Safety driving assistance method and device for commercial vehicle, electronic equipment and storage medium

Abstract

The invention provides a commercial vehicle safe driving assistance method and device, electronic equipment and a storage medium, and relates to the vehicle safe driving technology, and the method comprises the steps: obtaining the current motion track information of a target commercial vehicle in the driving process; inputting the current motion track information into the lane changing track prediction model, performing motion track global feature extraction on the current motion track information to obtain the current motion track feature of the target commercial vehicle, and performing optimal lane changing track prediction based on the current motion track feature and the historical motion track feature. Obtaining a lane changing track prediction result output by the lane changing track prediction model; and feeding back the lane changing track prediction result to other vehicles around the target operating vehicle, so that the other vehicles carry out safe driving early warning when identifying the lane changing intention of the target operating vehicle from the lane changing track prediction result. The method can help a driver make decisions and take measures in advance, and collision risks caused by manual lane changing are avoided.

Description

Technical Field

[0001] The present invention relates to vehicle safe driving technologies, and particularly to a safe driving assistance method, device, electronic device and storage medium for operating vehicles. Background Art

[0002] Currently, in traffic safety accidents related to road traffic, more than 80% of road traffic safety accidents are caused by human factors. Among them, lane changing is the most common operation of vehicle drivers during driving. Especially for large operating vehicles such as commercial vehicles and buses, the probability of accidents during lane changing is greater, and the consequences after accidents are more serious. Therefore, how to ensure the driving experience of operating vehicle drivers during driving and avoid traffic accidents caused by lane changing of surrounding vehicles has become a technical problem to be solved urgently at present. Summary of the Invention

[0003] The purpose of the present invention is to provide a safe driving assistance method, device, electronic device and storage medium for operating vehicles, so as to solve the technical problems that existing operating vehicle drivers have poor driving experience during driving and cannot avoid traffic accidents caused by lane changing of surrounding vehicles. By predicting the optimal lane change trajectory based on the current motion trajectory features extracted at the current moment and the historical motion trajectory features stored at the previous moment, the accuracy and stability of the lane change trajectory prediction of the vehicle can be greatly improved, and it can also help the driver make decisions and take measures in advance to avoid the collision risk caused by human lane changing as early as possible and improve the driver's driving experience. The many technical effects that can be produced by the preferred technical solutions among the many technical solutions provided by the present invention are described in detail below.

[0004] To achieve the above purpose, the present invention provides the following technical solutions: The present invention provides a safe driving assistance method for operating vehicles, including: Obtaining the current motion trajectory information of a target operating vehicle during driving; Inputting the current motion trajectory information into a lane change trajectory prediction model, extracting the global motion trajectory features of the current motion trajectory information through the lane change trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle, and then predicting the optimal lane change trajectory based on the current motion trajectory features and historical motion trajectory features to obtain the lane change trajectory prediction result of the target operating vehicle output by the lane change trajectory prediction model; Feeding back the lane change trajectory prediction result to other vehicles around the target operating vehicle, so that when the other vehicles recognize the lane change intention of the target operating vehicle from the lane change trajectory prediction result, they can give a safe driving warning.

[0005] According to a safe driving assistance method for operating vehicles provided by the present invention, the training process of the lane-changing trajectory prediction model includes: Determine the respective historical target lane-changing trajectory information of different types of sample operating vehicles from a preset vehicle movement trajectory information set; Use each of the historical target lane-changing trajectory information to train an initial prediction model containing the Transformer encoder-MLSTM network, and determine the lane-changing trajectory prediction model according to the training results.

[0006] According to a safe driving assistance method for operating vehicles provided by the present invention, the step of determining the respective historical target lane-changing trajectory information of different types of sample operating vehicles from a preset vehicle movement trajectory information set includes: Eliminate the useless attribute information that does not belong to the vehicle driving attributes in the preset vehicle movement trajectory information set to obtain a first candidate movement trajectory information set; Add the lateral speed and lateral acceleration of each sample operating vehicle in a preset section, the driving speed and longitudinal acceleration of the vehicle in front of each sample operating vehicle, and the driving speed and longitudinal acceleration of the vehicle behind each sample operating vehicle to the first candidate movement trajectory information set to obtain a second candidate movement trajectory information set; Select the movement trajectory information of large operating vehicles and small operating vehicles in a preset lane respectively from the second candidate movement trajectory information set, and determine each of the historical target lane-changing trajectory information based on the selection result.

[0007] According to a safe driving assistance method for operating vehicles provided by the present invention, the step of determining each of the historical target lane-changing trajectory information based on the selection result includes: Perform a determination process of vehicle ID reuse on the selection result; Extract lane-changing trajectory information from the determination process result to obtain each historical candidate lane-changing trajectory information; Perform a cleaning process on the missing data of each historical candidate lane-changing trajectory information, and perform wavelet denoising on the cleaning process result to obtain each of the historical target lane-changing trajectory information.

[0008] According to a safe driving assistance method for operating vehicles provided by the present invention, performing wavelet denoising on the cleaning process result includes: Perform wavelet denoising on the cleaning result using a first preset wavelet, or perform wavelet denoising on the cleaning result using a second preset wavelet; Wherein, the first preset wavelet is any wavelet between Sym4 wavelet and Sym11 wavelet, and the second preset wavelet is Db4 wavelet or Db5 wavelet.

[0009] A safe driving assistance method for operating vehicles provided by the present invention. The global motion trajectory features of the current motion trajectory information are extracted through the lane-changing trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle, including: When the lane-changing trajectory prediction model includes a trained multi-head attention mechanism layer, a trained feed-forward network layer, and a trained residual normalization layer, Use the trained multi-head attention mechanism layer to extract motion trajectory features of the current motion trajectory from multiple different angular orientations to obtain the first vehicle motion trajectory features; Use the trained feed-forward network layer to perform a fully connected process on the first vehicle motion trajectory features to obtain the second vehicle motion trajectory features; Use the trained residual normalization layer to perform feature normalization on the first vehicle motion trajectory features and the second vehicle motion trajectory features to obtain the current motion trajectory features.

[0010] A safe driving assistance method for operating vehicles provided by the present invention. The optimal lane-changing trajectory is predicted based on the current motion trajectory features and historical motion trajectory features, and the lane-changing trajectory prediction result of the target operating vehicle output by the lane-changing trajectory prediction model is obtained, including: When the lane-changing trajectory prediction model includes a trained MLSTM network, use the trained MLSTM network to extract motion trajectory sequence features from the current motion trajectory features and the historical motion trajectory features, and then perform a lane-changing trajectory prediction on the extracted motion trajectory sequence features to obtain the lane-changing trajectory prediction result.

[0011] The present invention also provides a safe driving assistance device for operating vehicles, including the following modules.

[0012] A motion trajectory acquisition module for acquiring the current motion trajectory information of a target operating vehicle during driving; A lane-changing trajectory prediction module for inputting the current motion trajectory information into a lane-changing trajectory prediction model, extracting the global motion trajectory features of the current motion trajectory information through the lane-changing trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle, and then predicting the optimal lane-changing trajectory based on the current motion trajectory features and historical motion trajectory features to obtain the lane-changing trajectory prediction result of the target operating vehicle output by the lane-changing trajectory prediction model; A safe driving assistance module for feeding back the lane-changing trajectory prediction result to other vehicles around the target operating vehicle, so that when the other vehicles recognize the lane-changing intention of the target operating vehicle from the lane-changing trajectory prediction result, a safe driving warning is issued.

