the present application;

[0062] Figure 4 A flowchart of another collision warning method provided in an embodiment of the present application;

[0063] Figure 5 A schematic diagram of a collision warning provided in an embodiment of the present application;

[0064] Figure 6 A cloud computing service system architecture diagram provided for an embodiment of the present application;

[0065] Figure 7 A schematic diagram of the structure of a vehicle-mounted terminal provided in an embodiment of the present application;

[0066] Figure 8 A schematic diagram of the structure of a cloud computing service system provided in an embodiment of the present application;

[0067] Fig.9A A schematic diagram of the structure of a collision warning system provided in an embodiment of the present application;

[0068] Fig. 9B A schematic diagram of the structure of another collision warning system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] As described above, in current vehicle applications, collision warning functions play an important role in improving driving safety, reducing traffic accidents, and promoting the development of intelligent transportation systems. However, in actual applications, V2X technology is usually used to monitor and coordinate vehicle driving in real time to avoid collisions and issue warnings. This processing solution relies on V2X equipment and infrastructure construction, and is also subject to the constraints of V2X fixed format protocols. It is not easy to popularize and promote, and the cost is high.

[0070] After research, the inventor proposed a collision warning method, a vehicle-mounted terminal, a cloud computing service system and a collision warning system. Figure 1A and Figure 1B The relevant scenarios of the implementation of the collision warning method are respectively presented, and the cloud computing service system and the vehicle-mounted terminals of multiple vehicles that are connected to the system for communication are involved in the scenarios.

[0071] First, the vehicle terminal is used to obtain the vehicle's current driving data. Then, the CRF model is used to generate the vehicle's trajectory prediction sequence using the current driving data and the historical driving data (the operation of generating the trajectory prediction sequence can be performed by the vehicle terminal, see Figure 1AThe scenario shown can also be implemented by a cloud computing service system, see Figure 1B Finally, the cloud computing service system determines whether a collision will occur based on the trajectory prediction sequence and the electronic fence mark where the vehicle is currently located, and sends the collision warning information to the vehicle terminal.

[0072] The process of generating the vehicle trajectory prediction sequence by the CRF model can be implemented in the vehicle terminal, such as Figure 1A As shown. The on-board terminal first obtains the driving data of the first vehicle at the current moment; then calls the conditional random field CRF model to use the driving data at the current moment and the driving data of the first vehicle at historical moments to generate a trajectory prediction sequence for the first vehicle; then sends the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system; finally, receives the collision warning information returned by the cloud computing service system. It should be noted that the on-board terminals of multiple vehicles including the first vehicle are all communicatively connected to the cloud computing service system, and the cloud computing service system can process the trajectory prediction sequences of multiple vehicles at the same time and provide collision warnings. The first vehicle here refers to any vehicle connected to the cloud computing service system. For ease of understanding, the subsequent description will be made from the perspective of the first vehicle.

[0073] The process of generating vehicle trajectory prediction sequences by CRF model can also be implemented in cloud computing service system, such as Figure 1B As shown. The cloud computing service system first receives the driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is located at the current moment; then calls the conditional random field CRF model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments; then predicts whether the first vehicle and other vehicles will have a spatiotemporal intersection in the electronic fence based on the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles in the electronic fence; if the prediction result is yes, then sends collision warning information to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminals of other vehicles respectively.

[0074] The technical solution of this application can achieve collision warning without relying on V2X technology, saving the cost of collision warning. At the same time, it gets rid of the constraints of the protocol, enhances the flexibility and popularity of technical implementation, and effectively improves the level of road traffic safety and driving safety performance.

[0075] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0076] Scenario 1 Implementation Example

[0077] This embodiment is a scenario in which the process of generating a vehicle trajectory prediction sequence using a CRF model is implemented in a vehicle terminal. Figure 1A shown.

[0078] The vehicle terminal refers to an intelligent device installed inside the vehicle for vehicle monitoring and management. The cloud computing service system refers to a system that allows users to obtain required computing resources and services (such as servers, storage, databases, software, etc.) in an on-demand and easily scalable manner through the Internet.

[0079] In an embodiment of the present application, the process of communication connection between the vehicle-mounted terminal and the cloud computing service system can be achieved through wireless communication technologies such as 4G / 5G network, WI-FI or Bluetooth.

[0080] See also Figure 1C , which is a flow chart of a collision warning method provided in an embodiment of the present application.

[0081] like Figure 1C As shown, the method comprises the following steps:

[0082] S101. Obtain driving data of the first vehicle at a current moment.

[0083] The driving data at the current moment includes driving parameters and location information at the current moment and a navigation path within a preset time period in the future from the current moment.

[0084] In the embodiment of the present application, the driving parameters at the current moment refer to various technical performance data displayed or recorded when the vehicle is running at a specific moment. For example, the vehicle model, the width and height of the vehicle, the vehicle identification code (VIN code), the current driving speed and the current driving acceleration. These parameters are of great significance for understanding the operating status, performance evaluation and safe driving of the vehicle.