[0013] The present invention provides a method, a device, an electronic device and a storage medium for assisting safe driving of operating vehicles. The method for assisting safe driving of operating vehicles can provide a lane-changing trajectory prediction model with a global feature extraction function, a historical feature data storage function and a final lane-changing trajectory prediction function. When obtaining the current motion trajectory information of a target operating vehicle during driving, the current motion trajectory information only needs to be input into the lane-changing trajectory prediction model to sequentially perform global feature extraction of the motion trajectory and optimal lane-changing trajectory prediction, so as to obtain the lane-changing trajectory prediction result of the target operating vehicle. Then, the lane-changing trajectory prediction result is fed back to other vehicles around the target operating vehicle, so that other vehicles can perform a safe driving warning when identifying the lane-changing intention of the target operating vehicle from the lane-changing trajectory prediction result. In this way, by means of optimal lane-changing trajectory prediction based on the current motion trajectory features extracted at the current moment and the historical motion trajectory features stored at the previous moment, the accuracy and stability of the lane-changing trajectory prediction of the vehicle can be greatly improved, and it can also help the driver make decisions and take measures in advance to avoid the collision risk caused by manual lane-changing as early as possible, and improve the driving experience of the driver. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0015] Figure 1 It is a schematic flow chart of the method for assisting safe driving of operating vehicles provided by the embodiments of the present invention; Figure 2 It is a schematic diagram of the road surface condition of the data collection section provided by the embodiments of the present invention; Figure 3 It is a schematic diagram for comparing the left lane-changing trajectory prediction results of different models provided by the embodiments of the present invention; Figure 4 It is a schematic diagram for comparing the right lane-changing trajectory prediction results of different models provided by the embodiments of the present invention; Figure 5 It is a schematic diagram for comparing the RMSE values of the lane-changing lateral trajectories provided by the embodiments of the present invention; Figure 6 It is a schematic diagram for comparing the RMSE values of the lane-changing longitudinal trajectories provided by the embodiments of the present invention; Figure 7 It is a schematic diagram for comparing the average deviation distances under different time window lengths provided by the embodiments of the present invention; Figure 8Schematic structural diagram of the safe driving assistance device for operating vehicles provided by the embodiments of the present invention; Figure 9 Schematic structural diagram of the electronic device provided by the embodiments of the present invention. Detailed implementation manners

[0016] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0017] Currently, road traffic accidents have always received wide attention from society. Each occurrence will cause irreparable harm and property losses to the drivers and passengers involved in the accident. Moreover, once a traffic accident occurs, it will immediately reduce the traffic capacity and traffic efficiency of the road, cause traffic jams of varying degrees, and at the same time have varying degrees of impact on adjacent vehicles, and even threaten the personal safety of surrounding drivers. Related technologies show that in traffic safety accidents related to road traffic, more than 80% of road traffic safety accidents are caused by human factors. Changing lanes is the most common operation during vehicle driving and is likely to cause accidents due to various reasons; especially large operating vehicles such as commercial vehicles and buses have significant differences in terms of contour size, load, power performance, and lane space occupancy compared with passenger cars. They have higher requirements for the lane-changing gap space during the lane-changing process, have a poorer field of vision during driving, have a greater probability of accidents during the lane-changing process, and more serious consequences after accidents occur.

[0018] In summary, in-depth research on the lane-changing intention recognition and trajectory prediction of vehicles can not only ensure the safe driving of vehicles on the road and effectively reduce the accident rate, but also have a better driving assistance effect on operating vehicles with a larger total mass such as buses and trucks, which is also the main research object of the present invention. Research on the lane-changing intention recognition and lane-changing trajectory prediction of vehicles also has great significance for aspects such as the driving safety of large vehicles and the improvement of road traffic efficiency. With the development of technology, the methods and efficiency of obtaining vehicle data have made great progress, and it is more reasonable and convenient to identify and predict the lane-changing behavior of vehicles through trajectory data.

[0019] Therefore, how to ensure the driving experience of the driver of the operating vehicle during driving and avoid traffic accidents caused by the lane-changing of surrounding vehicles has become a technical problem to be solved urgently at present.

[0020] To solve the above technical problems, the present invention provides a safe driving assistance method, device, electronic device and storage medium for operating vehicles. By predicting the optimal lane-changing trajectory based on the current motion trajectory features extracted at the current moment and the historical motion trajectory features stored at the previous moment, the accuracy and stability of the lane-changing trajectory prediction of the vehicle can be greatly improved, and it can also help the driver make decisions and take measures in advance to avoid the collision risk caused by manual lane-changing as early as possible, thereby improving the driving experience of the driver.

[0021] The following will be combined with Figures 1-9 Describe the safe driving assistance method, device, electronic device and storage medium for operating vehicles. The execution subject of the safe driving assistance method for operating vehicles is an electronic device or a server, and both the electronic device and the server can be communicatively connected to each operating vehicle in the lane to be measured, and both at least have a positioning function and a sensing function; the electronic device can be a personal computer (PC), a portable device, a laptop computer, a smart phone, a tablet computer, a portable wearable device and other devices cloud server, and the server can refer to a single server or a server cluster, a cloud server, etc. composed of multiple servers. The present invention does not specifically limit the specific form of the electronic device or the server. Further, the safe driving assistance method for operating vehicles can also be applied to a safe driving assistance device for operating vehicles provided in the electronic device or the server, and the safe driving assistance device for operating vehicles can be implemented by software, hardware or a combination of both. The following takes the execution subject of the safe driving assistance method as an electronic device as an example to describe the safe driving assistance method.

[0022] Next, the safe driving assistance method provided by the present invention will be described in detail through the following several exemplary embodiments.

[0023] Refer to Figure 1 , which is a schematic flow chart of the safe driving assistance method provided by the embodiment of the present invention. As Figure 1 shown, the safe driving assistance method includes the following steps 110 to 130.

[0024] Step 110: Obtain the current motion trajectory information of the target operating vehicle during driving.

[0025] Among them, the number of target operating vehicles is at least 1, and each target operating vehicle can be any operating vehicle currently driving on the lane to be studied, and each target operating vehicle can be a commercial vehicle, a passenger vehicle or other large operating transport vehicles, such as large trucks, big rigs.

[0026] The current motion trajectory information may include, but is not limited to, the current positioning data and current motion data of the target operating vehicle; the current motion data here includes, but is not limited to, the current speed and / or current acceleration of the target operating vehicle.

[0027] Specifically, the electronic device can utilize its positioning function, for example, use the Global Positioning System (GPS) to collect the position data of the target operating vehicle at the current moment. The position data here can be the geographical coordinate position of the target operating vehicle at the current moment, such as longitude and latitude; it can also be the specific location where the target operating vehicle is located at the current moment, such as a certain street or a certain lane.

[0028] Meanwhile, the electronic device can also utilize its sensing function, such as an acceleration sensor and / or a speed sensor to obtain the driving data of the target operating vehicle at the current moment. The driving data here can be the speed and / or acceleration of the target operating vehicle at the current moment.

[0029] In this way, the electronic device can generate the current motion trajectory information based on the position data and driving data of the target operating vehicle at the current moment during the driving process.

[0030] Alternatively, the electronic device can also directly utilize the existing advanced image recognition technology to obtain the current motion trajectory information of the target operating vehicle during driving. The present invention does not make specific limitations on this.

[0031] Step 120: Input the current motion trajectory information into the lane-changing trajectory prediction model, extract the global motion trajectory features of the current motion trajectory information through the lane-changing trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle, and then perform optimal lane-changing trajectory prediction based on the current motion trajectory features and historical motion trajectory features to obtain the lane-changing trajectory prediction result of the target operating vehicle output by the lane-changing trajectory prediction model.

[0032] Among them, the historical motion trajectory feature is the motion trajectory feature of the target operating vehicle at a historical moment; the historical moment here can be the previous moment of the current moment.