[0085] In an example implementation, static parameters such as vehicle model, width, height, and VIN code can be obtained from the vehicle's electronic control unit (ECU) or vehicle information system. Dynamic parameters, such as the current driving speed and acceleration, can be obtained in real time through the vehicle's sensors.

[0086] The vehicle's location information can be obtained by receiving signals from satellites in real time through the Global Positioning System (GPS) installed on the vehicle and analyzing these signals. It can also be obtained using a navigation system such as a vehicle navigation system, and no specific limitation is made here.

[0087] The navigation path within a preset time period in the future from the current moment can be obtained by the driving route planned by the navigation system such as the vehicle navigation according to the preset destination information. Specifically, the preset time period can be pre-set according to at least one of the information such as the size of the electronic fence, the average driving speed of the vehicle, the road condition (congestion degree), etc. As an example, the preset time period can be set to 3 minutes, 5 minutes, etc. In practical applications, by obtaining the navigation path within the preset time period, it can be used to predict the movement tendency of the vehicle within the electronic fence.

[0088] For example, the navigation path planned by the in-vehicle navigation system based on the pre-set destination information is as follows: starting from the current position, continue to go straight along Avenue A, turn right after the first traffic light to enter Street B, drive 1.5 kilometers along Street B and enter the ramp, enter the ramp and drive 300 meters and merge into Highway C, drive 3 kilometers along Highway C, exit the highway at the third exit and enter Road D, drive 500 meters along Road D, turn left at the first intersection and enter Lane E, and continue driving 100 meters to reach the destination. If the duration of the future preset period is set to 3 minutes, the navigation path in the future preset period from the current moment in the above path is: go straight along Avenue A, turn right after the first traffic light to enter Street B, and drive 1.5 kilometers along Street B.

[0089] It should be noted that during the actual driving process, the driver may not follow the navigation route pre-planned by the navigation system at certain times due to factors such as actual road conditions and personal driving habits. At this time, since the driving route deviates from the pre-planned route of the navigation system, the navigation system will re-plan a new navigation route based on the current traffic conditions, road information and destination. For example: the vehicle has deviated from the original route. Please follow the following instructions: Drive along Avenue F for about 800 meters, turn right at the intersection of J Road, drive along Road H for 1 kilometer and enter the ramp, enter the ramp and drive 400 meters and merge into Highway C. Drive along Highway C for 3 kilometers, exit the highway at the third exit and enter Road D, drive along Road D for 500 meters, turn left at the first intersection and enter Lane E, and continue driving for 100 meters to reach the destination. If the duration of the future preset time period is set to 3 minutes, the navigation path in the future preset time period from the current moment in the above path is: drive about 800 meters along F Avenue, turn right at J intersection, drive 1 kilometer along H Road and enter the ramp, enter the ramp and drive 400 meters before merging into C Highway.

[0090] Therefore, the navigation path in this step within the preset time period in the future from the current moment may change with the change of actual driving behavior and is not fixed.

[0091] When using the conditional random field (CRF) model to predict the trajectory of the first vehicle, it is crucial to obtain the current driving data. This data directly reflects the current driving status of the vehicle and the impact of the external environment. As the real-time input of the model, this data, combined with historical driving data, can more accurately capture the dynamic changes of vehicle behavior, thereby improving the accuracy and real-time performance of trajectory prediction, which is of great significance for the realization of collision warning functions and helps to improve safety and efficiency.

[0092] S102: Calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments.

[0093] In this step, using the driving data at the current moment obtained in the previous step and the driving data of the first vehicle at historical moments, the vehicle's on-board terminal models and predicts the acquired data by calling the conditional random field CRF model to generate a corresponding trajectory prediction sequence for the first vehicle. The vehicle's driving data can be collected in real time at a certain collection interval, for example, once every 0.5 seconds. For this solution, the driving data obtained by S101 can be the current moment, and the same type of data obtained at moments before the current moment can all be regarded as driving data at historical moments. By combining the driving data at historical moments and the driving data at the current moment, the law of vehicle state changes can be grasped more accurately.

[0094] In an embodiment of the present application, the trajectory prediction sequence may refer to a driving trajectory within a period of time in the future starting from the current moment, which may include vehicle predicted position information at multiple moments with fixed time intervals.

[0095] The generated trajectory prediction sequence can be a position point type trajectory sequence or a trajectory sequence combined with the shape. If the generated trajectory prediction sequence is a position point type trajectory sequence, the position point represents the predicted position of the vehicle at multiple moments in the future. It should be noted that the position point can be the center point or the center of gravity of the vehicle. Considering that the vehicle has a certain spatial form, it is likely to be difficult to capture the spatial intersection of multiple vehicles only through the position point type trajectory sequence, thereby ignoring the collision risk of the vehicle. To this end, for this form of trajectory sequence, the subsequent vehicle terminal can further send information representing the vehicle size to the cloud computing service system, so that the cloud computing service system can combine the trajectory sequence and the vehicle size to realize the collision warning function that conforms to the vehicle's spatial trajectory. In addition, the vehicle terminal itself can also combine the position point type trajectory sequence and the size information of the vehicle in which it is located to calculate the trajectory sequence combined with the shape. Thereby, the trajectory information combined with the shape is directly sent to the cloud computing service system, reducing the workload of the cloud. The aforementioned vehicle size information can be obtained through the factory parameters such as the width and height of the vehicle, or it can be obtained by querying the vehicle model information, or it can be obtained based on the vehicle identification code.

[0096] S103: Sending the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system.

[0097] In the embodiment of the present application, the electronic fence refers to a focus area divided for a specific occasion, such as an intersection, an accident-prone area, or any other place that requires special monitoring. It is a virtual fence created by electronic means, which is different from the traditional physical fence (such as iron fence, wall, etc.), but relies on GPS, Bluetooth, WI-FI or other wireless communication technology to identify and track vehicles entering or leaving the focus area.

[0098] Each electronic fence has a unique identifier corresponding to it, so that the system can accurately identify and distinguish different monitoring areas. For example, the electronic fence at the first intersection is identified as electronic fence 1, the electronic fence at the second intersection is represented as electronic fence 2, and the electronic fence at the first accident-prone area is identified as electronic fence 3. The identifiers of electronic fences are usually configured in sequence according to the direction of the road or the area to which they belong. For example, electronic fence 2 is set between electronic fence 1 and electronic fence 3.

[0099] In specific applications, the shape, size and location of the electronic fence can be flexibly set to meet the monitoring needs in different scenarios. For example, a rectangular electronic fence with a side length of 300 meters or a circular electronic fence with a diameter of 300 meters can be set at an intersection to cover the entire intersection area. In accident-prone sections, more complex electronic fence shapes can be set according to the shape of the road and the distribution of accidents. At the same time, users can also set the sensitivity of the electronic fence, that is, the conditions that trigger the recognition of vehicle entry or exit (such as the speed of the vehicle entering the fence, the duration of stay, etc.), which are not specifically limited here.

[0100] In the embodiment of the present application, sending the trajectory prediction sequence of the first vehicle to the cloud computing service system also includes:

[0101] The vehicle identification code of the first vehicle is sent to the cloud computing service system.

[0102] It can be understood that the VIN code, as an example of a vehicle identification code, can play a one-to-one identification role for the vehicle.

[0103] The vehicle identification code is used by the cloud computing service system to extract the size information and / or model information of the first vehicle, and to construct the spatial prediction trajectory of the first vehicle using the trajectory prediction sequence of the first vehicle.

[0104] The spatial prediction trajectory is a prediction trajectory with a sense of space generated by obtaining the size and / or model information of the vehicle based on the VIN code of the vehicle, which reflects the three-dimensional image of the vehicle in space, making the collision warning more accurate and effective.

[0105] In the embodiment of the present application, after receiving the data sent in the above steps, the cloud computing service system processes and analyzes the data. If it is found that there is a time-space intersection, the cloud computing service system will generate a collision warning information.

[0106] The collision warning information is sent by the cloud computing service system when it is predicted that the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence based on the trajectory prediction sequence of the first vehicle and the trajectory prediction sequence of other vehicles within the electronic fence. Here, the occurrence of spatiotemporal intersection means that the spatial prediction trajectories of multiple vehicles overlap at the same time.

[0107] These collision warning information may include the estimated time and location of the collision, the vehicles involved, and recommended collision avoidance measures (such as deceleration, lane change, etc.). For example: The estimated time of the collision is 14:30:05 on April 15, 2023; the estimated location of the collision is 200 meters north of the intersection of City Center Avenue and Heping Road. Your vehicle is currently traveling north, and another vehicle of model A is currently traveling west and is expected to turn right from Heping Road to City Center Avenue. Please gradually reduce your speed to allow time for response to possible emergencies. If conditions permit and it is safe, please observe the conditions in the left and right lanes and prepare to change lanes to avoid collision with the other vehicle. In an example implementation, before sending data to the cloud computing service system, the data can be encapsulated and encrypted to ensure the security and integrity of the data during transmission. Encapsulation is to package the data in a certain format so that the recipient can parse it correctly; encryption is to encrypt the data using an encryption algorithm to prevent the data from being stolen or tampered with during transmission.

[0108] In an embodiment of the present application, during the data transmission process, the vehicle-mounted terminal can monitor the network status and data transmission status to ensure that the data can be successfully delivered to the cloud computing service system.

[0109] S104: Receive collision warning information returned by the cloud computing service system.

[0110] In the embodiment of the present application, the collision warning information can be sent to the vehicle terminal of the relevant vehicle via a wireless network. After receiving the collision warning information, the vehicle terminal will remind the driver of the potential collision risk through sound, vision or touch. After receiving the warning information, the driver can take corresponding collision avoidance measures to reduce or avoid the occurrence of collision.

[0111] In an example implementation, the vehicle terminal may also feed the driver's response (such as whether collision avoidance measures have been taken) to the cloud computing service system to further optimize the prediction and warning algorithms.

[0112] In the technical solution of this application, the above steps can realize collision warning without relying on V2X technology. Different from the collision warning method implemented by existing technology relying on V2X technology, it significantly saves the cost of collision warning, gets rid of the protocol constraints, enhances the flexibility and popularity of technology implementation, and effectively improves the road traffic safety level and driving safety performance. Compared with the collision warning method in the existing technology, it has obvious advantages.