[0033] Specifically, when the electronic device obtains the current movement trajectory information of the target operating vehicle, it can use the pre-trained lane-changing trajectory prediction model to obtain the lane-changing trajectory prediction result of the target operating vehicle. That is, when the lane-changing trajectory prediction model at least has the functions of global feature extraction, historical feature data storage, and final lane-changing trajectory prediction, the current movement trajectory information of the target operating vehicle can be input into the lane-changing trajectory prediction model for global feature extraction of the movement trajectory, and then combined with the movement trajectory characteristics of the target operating vehicle at historical moments to predict the optimal lane-changing trajectory. The optimal lane-changing trajectory prediction here is to perform multiple lane-changing position predictions on the movement trajectory feature sequence composed of the movement trajectory characteristics of multiple target time windows between the previous moment and the current moment.

[0034] Step 130: Feed back the lane-changing trajectory prediction result to other vehicles around the target operating vehicle, so that when other vehicles recognize the lane-changing intention of the target operating vehicle from the lane-changing trajectory prediction result, they can give a safety driving warning.

[0035] Among them, there is at least one other vehicle around the target operating vehicle; each other vehicle can be other non-operating vehicles such as small passenger cars, or it can also be an operating vehicle; no specific limitation is made here.

[0036] Specifically, for the lane-changing trajectory prediction result of the target operating vehicle output by the lane-changing trajectory prediction model, it can be fed back to other vehicles around the target operating vehicle to prompt other vehicles, and then the intelligent driving function on the vehicle can recognize the lane-changing intention of the target operating vehicle from the lane-changing trajectory prediction result, so as to timely remind the driver of the corresponding vehicle of safe driving when recognizing the lane-changing intention of the target operating vehicle, thereby avoiding the occurrence of traffic accidents.

[0037] The operating vehicle safety driving assistance method provided by the present invention can greatly improve the accuracy and stability of the lane-changing trajectory prediction of the vehicle by predicting the optimal lane-changing trajectory based on the current movement trajectory features extracted at the current moment and the historical movement trajectory features stored at the previous moment. It can also help the driver make decisions and take measures in advance to avoid the collision risk caused by manual lane-changing as early as possible, and improve the driving experience of the driver.

[0038] Based on the above Figure 1 In an example embodiment of the operating vehicle safety driving assistance method shown, in step 120, the lane-changing trajectory prediction model is used to extract the global features of the movement trajectory of the current movement trajectory information to obtain the current movement trajectory features of the target operating vehicle, and its specific implementation process can be achieved through the following steps.

[0039] When the lane-changing trajectory prediction model includes a trained multi-head attention mechanism layer, a trained feed-forward network layer, and a trained residual normalization layer, first, use the trained multi-head attention mechanism layer to extract motion trajectory features from the current motion trajectory in multiple different angular orientations to obtain the first vehicle motion trajectory features; then further use the trained feed-forward network layer to perform a fully connected process on the first vehicle motion trajectory features to obtain the second vehicle motion trajectory features; then, use the trained residual normalization layer to perform feature normalization on the first vehicle motion trajectory features and the second vehicle motion trajectory features to obtain the current motion trajectory features.

[0040] Specifically, the lane-changing trajectory prediction model includes a trained (Transformer) encoder and a trained multi-sequence long short-term memory (Multi-Layer Long Short-Term Memory, MLSTM), and the trained MLSTM network serves as a post-processing layer of the trained Transformer encoder. The trained Transformer encoder includes a trained multi-head attention mechanism layer, a trained feed-forward network layer, and a trained residual normalization layer. In this way, the trained multi-head attention mechanism layer can use the multi-head attention mechanism to extract motion trajectory features from the current motion trajectory in multiple different angular orientations to obtain the first vehicle motion trajectory features; then use the trained feed-forward network layer to perform a fully connected process on the first vehicle motion trajectory features to obtain the second vehicle motion trajectory features; the feed-forward network layer can include two fully connected layers; finally, use the trained residual normalization layer to perform feature normalization on the first vehicle motion trajectory features and the second vehicle motion trajectory features to obtain the current motion trajectory features; among them, the residual normalization layer is composed of a residual connection and layer normalization. The residual connection is set to prevent the model gradient from vanishing, and the layer normalization function used for layer normalization can be used to reduce data bias and prevent overfitting.

[0041] Based on the above Figure 1 For the commercial vehicle safe driving assistance method shown above, in one example embodiment, in step 120, based on the current motion trajectory features and historical motion trajectory features, an optimal lane-changing trajectory prediction is performed to obtain the lane-changing trajectory prediction result of the target commercial vehicle output by the lane-changing trajectory prediction model, and its specific implementation process can be achieved through the following steps.

[0042] When the lane-changing trajectory prediction model includes a trained MLSTM network, use the trained MLSTM network to extract motion trajectory sequence features from the current motion trajectory features and historical motion trajectory features, and then perform a lane-changing trajectory prediction on the extracted motion trajectory sequence features to obtain the lane-changing trajectory prediction result.

[0043] Specifically, the MLSTM network is obtained by stacking multiple LSTM network layers, which is convenient for long-term continuous memory of vehicle information. At the same time, it can also predict the vehicle's motion trajectory information more accurately. The LSTM based on layer stacking can better predict the transformation trend of the surrounding vehicle trajectory data sequence information. In this way, by using each trained LSTM network to perform multiple lane change position predictions on the motion trajectory feature sequence composed of the motion trajectory features of multiple target time windows between the previous moment and the current moment, the optimal lane change trajectory prediction can be realized, and thus the lane change trajectory prediction result of the target operating vehicle can be obtained.

[0044] Based on the above Figure 1 For the safe driving assistance method of the operating vehicle shown above, in an exemplary embodiment, the training process of the lane change trajectory prediction model can be realized through the following steps.

[0045] First, determine the respective historical target lane change trajectory information of different types of sample operating vehicles from the preset vehicle motion trajectory information set; then, use each historical target lane change trajectory information to train the initial prediction model containing the Transformer encoder-MLSTM network, and determine the lane change trajectory prediction model according to the training result.

[0046] Specifically, considering the long-term dependence learning ability of the MLSTM network, as well as the efficient parallel processing ability and powerful context representation ability of the Transformer model, the initial prediction model can be constructed by taking the MLSTM network as the post-processing layer of the Transformer encoder, so as to ensure that the initial prediction model has both the advantages of the Transformer encoder in extracting vehicle motion trajectory features and the advantages of the MLSTM network in time series reconstruction. In this way, by training the initial prediction model until the training end condition is met and then ending the training, and determining the intermediate prediction model corresponding to the end of training as the lane change trajectory prediction model, it is convenient to directly use the lane change trajectory prediction model for lane change trajectory prediction in the future.

[0047] It should be noted that the preset vehicle motion trajectory information set is part of the data in the open-source trajectory data set. The trajectory data set is obtained by collecting the vehicle trajectory information of vehicles driving on four types of roads in different regions of the United States through high-definition cameras; for example, the four types of roads are the US-101 highway in Los Angeles, the Interstate 80 (I-80) in San Francisco, the Lankershim Highway in Los Angeles, and the Peachtree Highway in the state of Georgia; the data sampling frequency is 10Hz, and the accurate position of each vehicle in the corresponding road section area can be determined at an interval of 0.1s, so as to generate the continuous motion trajectory information of each vehicle for studying the real behavior between vehicles; the trajectory data set can be generated according to the data collection situation of each road section shown in Table 1.