[0113] Figure 2 Another flow chart of a collision warning method provided in an embodiment of the present application. In the embodiment introduced by this figure, a more detailed description is provided for the implementation of the collision warning applied to the vehicle-mounted terminal.

[0114] like Figure 2 The collision warning method shown includes the following steps:

[0115] S201. Obtain driving data of the first vehicle at the current moment.

[0116] This step is basically the same as the implementation method of S101 in the example described above, and will not be described in detail here. For related technical implementations, please refer to the introduction to S101 above.

[0117] S202: Utilize the driving data at the historical moments to learn the parameters of the CRF model.

[0118] In the embodiment of the present application, historical driving data is used as a training set to train the CRF model through maximum likelihood estimation, gradient descent or other optimization algorithms. At the same time, as new data is continuously generated, the parameters of the CRF model are regularly updated to adapt to changes in the traffic environment and changes in vehicle driving modes.

[0119] S203: Perform preprocessing on the basis of the driving data at the current moment to construct an observation sequence at the current moment.

[0120] In an exemplary implementation, the preprocessing based on the driving data at the current moment may be implemented by the following steps:

[0121] Step 1: First, verify the driving data obtained at the current moment to ensure the integrity and accuracy of the data. Check for missing values, outliers, or unreasonable data points. Remove or correct data records that are obviously wrong. For example, if the speed suddenly jumps to an unrealistic value, it may be necessary to treat it as an outlier and handle it. Make sure that the timestamps of all data are synchronized, or at least can be aligned to a common time base.

[0122] Step 2: Cluster the data into trajectories. Select a suitable trajectory clustering algorithm based on the data characteristics and analysis requirements. Optionally, a clustering algorithm such as K-means or DBSCAN can be selected. Apply the selected clustering algorithm to cluster the trajectory data and divide similar trajectories into the same cluster. Group the trajectories based on the clustering results, and possibly extract the central trajectory or representative trajectory of each cluster.

[0123] Step 3: Grid the data. Define the resolution and range of the grid based on the analysis requirements and data characteristics. The grid can be regular (such as a square or rectangular grid) or irregular (customized according to factors such as terrain, road network, etc.). Map the clustered trajectory data onto the defined grid. Each grid cell can contain trajectory points or trajectory segments that fall into the cell. Extract useful features from each grid cell, such as the number of trajectory points within the grid, average speed, direction changes, etc.

[0124] Representative observation points are selected in each grid cell. These observation points can be the center point of the grid, the average value of the trajectory points, or the projection of the cluster center.

[0125] Step 4: Sort the selected observation points by time or spatial relationship (such as the traversal order of the grid) to form an observation sequence.

[0126] Step 5: Encapsulate the observation sequence into a format suitable for CRF model input, including necessary metadata (such as timestamp, grid index, etc.).

[0127] Through the above steps, we can preprocess the current driving data and finally construct an observation sequence suitable for CRF model input. This process helps to extract and represent the vehicle's driving status and environmental information at the current moment, providing strong support for subsequent tasks such as trajectory prediction and collision warning.

[0128] S204: Using the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and constructing the state sequence at the current moment.

[0129] The feature function is the core part of the CRF model, which defines the relationship between the observation sequence and the state sequence. The feature function includes state features (depending on the observation and state of the current position) and transition features (depending on the state transition of the adjacent position).

[0130] In the embodiment of the present application, the state sequence at the current moment is constructed specifically through the following steps.

[0131] Step 1: First, define appropriate feature functions based on task requirements and data characteristics. These feature functions can capture important features in the observation sequence that are useful for predicting the state sequence, such as the distance, direction, speed, etc. between different positions.

[0132] Step 2: For each element in the observation sequence, apply the selected feature function to extract features. The feature function can calculate the relative position, velocity change, acceleration, direction change, etc. between observation points, as well as other features related to the task. The feature extraction results of each observation point are combined into a feature vector.

[0133] Step 3: Determine the initial state of the state sequence. In some cases, the initial state may be known, such as a parameter that has been confirmed in feature extraction. Or it may be uncertain and require random initialization, such as an unconfirmed parameter in feature extraction.

[0134] Step 4: Use characteristic functions and parameters to construct the state sequence at the current moment.

[0135] S205: Utilize the CRF model to output a state transfer matrix based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment.

[0136] In an embodiment of the present application, the state space includes a variety of states to be predicted in trajectory prediction, for example, all possible positions of the vehicle, all braking and starting state sets, etc.

[0137] The state transition matrix describes the predicted probability of transitioning from one state to another in the trajectory sequence.

[0138] In an embodiment of the present application, the transfer matrix can be learned and updated based on historical data.

[0139] S206. Obtain a trajectory prediction sequence of the first vehicle by decoding the state transfer matrix.

[0140] In an embodiment of the present application, the process of decoding the state transition matrix can be decoded using the Viterbi algorithm to find multiple possible trajectory sequences and their corresponding probabilities. However, in actual applications, there may be sequence trajectories among these predicted trajectory sequences that do not conform to the navigation route or road condition information. Therefore, a reinforcement learning model can be defined to make decisions on multiple possible trajectory sequences. The specific steps are as follows:

[0141] If a plurality of trajectory prediction sequences of the first vehicle are obtained by decoding the state transfer matrix, a trajectory prediction sequence that maximizes the value of the reward function is determined from the plurality of trajectory prediction sequences through a reinforcement learning model as a recommended trajectory prediction sequence according to the navigation path of the first vehicle within a preset period of time in the future from the current moment and / or the road condition information within a preset period of time in the future from the current moment.

[0142] The reward function is a function set according to the navigation path of the first vehicle within a preset time period in the future from the current moment, and / or the road condition information within a preset time period in the future from the current moment. For example, according to the degree of coincidence between the trajectory prediction sequence and the navigation path of the first vehicle within a preset time period in the future from the current moment, a higher reward is given to the trajectory prediction sequence with a high degree of coincidence, so as to encourage the model to choose a route that is gentler than the navigation path within the preset time period. According to the analysis of the road condition information within the preset time period in the future, such as the degree of congestion, traffic accidents, road construction, etc., the reward function can give positive rewards to the behavior of avoiding congested sections and choosing unobstructed paths, thereby prompting the model to make better decisions in real-time traffic conditions. If there are safety hazards in the navigation path or road condition information (such as sharp turns, steep downhills, accident-prone areas, etc.), the reward function can also be adjusted accordingly to give higher rewards to safe driving behaviors to ensure the safety of vehicle driving.

[0143] By setting such a reward function, the reinforcement learning model can maximize the cumulative probability to select the final trajectory sequence in the process of continuous trial and error and learning. A trajectory prediction sequence that maximizes the value of the reward function is used as the recommended trajectory prediction sequence to prevent the reinforcement learning model from falling into a local optimal solution and failing to find a global optimal solution.

[0144] S207: Sending the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system.

[0145] S208: Receive collision warning information returned by the cloud computing service system.

[0146] S207 and S208 are basically the same as the implementation methods of S103 and S104 in the example as the methods introduced above, and will not be described in detail here. For the relevant technical implementation, please refer to the introduction and description of S103 and S104 above.

[0147] In the embodiment of the present application, as the vehicle-mounted computing power is improved, the vehicle-mounted terminal can support CRF to realize the prediction trajectory sequence at the vehicle-mounted terminal. Doing so can greatly reduce the amount of data transmission, because the data does not need to be uploaded to the cloud computing service system for processing. The cloud computing service system only needs to judge the uploaded trajectory prediction sequence, which reduces the requirements for the cloud computing service system, thereby improving the real-time nature of the collision warning. At the same time, because all vehicle driving data is processed at the vehicle-mounted terminal, the risk of data leakage is reduced and user privacy protection is enhanced. Furthermore, the CRF trajectory prediction is performed by the vehicle-mounted terminal, and each vehicle has its own set of training models, which improves the adaptability of the vehicle and makes the prediction more accurate.

[0148] Scenario 2 Example

[0149] This embodiment is a scenario in which the process of generating a vehicle trajectory prediction sequence using a CRF model is implemented in a cloud computing service system. Figure 1B shown.

[0150] The cloud computing service system is connected to the vehicle terminals of multiple vehicles including the first vehicle in a communication connection process, which can be achieved through wireless communication technologies such as 4G / 5G network, WI-FI or Bluetooth.

[0151] See also Figure 3 , which is a flow chart of another collision warning method provided in an embodiment of the present application.

[0152] like Figure 3 As shown, the method comprises the following steps:

[0153] S301, receiving driving data of a first vehicle at a current moment and an identifier of an electronic fence where the first vehicle is currently located.

[0154] S302 , calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments.

[0155] S303: predict whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence.

[0156] S304: If the prediction result is yes, collision warning information is sent to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminals of other vehicles respectively.

[0157] S301-S304 is basically the same as the implementation method of S101-S104 in the example as the method introduced above, the only difference is that S101-S104 are applied to the vehicle's on-board terminal, and S301-S304 are applied to the cloud computing service system, which will not be elaborated here, and the relevant technical implementation can refer to the introduction of S101-S104 above.

[0158] Figure 4 A flowchart of another collision warning method provided in an embodiment of the present application. In the embodiment introduced by this figure, a more detailed description is provided for the implementation of collision warning applied to a cloud computing service system.

[0159] like Figure 4 The collision warning method shown includes the following steps:

[0160] S401, receiving driving data of a first vehicle at a current moment and an identifier of an electronic fence where the first vehicle is currently located.

[0161] S402 , calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments.

[0162] S403: Perform preprocessing on the basis of the driving data at the current moment to construct an observation sequence at the current moment.

[0163] S404: Use the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and construct the state sequence at the current moment.

[0164] S405: Utilize the CRF model to output a state transfer matrix based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment.

[0165] S406: Obtain a trajectory prediction sequence of the first vehicle by decoding the state transfer matrix.

[0166] S407 . Predict whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence based on the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence.

[0167] S408: If the prediction result is yes, collision warning information is sent to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminals of other vehicles respectively.