[0048]

[0049] Since the research object of the present invention is the lane-changing intention of drivers of large commercial vehicles such as trucks and buses on highways and the vehicle lane-changing trajectories, the vehicles on highways US-101 and I-80 can be the main research objects. By comparing the data of the two highways, the length of the road sections included in the data on highway US-101 is greater than that of the data on highway I-80, which is more conducive to the analysis of the lane-changing intention and real-time trajectory prediction of large commercial vehicles when driving at high speeds. At the same time, there are more lanes in the research road sections of this trajectory dataset, and the number and density of the driving vehicles collected are very large. There are also more large commercial vehicles such as trucks and buses, which has a high degree of conformity with the current situation of vehicle driving on the highways in China, making the research of the present invention more in line with the current situation. In summary, the present invention is based on the trajectory data of the vehicles driving on highway US-101. That is to say, the US-101 highway dataset contained in the trajectory dataset is the preset vehicle motion trajectory information set for subsequent selection of model training data.

[0050] Refer to Figure 2 , which is a schematic diagram of the road surface conditions of the data collection road section provided by an embodiment of the present invention. As Figure 2 shown, highway US-101 is a ten-lane expressway in both directions. The innermost lane of highway US-101 is numbered as lane 1, and the numbers of other lanes increase from the inside out. Lane 1 is the high-speed driving lane, lane 6 is the lane-changing lane when the vehicle enters the highway, and lanes 7 and 8 are the approach roads when entering and leaving the highway.

[0051] Furthermore, a large amount of vehicle information is collected in the trajectory dataset, and each vehicle is numbered and distinguished. The continuous motion trajectory information of each vehicle includes 18 categories of information such as the longitude and latitude, horizontal and vertical coordinates, speed, acceleration, lane ID, the position and relative distance of the vehicle in front and behind, etc. The main parameter descriptions of the trajectory dataset are shown in Table 2.

[0052]

[0053] For the above-mentioned preset vehicle motion trajectory information set, a training dataset for training the initial prediction model can be selected from it. That is, for highway US-101 sections, different types of commercial vehicles participating in model training are selected as sample commercial vehicles. Thus, through selection, the historical target lane-changing trajectory information of different types of sample commercial vehicles can be used as the training dataset, and the historical target lane-changing trajectory information of each sample commercial vehicle is a training data in the training dataset.

[0054] It can be understood that, in order to improve the accuracy of model training, a threshold for the number of model training times can be preset, so that a trained lane-changing trajectory prediction model can be obtained when the number of model training times reaches the training times threshold.

[0055] In this way, after the initial prediction model is trained for a preset number of times, it can be judged whether the model loss of the intermediate prediction model after the preset number of times of training matches the preset model loss. If the model loss of the intermediate prediction model after the preset number of times of training matches the preset model loss, the model training is stopped, and the intermediate prediction model corresponding to the model loss when the model training stops is determined as the trained initial prediction model. Among them, the preset model loss can be a loss experience value, a loss threshold range, or an ideal loss level that can be achieved after being trained according to the preset number of times set by experience. There is no specific limitation here.

[0056] On the contrary, when the model effect after the preset number of times of training does not meet the preset requirements, the training can be carried out for another preset number of times. That is, if the model loss of the intermediate prediction model after the preset number of times of training does not match the preset model loss, the intermediate prediction model corresponding to the model loss can be trained using the next preset batch of training data in the training dataset until the lane-changing trajectory prediction model is determined.

[0057] It can be understood that for the situation where the initial prediction model is trained for a preset number of times, if it is determined that the model loss of the intermediate prediction model after the preset number of times of training does not match the preset model loss, it means that the intermediate prediction model obtained at this time does not meet the preset requirements. At this time, 200 batches of sample point cloud data can be selected from the remaining sample point cloud data of the training dataset to participate in the subsequent training. Among them, the next preset batch of training data selected again can be the same as or different from the batch volume of the previous preset batch of training data. There is no specific limitation here. And when selecting the next preset batch of training data from the remaining training data of the training dataset, it can be selected sequentially or at intervals. There is no specific limitation here either.

[0058] It should be noted that since the historical target lane-changing trajectory information of each sample operating vehicle contains multiple vehicles around the lane-changing sample operating vehicle and their respective parameter characteristics, the initial prediction model to be trained needs to pay attention to the changes in the driving characteristics of multiple vehicles in the historical target lane-changing trajectory information of each sample operating vehicle and their influence on the trajectory of the lane-changing sample operating vehicle. In view of this, a multi-head attention mechanism can be used in the Transformer encoder, and at the same time, a residual connection method is adopted to solve the problem that the original information is easily lost in multi-head attention, so that the model can better maintain information transmission while retaining the original input information.

[0059] The input and output layers of the multi-head attention mechanism are connected by residual connection through layer normalization. After the multi-head attention outputs data, it passes through a feed-forward network layer composed of two one-dimensional convolutional layers that can capture more local features of the data to regularize the data, which can improve the accuracy of non-linear data modeling.

[0060] For each training of the initial prediction model, first use the multi-head attention mechanism layer in the Transformer encoder to extract the first vehicle sample motion trajectory features of the corresponding sample operating vehicle from the historical target lane-changing trajectory information currently participating in training. Here, the multi-head attention mechanism layer maps the historical target lane-changing trajectory information to three data sets of query, key, and value, and obtains the proportion of different trajectory features during vehicle driving through continuous fitting and training methods. Further, the weight size is determined by analyzing the correlation degree of different trajectory points during vehicle operation. Among them, the three data sets of query, key, and value are obtained by multiplying the position encoding of the historical target lane-changing trajectory information by its weight matrix, and the required model parameters are trained by the gradient descent method.

[0061] For each of the first sample vehicle motion trajectory features extracted by the Transformer encoder, all the extracted first sample vehicle motion trajectory features can be output as the second sample vehicle motion trajectory features after passing through a feed-forward network layer containing two fully connected layers, and then all the output sample second vehicle motion trajectory features are input into a residual normalization layer composed of a residual connection and a layer normalization layer, and the third sample vehicle motion trajectory features suitable for use in the MLSTM network are output. Among them, the residual connection is set to prevent the model gradient from vanishing, and it adds all the original input first sample vehicle motion trajectory features and all the second sample vehicle motion trajectory features on the basis of outputting the third sample vehicle motion trajectory features. The existence of the layer normalization function can reduce data bias to prevent overfitting.

[0062] It can be understood that considering that the prediction of the target vehicle's lane-changing trajectory needs to comprehensively consider the influence of the surrounding vehicle information feature variables on the driving of the target vehicle, and consider the influence of the surrounding vehicles on the target vehicle within a certain time range. Therefore, in order to better consider the influence of these vehicles, the present invention selects the LSTM network structure for calculation, so as to provide more comprehensive vehicle motion trajectory feature information for the next layer.

[0063] The LSTM network is composed of different numbers of memory units, which can screen all the input third-party vehicle motion trajectory features. Each memory unit filters and outputs all the third-party vehicle motion trajectory features through the forget gate, input gate, and output gate. The MLSTM network is obtained by superimposing multiple LSTM network layers, which can not only continuously memorize different vehicle motion trajectory features for a long time, but also ensure that the final motion trajectory prediction results of the sample operating vehicles are more accurate. The LSTM network based on layered superposition can better predict the transformation trend of the surrounding vehicle trajectory data sequence information.

[0064] The characteristic parameter data of the horizontal and vertical coordinates, speed, acceleration and other characteristic parameters of the surrounding vehicles of the sample operating vehicle are converted into three time steps as input, and the horizontal and vertical coordinates of the sample operating vehicle are used as output targets. Two independent modules, the Transformer independent module and the LSTM network, are constructed, and the two independent modules are integrated to better output the characteristic results of the vehicle motion trajectory.