[0168] S401-S408 is basically the same as the implementation method of S201-S208 in the example introduced above, the only difference is that S201-S208 is applied to the vehicle's on-board terminal, and S401-S408 is applied to the cloud computing service system, which will not be elaborated here, and the relevant technical implementation can refer to the above introduction to S201-S208.

[0169] In an embodiment of the present application, CRF trajectory prediction is performed through a cloud service system. The cloud computing service system has powerful computing power and scalability, and can process large-scale vehicle trajectory data, thereby realizing a more complex and accurate collision warning algorithm. At the same time, in the cloud computing service system, the trajectory data of different vehicles and traffic participants can be shared and collaboratively processed, thereby building a more comprehensive traffic situation awareness capability. It helps to achieve cross-regional vehicle collision warning and collaborative control, and improve the overall traffic safety level. Furthermore, for vehicle manufacturers and traffic management departments, the use of cloud computing service systems can reduce the hardware cost and development difficulty of vehicle terminals, and the pay-as-you-go model of cloud computing can also help reduce long-term operating costs.

[0170] Figure 5 A collision warning schematic diagram is provided for an embodiment of the present application. Figure 5In the overhead view of the intersection shown, there is a vehicle 501 on the left side of the intersection and a vehicle 502 on the right side of the intersection. Vehicles 501 and 502 are located in the same electronic fence 503. By implementing this solution, it can be obtained that the estimated position of vehicle 501 in 3 seconds is 504, and the estimated position of vehicle 502 in 3 seconds is 505. A collision may occur. At this time, the cloud computing service system sends a collision warning to the on-board terminals of vehicle 501 and vehicle 502, indicating that a collision is expected to occur in 3 seconds, reminding the driver to take collision avoidance actions in advance, thereby effectively realizing collision warning.

[0171] Apart from Figure 5 In addition to the intersection scenario shown, collision warning can also be implemented on the road of moving vehicles. For example, on two adjacent lanes A and lane B, there are vehicles A and B running one after the other on lane A, and vehicle C is driving in the opposite direction on lane B. Vehicle B intends to overtake by entering lane B, and vehicle C is driving normally. Based on the technical solution of this application, a collision warning method is provided, which can realize the trajectory prediction of vehicles A, B and C, and predict and warn whether vehicle B will overtake and collide with vehicle C.

[0172] In another possible scenario, still in adjacent lanes A and B, vehicles a, b, and c are driving in lane A from front to back, where vehicle b stops somewhere in lane A after driving for a period of time due to a malfunction. Vehicles a and c still need to continue driving. The collision warning method provided by the technical solution of the present application can predict the trajectories of vehicles a, b, and c, and predict and warn whether vehicle c will collide with vehicle b when vehicle b stops midway.

[0173] In the embodiment of the present application, the cloud computing service system is deployed with multiple computing clusters and multiple electronic fence computing proxy units. The computing clusters are connected one-to-one with the electronic fence computing proxy units. Figure 6 The cloud computing service system architecture diagram shown in FIG.

[0174] The computing cluster is responsible for computing the data, predicting the future trajectory sequence and generating collision warning information. The electronic fence computing agent unit is responsible for reporting the data acquired by the vehicle terminal to the computing cluster in real time, and sending the collision warning information generated by the computing cluster to the vehicle terminal.

[0175] Based on the above deployment architecture, in an exemplary implementation, the above collision warning method applied to the cloud computing service system further includes:

[0176] According to the traffic busyness of different electronic fences, the corresponding relationship between the electronic fence logo and the electronic fence computing agent unit is constructed. In view of the great difference in traffic busyness of different electronic fences, for economic considerations, an electronic fence with very busy traffic can occupy a computing agent unit exclusively, while multiple electronic fences with less busy traffic can share a computing agent unit.

[0177] Based on the above deployment architecture, step S401 specifically includes:

[0178] The driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is currently located are received through the first electronic fence calculation agent unit. The first electronic fence calculation agent unit is an electronic fence calculation agent unit among the multiple electronic fence calculation agents that has a corresponding relationship with the identifier of the electronic fence where the first vehicle is currently located.

[0179] The first electronic fence calculation agent unit reports the driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is currently located to the first computing cluster connected to the first electronic fence calculation agent unit.

[0180] By executing the corresponding steps of the above method, collision warning can be realized, which can provide early warning of collision risks to the driver. Through the cloud server, a comprehensive and coordinated collision warning is realized, replacing the existing economically expensive signal base stations. At the same time, through the trajectory prediction and reinforcement learning decision of the CRF model, accurate and reliable collision warning is realized.

[0181] Based on the collision warning method provided in the above embodiment, the present application also provides a vehicle-mounted terminal for collision warning. Figure 7 Figure 1 is a schematic diagram of the structure of the vehicle-mounted terminal. Figure 7 As shown, the vehicle-mounted terminal includes the following components: a processor 701 and a memory 702 .

[0182] The memory 702 stores at least one instruction or at least one program.

[0183] The processor 701 is used to run the instructions or programs stored in the memory 702, and implement any one of the collision warning methods applied to the vehicle-mounted terminal provided by the embodiment of the present invention when executing the instructions or programs.

[0184] In an exemplary implementation, the processor 701 is used to run instructions or programs stored therein, and implement the following steps when executing the instructions or programs:

[0185] Acquire driving data of the first vehicle at the current moment;

[0186] The conditional random field (CRF) model is called to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at the historical moment;

[0187] Sending the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system;

[0188] Receive collision warning information returned by the cloud computing service system.

[0189] In another example implementation, the processor 701 is used to run instructions or programs stored therein, and implement the following steps when executing the instructions or programs:

[0190] Acquire driving data of the first vehicle at the current moment;

[0191] Using the driving data at the historical moment to learn the parameters of the CRF model;

[0192] Preprocessing is performed on the basis of the driving data at the current moment to construct an observation sequence at the current moment;

[0193] Using the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and constructing the state sequence at the current moment;

[0194] Outputting a state transfer matrix using the CRF model based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment;

[0195] Obtaining a trajectory prediction sequence of the first vehicle by decoding the state transfer matrix;

[0196] Sending the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system;

[0197] Receive collision warning information returned by the cloud computing service system.

[0198] For the convenience and simplicity of description, the specific working process of the vehicle-mounted terminal described above can refer to the corresponding process in the aforementioned collision warning method embodiment, and will not be repeated here.

[0199] Based on the collision warning method provided in the above embodiment, the present application also provides a cloud computing service system. Figure 8 This is a schematic diagram of the structure of the cloud computing service system. Figure 8 As shown, the cloud computing service system includes the following components: a processor 801 and a memory 802 .

[0200] The memory 802 stores at least one instruction or at least one program.

[0201] The processor 801 is used to run the instructions or programs stored in the memory 802, and implement any one of the collision warning methods applied to the cloud computing service system provided by the embodiment of the present invention when executing the instructions or programs.

[0202] In an exemplary implementation, the processor 801 is used to run instructions or programs stored therein, and implement the following steps when executing the instructions or programs:

[0203] Receiving driving data of the first vehicle at the current moment and an identifier of the electronic fence where the first vehicle is located at the current moment;

[0204] The conditional random field (CRF) model is called to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at the historical moment;

[0205] Predicting whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence;

[0206] If the prediction result is yes, collision warning information is sent to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminals of other vehicles respectively.

[0207] In another example implementation, the processor 801 is used to run instructions or programs stored therein, and implement the following steps when executing the instructions or programs:

[0208] Receiving driving data of the first vehicle at the current moment and an identifier of the electronic fence where the first vehicle is located at the current moment;

[0209] The conditional random field (CRF) model is called to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at the historical moment;

[0210] Preprocessing is performed on the basis of the driving data at the current moment to construct an observation sequence at the current moment;

[0211] Using the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and constructing the state sequence at the current moment;

[0212] Outputting a state transfer matrix using the CRF model based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment;

[0213] Obtaining a trajectory prediction sequence of the first vehicle by decoding the state transfer matrix;

[0214] Predicting whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence;

[0215] If the prediction result is yes, collision warning information is sent to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminals of other vehicles respectively.

[0216] For the convenience and simplicity of description, the specific working process of the cloud computing service system described above can refer to the corresponding process in the aforementioned collision warning method embodiment, and will not be repeated here.

[0217] Based on the collision warning method provided in the above embodiment, the present application also provides a collision warning system. The collision warning system includes a cloud computing service system and a vehicle-mounted terminal provided in the above embodiment, such as Fig.9A The specific working process of the vehicle terminal in the collision warning system described above can refer to the introduction in the embodiment of scenario 1, and will not be repeated here.

[0218] The present application also provides another collision warning system, which includes a vehicle-mounted terminal and a cloud computing service system provided in the above embodiment, such as Fig. 9B The specific working process of the cloud computing service system in the collision warning system described above can refer to the introduction in the second embodiment of the scenario, which will not be repeated here.

[0219] For the convenience and simplicity of description, the specific working process of the vehicle-mounted terminal in the collision warning system described above can refer to the corresponding process in the aforementioned collision warning method embodiment, and will not be repeated here.

[0220] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The system embodiment described above is merely schematic, in which the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0221] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A collision warning method, characterized in that: The method is applied to an on-board terminal of a first vehicle, and the on-board terminals of multiple vehicles including the first vehicle are all communicatively connected to a cloud computing service system; the method comprises: Acquiring driving data of the first vehicle at the current moment; the driving data at the current moment includes driving parameters, location information, and a navigation path within a preset time period in the future from the current moment; Calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments; Sending the trajectory prediction sequence of the first vehicle and the identifier of the electronic fence where the first vehicle is currently located to the cloud computing service system; Receive collision warning information returned by the cloud computing service system; the collision warning information is sent by the cloud computing service system under the condition that the first vehicle and the other vehicles in the electronic fence will have a temporal and spatial intersection within the electronic fence based on the trajectory prediction sequence of the first vehicle and the trajectory prediction sequence of other vehicles in the electronic fence.