[0065] In summary, the core of the model establishment in the present invention is to use the multi-head attention mechanism layer in the Transformer module to calculate the weights of the vehicle motion trajectory features, and then continuously learn and adjust the vehicle motion trajectory features through the MLSTM module, which can improve the accuracy and stability of the vehicle's lane change trajectory prediction.

[0066] Based on the above Figure 1 The commercial vehicle safe driving assistance method shown, in an exemplary embodiment, determines the historical target lane change trajectory information of each of the different types of sample commercial vehicles from a preset vehicle motion trajectory information set, and the specific process is implemented through the following steps.

[0067] Firstly, useless attribute information that does not belong to the vehicle driving attributes in the preset vehicle motion trajectory information set is eliminated to obtain the first candidate motion trajectory information set; further, the lateral speed and lateral acceleration of each sample operating vehicle in the preset road section, the front vehicle driving speed and longitudinal acceleration of each sample operating vehicle, and the rear vehicle driving speed and longitudinal acceleration of each sample operating vehicle are added to the first candidate motion trajectory information set to obtain the second candidate motion trajectory information set; then, the motion trajectory information of large operating vehicles and small operating vehicles in the preset lanes are selected from the second candidate motion trajectory information set, and the lane change trajectory information of each historical target is determined based on the selection results.

[0068] It is understandable that, considering that the preset vehicle motion trajectory information set contains data information of more than 20 types of vehicles, a considerable part of it is invalid information such as road names and vehicle types. Since the present invention mainly studies the relevant trajectories of free lane changes of large commercial vehicles and buses, in order to make the data contain as much information related to model training as possible, it is necessary to classify the vehicle types in the preset vehicle motion trajectory information set, and to minimize the impact of forced lane changes, obstacles and other factors on the analysis of vehicle lane change behavior, so it is necessary to delete the data of vehicles traveling on special lanes, so as to screen out vehicle driving data suitable for model training.

[0069] Specifically, the preset vehicle motion trajectory information set can be re-edited first, that is, useless attribute information in the preset vehicle motion trajectory information set is eliminated to ensure that the first candidate motion trajectory information set only includes useful data related to vehicle lane change intention recognition and trajectory prediction, that is, mainly including vehicle number, lane number, position of the moving vehicle in the road section, namely the horizontal coordinate and vertical coordinate, driving speed, longitudinal acceleration, driving data (position, speed, etc.) of the front and rear vehicles and vehicles in other lanes, time, etc.

[0070] In addition, considering that the vehicle numbers in the preset vehicle motion trajectory information set are not continuous, they need to be re-sorted, and the lateral speed and lateral acceleration of the vehicle, as well as the speed and longitudinal acceleration of the vehicle in front and the speed and longitudinal acceleration of the vehicle behind are calculated through the vehicle's position and time information. The process of calculating the data column and inserting the results into the original data is carried out in Python, using the pandas and numpy libraries to load the trajectory data into the DataFrame, and re-sort and calculate row by row.

[0071] Exemplarily, the calculation formula of the lateral acceleration of the vehicle is shown in formula (1).

[0072] (1)

[0073] In formula (1), and Indicates the lateral coordinate of the vehicle at the current moment and the lateral coordinate of the next moment, is the data interval between two adjacent moments, = 0.1s, represents the lateral instantaneous acceleration of the moving vehicle; the vehicle in formula (1) can be any one of the moving vehicle, the vehicle involved in the moving vehicle, and the vehicle behind the moving vehicle. Similarly, the longitudinal acceleration of the vehicle in front of the moving vehicle and the longitudinal acceleration of the vehicle behind the moving vehicle can be calculated. Further, the lateral speed of the vehicle, the speed of the vehicle in front of the moving vehicle, and the speed of the vehicle behind the moving vehicle can be obtained through the conversion relationship between acceleration and speed.

[0074] Further, since the preset vehicle motion trajectory information set contains three types of vehicles: passenger cars and small and micro passenger cars, commercial vehicles and large buses, and motorcycles. Since the number of motorcycles traveling on highways is small and the vehicle types are special, for the convenience of research, motorcycles are removed from the preset vehicle motion trajectory information set. At the same time, the driving habits and motion states of different types of vehicles, such as large vehicles and small vehicles, vary greatly. The main vehicles studied in the present invention are large commercial vehicles such as commercial vehicles and large buses. However, for the accuracy of research, the present invention also retains the continuous motion trajectories of large vehicles and small vehicles.

[0075] In addition, considering that when a driver changes lanes while driving a vehicle, the driver mainly considers information such as driving speed and driving comfort. Therefore, there will be certain constraint information affecting the driving judgment of the vehicle. So, it is necessary to distinguish the parameters of vehicles driving on different roads in combination with the actual driving situation. For example, the average headway of lane 6 is significantly higher than that between other lanes. The main reason may be that lane 6 belongs to a ramp and the vehicle motion situation is more complex, requiring a longer distance to compensate for the reaction time of the driver during the lane change operation. It can be considered that the continuous motion trajectory of lane 6 has a greater impact on the subsequent research on the recognition and prediction of the lane change intention of subsequent vehicles. Therefore, it is necessary to remove the vehicles on lane 6 and the subsequent lanes 7 and 8 to prevent inaccurate variances and averages caused by excessive data gaps and eliminate the impact on vehicle lane change data; that is, the lanes before lane 6 (excluding lane 6) can be used as preset lanes. In this way, the motion trajectory information of large commercial vehicles and small commercial vehicles in the preset lanes can be selected from the second candidate motion trajectory information set, so as to determine each historical target lane change trajectory information based on the selection result.

[0076] Based on the above Figure 1 In an example embodiment of the commercial vehicle safe driving assistance method shown, the electronic device determines each historical target lane change trajectory information based on the selection result, and its specific process can be implemented through the following steps.

[0077] First, perform a determination process for vehicle ID reuse on the selection result; further, extract lane change trajectory information from the determination process result to obtain each historical candidate lane change trajectory information; then, perform a cleaning process on the field missing data of each historical candidate lane change trajectory information, and perform wavelet denoising on the cleaning process result to obtain each historical target lane change trajectory information.

[0078] Specifically, vehicle data is extracted by using the Vehicle_ID item in the preset vehicle motion trajectory information set. Considering that there is repeated use of vehicle IDs in the trajectory data set, it is necessary to determine whether the vehicle ID is reused through the Global_Time information in the preset vehicle motion trajectory information set, and process the vehicle ID reuse to solve the vehicle ID reuse in the preset vehicle motion trajectory information set.

[0079] At this time, the lane change trajectory information can be extracted from the judgment processing result by judging whether the lane has changed through the Lane_ID lane difference list to obtain the lane change trajectory information of each historical candidate.

[0080] In addition, by searching for the lane change time, the numbers of the front and rear vehicles of the research sample operating vehicle when it is driving in the lane and the numbers and driving data of the front and rear vehicles driving in the target lane are obtained, so as to obtain the historical candidate lane change trajectory information. In the historical candidate lane change trajectory information obtained, V_0 represents the historical candidate lane change trajectory information of a sample operating vehicle (referred to as the vehicle) when changing lanes, V_before_P represents the speed of the vehicle in front of the vehicle in the target lane, V_after_P represents the speed of the vehicle in front of the vehicle in the target lane, V_after_F represents the vehicle data after the vehicle in the target lane, distance_before_0_P represents the distance between the vehicle and the vehicle in front of the target lane when changing lanes, distance_after_0_P represents the distance between the vehicle in front of the target lane and the vehicle in front of the target lane when changing lanes, and distance_after_0_F represents the distance between the vehicle in the back of the target lane and the vehicle in the back of the target lane when changing lanes. At the same time, in order to ensure that the final data is available, the vehicle data with missing data is eliminated, that is, the field missing data of each historical candidate lane change trajectory information is cleaned.

[0081] Exemplarily, taking Table 2 as an example, a determination process for vehicle ID reuse of the selected results is provided, and the specific process can be implemented through the following steps.

[0082] First, whether the FrameID number in the preset vehicle motion trajectory information set is continuous is used as a judgment condition to determine whether there is vehicle ID reuse in the selection result; then, if it is determined that there is vehicle ID reuse in the selection result, the value of Global_Time in the row corresponding to the continuous FrameID number in the judgment result is changed to the value of the corresponding row TotalFrame.

[0083] Specifically, the reuse of vehicle IDs is determined based on the Global_Time information. Since Global_Time is in scientific notation and the value of Global_Time in some of the original data is the same for each line, at this time, whether the FrameID numbers in the preset vehicle movement trajectory information set are consecutive can be used as a discrimination condition. This is to facilitate the determination of the existence of vehicle ID reuse when the FrameID numbers are consecutive, and the values of Global_Time in the lines corresponding to the consecutive FrameID numbers are changed to the values of TotalFrame in the corresponding lines, thus solving the problem of vehicle ID reuse in the preset vehicle movement trajectory information set.

[0084] In summary, after removing invalid data, calculating valid data, extracting relevant vehicle models, and extracting lane-changing trajectory information from the preset vehicle movement trajectory information set in the trajectory dataset, the trajectory information during the vehicle lane-changing process can be initially obtained. However, there are still a large number of errors in the preset vehicle movement trajectory information set. These errors will not only cause misjudgments of the actual motion state by the model but also increase the loss during model training, resulting in problems such as gradient explosion. Therefore, it is necessary to perform filtering processing on the extracted vehicle movement data, that is, perform wavelet denoising on the cleaning processing results to obtain the lane-changing trajectory information of each historical target, so that the lane-changing trajectory information of the historical target in the training dataset is smoother and convenient for model training.

[0085] Based on the above Figure 1 In one exemplary embodiment based on the above-described commercial vehicle safe driving assistance method, the electronic device extracts lane-changing trajectory information from the determination processing result to obtain each historical candidate lane-changing trajectory information, and its specific process can be implemented through the following steps.

[0086] Extract the lane-changing trajectory information of the same sample commercial vehicle that changes lanes multiple times and the lane-changing interval time meets the preset minimum interval frame number condition from the determination processing result to obtain each historical candidate lane-changing trajectory information.

[0087] Specifically, through the Lane_ID lane difference list, it is judged whether the lane changes. In addition, if the same sample commercial vehicle changes lanes multiple times, then multiple lane changes are regarded as single lane changes for extraction. In this case, if the lane-changing interval time is too short, it obviously has no research significance. Therefore, by setting the preset minimum interval frame number condition, the minimum interval frame number of consecutive lane changes is restricted. If the lane-changing interval time does not meet the minimum interval frame number, the corresponding lane change is removed; thus, the purpose of extracting the lane-changing trajectory information of the same sample commercial vehicle that changes lanes multiple times and the lane-changing interval time meets the preset minimum interval frame number condition is achieved.

[0088] Based on the above Figure 1The disclosed operation vehicle safe driving assistance method, in one exemplary embodiment, performs wavelet noise reduction on the cleaning processing result. The specific process may include: performing wavelet noise reduction on the cleaning result using a first preset wavelet, or performing wavelet noise reduction on the cleaning result using a second preset wavelet.

[0089] Among them, the first preset wavelet is any wavelet between the Sym4 wavelet and the Sym11 wavelet, and the second preset wavelet is the Db4 wavelet or the Db5 wavelet.

[0090] It should be noted that when performing wavelet noise reduction, the selection of the wavelet basis function needs to be made according to actual requirements. Through the analysis and research of the existing commonly used wavelet basis functions such as the Haar function, Daubechies (Db) function, Symlets (Syms) function, and Morlet function, when the present invention uses the Symlets series basis function for noise reduction, the filter length between Sym4 and Sym11 has a good noise reduction effect, that is, Syms4 to Syms11; or when using the Daubechies series function for noise reduction, the filter length between Syms4 and Syms5, that is, Db4 and Db5 have a good effect. Since the speed can be obtained by differential calculation using the position coordinates, the original lateral trajectory in the dataset is respectively subjected to Db4 to Db5 and Syms4 to Syms11 wavelet transforms.

[0091] According to the signal-to-noise ratio and root mean square error results, it can be found that for the lateral trajectory, among the 10 groups of wavelet basis functions in the experiment, after using Syms11 as the basis function for noise reduction processing, there is the largest signal-to-noise ratio between the noise-reduced data and the original data, and at the same time, there is the smallest root mean square error. Therefore, for the lateral trajectory, the present invention preferably uses Syms11 as the basis function for noise reduction processing.

[0092] Exemplarily, for the lane change trajectory prediction model obtained after training, its prediction effect on the vehicle lane change trajectory can be compared with the model established by a single model network to verify whether the joint model plays a role in trajectory prediction, and the accuracy of the joint network and the single network is respectively compared. Since the optimal hyperparameter combinations in the training of different network models may be different, the prediction accuracies under the same hyperparameter combination are not comparable, and different hyperparameter combinations will also affect the training speed. Therefore, under the same hyperparameter combination of each model, the trajectories in the dataset are randomly selected for prediction, and the prediction results of the trajectories with normalized lateral coordinates in the test set are visually compared.

[0093] In the lane change trajectory prediction model of the vehicle, the predicted results of the target vehicle lane change trajectory at the next 50 moments are used to compare with the actual lane change trajectory of the vehicle. Figure 3 and Figure 4They are the trajectory prediction results of a vehicle randomly selected for a left lane change and a right lane change respectively, and are compared with the actual vehicle operation trajectory results. From the perspective of the prediction results of each model, the vehicle lane change trajectory prediction result of the lane change trajectory prediction model obtained by training the initial prediction model containing a Transformer encoder - MLSTM network is the most consistent with the lane change trajectory during the actual vehicle process, indicating that the Transformer encoder - MLSTM network vehicle lane change trajectory prediction model is superior to the single model in terms of prediction accuracy. The trajectory prediction accuracy of the single model is significantly worse, mainly because different vehicles are not distinguished, so the error of the vehicle trajectory prediction result is relatively large. At the same time, it can be found that the trajectory prediction result of the vehicle with a left lane change is more consistent with the actual lane change trajectory of the vehicle than that with a right lane change, and when the actual lane change trajectory data of the vehicle is smooth, the trajectory prediction result is also more accurate.

[0094] In addition, in the training of the established neural network model, hyperparameters such as the learning rate, the number of model training epochs, the number of data for one - time training, hidden neurons, multi - head attention mechanism, the number of LSTM stacked layers, etc. are neural network parameters set before model learning and training. They are not the internal calculation parameters of the neural network but the framework parameters for neural network training. Selecting different hyperparameters will affect the learning of the neural network model and thus the prediction results. The optimization of hyperparameters is to select a set of hyperparameters that can make the training results optimal for the neural network model, which is the most important part in the training of the entire neural network model.

[0095] Among them, the selection of the number of data for one - time training plays an important role in network training. A suitable number of data for one - time training can improve the utilization rate of the computer GPU and memory during calculation, and at the same time accurately determine the gradient descent direction, which is directly related to the overall effect of model training. Stacking LSTM network layers can not only continuously remember long - term information but also better capture the changing trend of the surrounding vehicle trajectory data sequence information. However, it is not always the case that the more stacked layers, the better the actual effect. Since there is currently no method to directly determine the optimal hyperparameters, it is mainly based on experience to compare the model training results under different hyperparameter combinations to obtain a set of optimal hyperparameters. In view of the characteristics of the Transformer encoder - MLSTM network, different experimental combinations of the number of one - time training samples, the number of LSTM stacked layers, and the number of training epochs are selected to compare their impacts on the neural network.