2. The method according to claim 1, characterized in that The calling of the conditional random field CRF model generates a trajectory prediction sequence of the first vehicle by using the driving data at the current moment and the driving data of the first vehicle at historical moments, including: Using the driving data at the historical moment to learn the parameters of the CRF model; Preprocessing is performed on the basis of the driving data at the current moment to construct an observation sequence at the current moment; Using the characteristic function of the CRF model to extract features from the observation sequence at the current moment, and constructing the state sequence at the current moment; The CRF model is used to output a state transfer matrix based on a predefined state space, the observation sequence at the current moment, and the state sequence at the current moment; the state space includes a plurality of states to be predicted in trajectory prediction; the state transfer matrix describes the prediction probability of trajectory prediction transferring from one state to another state; The trajectory prediction sequence of the first vehicle is obtained by decoding the state transfer matrix.

3. The method according to claim 2, characterized in that The step of obtaining a trajectory prediction sequence of the first vehicle by decoding the state transfer matrix includes: If a plurality of trajectory prediction sequences of the first vehicle are obtained by decoding the state transfer matrix, a recommended trajectory prediction sequence is determined from the plurality of trajectory prediction sequences according to the navigation path of the first vehicle within a preset time period in the future from the current moment and / or the road condition information within a preset time period in the future from the current moment.

4. The method according to claim 3, characterized in that The step of determining a recommended trajectory prediction sequence from the plurality of trajectory prediction sequences based on the navigation path of the first vehicle within a preset time period in the future from the current moment and / or the road condition information within a preset time period in the future from the current moment comprises: A trajectory prediction sequence that maximizes the value of a reward function is determined from the multiple trajectory prediction sequences through a reinforcement learning model as a recommended trajectory prediction sequence; the reward function is a function set according to the navigation path of the first vehicle within a preset time period in the future from the current moment, and / or the road condition information within a preset time period in the future from the current moment.

5. The method according to any one of claims 1 to 4, characterized in that When sending the trajectory prediction sequence of the first vehicle to the cloud computing service system, the method further includes: The vehicle identification code of the first vehicle is sent to the cloud computing service system; the vehicle identification code is used by the cloud computing service system to extract the size information and / or model information of the first vehicle, and to construct the spatial prediction trajectory of the first vehicle using the trajectory prediction sequence of the first vehicle; the collision warning information is specifically sent by the cloud computing service system under the condition that the spatial prediction trajectory of the first vehicle and the spatial prediction trajectory of other vehicles in the electronic fence have a temporal and spatial intersection.

6. A collision warning method, characterized in that: Applied to a cloud computing service system, the cloud computing service system is communicatively connected with vehicle terminals of a plurality of vehicles including a first vehicle; the method comprises: Receive driving data of the first vehicle at the current moment and an identifier of the electronic fence where the first vehicle is located at the current moment; the driving data at the current moment includes driving parameters, location information at the current moment, and a navigation path within a preset time period in the future from the current moment; Calling a conditional random field (CRF) model to generate a trajectory prediction sequence of the first vehicle using the driving data at the current moment and the driving data of the first vehicle at historical moments; Predicting whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence; If the prediction result is yes, collision warning information is sent to the vehicle-mounted terminal of the first vehicle and the vehicle-mounted terminal of the other vehicle respectively.

7. The method according to claim 6, characterized in that When receiving the driving data of the first vehicle at the current moment, the method further includes: Receiving a vehicle identification code of the first vehicle sent by the vehicle-mounted terminal of the first vehicle; Extracting the size information and / or model information of the first vehicle according to the mapping relationship between the vehicle identification code, the vehicle size information and the vehicle model information; constructing a spatial prediction trajectory of the first vehicle using the trajectory prediction sequence of the first vehicle and the extracted size information and / or model information; The step of predicting whether the first vehicle and the other vehicles will have a spatiotemporal intersection within the electronic fence according to the trajectory prediction sequence of the first vehicle and the trajectory prediction sequences of other vehicles within the electronic fence is specifically as follows: Analyze whether the spatial prediction trajectory of the first vehicle and the spatial prediction trajectory of other vehicles in the electronic fence have a temporal and spatial intersection.

8. The method according to claim 6, characterized in that The cloud computing service system is deployed with multiple computing clusters and multiple electronic fence computing proxy units; the computing clusters are connected one-to-one with the electronic fence computing proxy units; The method further comprises: According to the traffic busyness of different electronic fences, a corresponding relationship between the electronic fence logo and the electronic fence calculation agent unit is constructed; The receiving the driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is located at the current moment includes: The driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is currently located are received through a first electronic fence calculation agent unit; the first electronic fence calculation agent unit is an electronic fence calculation agent unit among the multiple electronic fence calculation agents that has a corresponding relationship with the identifier of the electronic fence where the first vehicle is currently located; The first electronic fence calculation agent unit reports the driving data of the first vehicle at the current moment and the identifier of the electronic fence where the first vehicle is currently located to the first computing cluster connected to the first electronic fence calculation agent unit.

9. A vehicle-mounted terminal, characterized in that: It comprises a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the collision warning method as described in any one of claims 1-5.

10. A cloud computing service system, characterized in that: It comprises a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the collision warning method as described in any one of claims 6-8.

Images