[0096] After training the Transformer encoder-MLSTM network with 24 sets of proposed hyperparameter combinations, the 24 models converged within a certain number of training rounds from the training set to the test set and can be used for predicting the vehicle lane change trajectory. The present invention uses RMSE (a deviation evaluation index of the true value and the predicted value) to evaluate the vehicle lane change trajectory results predicted by the Transformer encoder-MLSTM network under 24 different hyperparameter combinations, so as to obtain the optimal hyperparameter combination.

[0097] Figure 5 and Figure 6 respectively compare the RMSE values between the vehicle lane change trajectories predicted by the Transformer-MLSTM network models under different hyperparameter combinations and the true vehicle lane change trajectories.

[0098] From Figure 5 and Figure 6 in the comparison, the trends of the RMSE values calculated when the Transformer encoder-MLSTM network predicts the lateral and longitudinal trajectories of the vehicle are different, but the RMSE value calculated when predicting the longitudinal trajectory of the vehicle is significantly larger. The main reason is that the trajectory change of the vehicle in the longitudinal direction is much greater than that in the lateral direction of the vehicle, so the error will be higher. Generally speaking, by comparing the RMSE values of the prediction results, the number of stacked layers of LSTM in the established Transformer encoder-MLSTM network is 2, and the single training data volume is 1500. Under this hyperparameter combination, the RMSE values of the model predicting the lateral trajectory change and the longitudinal trajectory change of the vehicle are 0.364 and 1.492 respectively, and better results can be obtained.

[0099] Furthermore, when predicting the vehicle lane change trajectory, users hope to identify the future vehicle lane change trajectory earlier to leave more reaction time to avoid risks. For the vehicle lane change trajectory prediction model, it is to use as little data as possible to predict a longer vehicle trajectory, that is, to use the shortest time window as much as possible while ensuring the prediction accuracy.

[0100] The specific sampling method of the trajectory is determined according to the lane change trajectory. Referring to relevant research literature, the time interval for an operating vehicle to complete a lane change in a highway scenario is 3-7 s (generally longer than that of a passenger car), that is, the average time taken for an operating vehicle to complete a complete lane change behavior is 5 s. The trajectory of the first 3 s connected to the predicted target trajectory is the historical trajectory of the vehicle. That is, all the above studies are based on the time length of each complete sample trajectory being 8 s, so as to obtain that the length of the model input data sequence is 30, and the number of future vehicle lane change trajectory points to be predicted is 50.

[0101] To further study the relationship between different time window lengths and the results of predicted vehicle lane change trajectories, time windows of 2s, 3s, and 4s were respectively selected to predict the vehicle trajectories when the vehicle changed lanes to the left. As Figure 7 shown, it is a comparison chart of the deviation distances between the predicted trajectories and the true trajectories of the vehicle calculated by the model under different time window settings.

[0102] As Figure 7 shown, when predicting for a certain duration, the longer the time window length is set, the smaller the deviation distance of the obtained trajectory prediction. According to the above results, in the setting of network model parameters, when other parameter settings are the same, a longer time window length should be set as much as possible. However, a longer time window length requires more historical lane change trajectory data. But when the vehicle just starts to change lanes, the historical trajectory data of the vehicle is basically the straight trajectory feature, which will have a certain impact on the lane change trajectory of the model. Therefore, it can be considered that the model uses a short time window at the beginning of lane change, which requires less data but can make predictions faster to remind of lane change, and increase the time window length as the lane change progresses to improve the accuracy of trajectory prediction.

[0103] In summary, the present invention uses a Transformer encoder-MLSTM network to establish a lane change trajectory prediction model. By using the multi-head attention mechanism and high-efficiency parallel processing ability of the Transformer encoder, weight calculation is performed on the motion trajectory features of surrounding vehicles. At the same time, the MLSTM network can capture long-term dependencies, and strengthen the parallel information transmission between stacked layers to enhance the overall expression ability of the model. The combined model can make the prediction of vehicle lane change trajectories more accurate and stable. A sliding time window is used to predict the future driving trajectory based on the vehicle's historical trajectory information, and the left and right lane change trajectories are trained separately, verifying that the model is superior to single network models established by methods such as the LSTM network, MLSTM network, and Transformer encoder in terms of prediction accuracy. By comparing the RMSE between the trajectories predicted by the model under different hyperparameter combinations and the true trajectories, the optimal hyperparameter combination for each is determined. Finally, by comparing and analyzing the trajectory prediction situations under different time window lengths, it is concluded that using a long time window contains more lane change trajectories, and the prediction accuracy is better than that of a short time window. However, the historical lane change trajectories required for prediction using a short time window are shorter, and the prediction starting point is earlier than that of a long time window. The present invention greatly helps drivers make early decisions and take measures, and can avoid possible collision risks in advance.

[0104] The operating vehicle safety driving assistance device provided by the present invention will be described below. The operating vehicle safety driving assistance device described below can be mutually referred to the operating vehicle safety driving assistance method described above.

[0105] Referring to Figure 8, is a schematic diagram of the structure of a safe driving assistance device for commercial vehicles provided by an embodiment of the present invention, such as Figure 8 As shown, the commercial vehicle safe driving assistance device 800 includes: a motion trajectory acquisition module 810, a lane change trajectory prediction module 820 and a safe driving assistance module 830.

[0106] The motion trajectory acquisition module 810 is used to obtain the current motion trajectory information of the target operating vehicle during the driving process.

[0107] The lane changing trajectory prediction module 820 is used to input the current motion trajectory information into the lane changing trajectory prediction model, extract the global motion trajectory features of the current motion trajectory information through the lane changing trajectory prediction model, obtain the current motion trajectory features of the target operating vehicle, and then perform optimal lane changing trajectory prediction based on the current motion trajectory features and historical motion trajectory features to obtain the lane changing trajectory prediction result of the target operating vehicle output by the lane changing trajectory prediction model.

[0108] The safe driving assistance module 830 is used to feed back the lane change trajectory prediction results to other vehicles around the target operating vehicle, so that other vehicles can issue a safe driving warning when they recognize the lane change intention of the target operating vehicle from the lane change trajectory prediction results.

[0109] Optionally, the lane change trajectory prediction module 820 is specifically used to determine the historical target lane change trajectory information of different types of sample operating vehicles from a preset vehicle motion trajectory information set; use each historical target lane change trajectory information to train the initial prediction model containing the Transformer encoder-MLSTM network, and determine the lane change trajectory prediction model based on the training results.

[0110] Optionally, the lane change trajectory prediction module 820 is specifically used to eliminate useless attribute information that does not belong to the vehicle driving attributes in the preset vehicle motion trajectory information set to obtain a first candidate motion trajectory information set; the lateral speed and lateral acceleration of each sample operating vehicle in the preset road section, the front vehicle driving speed and longitudinal acceleration of each sample operating vehicle, and the rear vehicle driving speed and longitudinal acceleration of each sample operating vehicle are added to the first candidate motion trajectory information set to obtain a second candidate motion trajectory information set; the motion trajectory information of large operating vehicles and small operating vehicles in the preset lanes are selected from the second candidate motion trajectory information set, and the lane change trajectory information of each historical target is determined based on the selection results.

[0111] Optionally, the lane change trajectory prediction module 820 is specifically configured to perform a determination process for reusing vehicle IDs on the selection result; extract lane change trajectory information from the determination process result to obtain each historical candidate lane change trajectory information; perform a cleaning process on the missing data fields of each historical candidate lane change trajectory information, and perform wavelet denoising on the cleaning process result to obtain each historical target lane change trajectory information.

[0112] Optionally, the lane change trajectory prediction module 820 is specifically configured to perform wavelet denoising on the cleaning result using a first preset wavelet, or perform wavelet denoising on the cleaning result using a second preset wavelet; wherein, the first preset wavelet is any wavelet between Sym4 wavelet and Sym11 wavelet, and the second preset wavelet is Db4 wavelet or Db5 wavelet.

[0113] Optionally, when the lane change trajectory prediction model includes a trained multi-head attention mechanism layer, a trained feed-forward network layer, and a trained residual normalization layer, the lane change trajectory prediction module 820 is specifically configured to use the trained multi-head attention mechanism layer to extract motion trajectory features of multiple different angular orientations from the current motion trajectory to obtain the first vehicle motion trajectory features; use the trained feed-forward network layer to perform a fully connected process on the first vehicle motion trajectory features to obtain the second vehicle motion trajectory features; use the trained residual normalization layer to perform feature standardization on the first vehicle motion trajectory features and the second vehicle motion trajectory features to obtain the current motion trajectory features.

[0114] Optionally, when the lane change trajectory prediction model includes a trained MLSTM network, the lane change trajectory prediction module 820 is specifically configured to use the trained MLSTM network to extract motion trajectory sequence features from the current running trajectory features and historical motion trajectory features, and then perform lane change trajectory prediction on the extracted motion trajectory sequence features to obtain a lane change trajectory prediction result.

[0115] Figure 9 An example of the physical structure diagram of an electronic device is shown as Figure 9As shown in the figure, the electronic device may include: a processor 910, a communications interface 920, a memory 930, and a communication bus 940. Among them, the processor 910, the communications interface 920, and the memory 930 complete communication with each other through the communication bus 940. The processor 910 may call the logical instructions in the memory 930 to execute the safety driving assistance method for operating vehicles, and the method includes: obtaining the current motion trajectory information of the target operating vehicle during driving; inputting the current motion trajectory information into the lane change trajectory prediction model, and extracting the global motion trajectory features of the current motion trajectory information through the lane change trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle, and then predicting the optimal lane change trajectory based on the current motion trajectory features and the historical motion trajectory features to obtain the lane change trajectory prediction result of the target operating vehicle output by the lane change trajectory prediction model; feeding back the lane change trajectory prediction result to other vehicles around the target operating vehicle, so that when other vehicles recognize the lane change intention of the target operating vehicle from the lane change trajectory prediction result, they can perform safety driving warnings.

[0116] In addition, when the logical instructions in the above-mentioned memory 930 can be implemented in the form of software function units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0117] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the operation vehicle safe driving assistance method provided by each of the above methods. The method includes: obtaining the current motion trajectory information of a target operation vehicle during driving; inputting the current motion trajectory information into a lane change trajectory prediction model, and extracting the global motion trajectory features of the current motion trajectory information through the lane change trajectory prediction model to obtain the current motion trajectory features of the target operation vehicle, and then predicting the optimal lane change trajectory based on the current motion trajectory features and historical motion trajectory features to obtain the lane change trajectory prediction result of the target operation vehicle output by the lane change trajectory prediction model; and feeding back the lane change trajectory prediction result to other vehicles around the target operation vehicle, so that when other vehicles recognize the lane change intention of the target operation vehicle from the lane change trajectory prediction result, a safe driving warning is issued.

[0118] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the operation vehicle safe driving assistance method provided by each of the above methods. The above is only used to illustrate the technical solution of the present invention and is not intended to limit it.

Claims

1. A safety driving assistance method for operating vehicles, characterized in that, Including: Obtaining the current motion trajectory information of a target operating vehicle during driving; Inputting the current motion trajectory information into a lane-changing trajectory prediction model, extracting the global motion trajectory features of the current motion trajectory information through the lane-changing trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle, and then predicting the optimal lane-changing trajectory based on the current motion trajectory features and historical motion trajectory features to obtain the lane-changing trajectory prediction result of the target operating vehicle output by the lane-changing trajectory prediction model; Feeding back the lane-changing trajectory prediction result to other vehicles around the target operating vehicle, so that when other vehicles recognize the lane-changing intention of the target operating vehicle from the lane-changing trajectory prediction result, a safety driving warning is issued.

2. The safety driving assistance method for operating vehicles according to claim 1, wherein The training process of the lane-changing trajectory prediction model includes: Determining the respective historical target lane-changing trajectory information of different types of sample operating vehicles from a preset vehicle motion trajectory information set; Training an initial prediction model containing a Transformer encoder-MLSTM network using each of the historical target lane-changing trajectory information, and determining the lane-changing trajectory prediction model according to the training result.

3. The method for assisting safe driving of an operating vehicle according to claim 2, wherein The determining the respective historical target lane-changing trajectory information of different types of sample operating vehicles from a preset vehicle motion trajectory information set includes: Removing the useless attribute information that does not belong to the vehicle driving attributes in the preset vehicle motion trajectory information set to obtain a first candidate motion trajectory information set; Adding the lateral speed and lateral acceleration of each sample operating vehicle in a preset section, the driving speed and longitudinal acceleration of the vehicle in front of each sample operating vehicle, and the driving speed and longitudinal acceleration of the vehicle behind each sample operating vehicle to the first candidate motion trajectory information set to obtain a second candidate motion trajectory information set; Selecting the motion trajectory information of large operating vehicles and small operating vehicles in a preset lane respectively from the second candidate motion trajectory information set, and determining each of the historical target lane-changing trajectory information based on the selection result.

4. The method for assisting safe driving of an operating vehicle according to claim 3, wherein, The determining each of the historical target lane-changing trajectory information based on the selection result includes: Performing a determination process of vehicle ID reuse on the selection result; Extracting lane-changing trajectory information from the determination process result to obtain each historical candidate lane-changing trajectory information; Performing a cleaning process on the missing data in the fields of each historical candidate lane-changing trajectory information, and performing wavelet denoising on the cleaning process result to obtain each of the historical target lane-changing trajectory information.

5. The method for assisting safe driving of an operating vehicle according to claim 4, wherein, The performing wavelet denoising on the cleaning process result includes: Performing wavelet denoising on the cleaning result using a first preset wavelet, or performing wavelet denoising on the cleaning result using a second preset wavelet; Wherein, the first preset wavelet is any wavelet between Sym4 wavelet and Sym11 wavelet, and the second preset wavelet is Db4 wavelet or Db5 wavelet.

6. The safety driving assistance method for operating vehicles according to any one of claims 1 to 5, characterized in that, The extracting the global motion trajectory features of the current motion trajectory information through the lane-changing trajectory prediction model to obtain the current motion trajectory features of the target operating vehicle includes: When the lane-changing trajectory prediction model includes a trained multi-head attention mechanism layer, a trained feed-forward network layer, and a trained residual normalization layer, using the trained multi-head attention mechanism layer to extract motion trajectory features of the current motion trajectory from multiple different angular orientations to obtain first vehicle motion trajectory features; using the trained feed-forward network layer to perform a fully-connected process on the first vehicle motion trajectory features to obtain second vehicle motion trajectory features; using the trained residual normalization layer to perform feature normalization on the first vehicle motion trajectory features and the second vehicle motion trajectory features to obtain the current motion trajectory features.

7. The method for assisting safe driving of an operating vehicle according to any one of claims 1 to 5, characterized in that The optimal lane-changing trajectory prediction based on the current motion trajectory features and historical motion trajectory features to obtain the lane-changing trajectory prediction result of the target commercial vehicle output by the lane-changing trajectory prediction model includes: When the lane-changing trajectory prediction model includes a trained MLSTM network, using the trained MLSTM network to extract motion trajectory sequence features from the current motion trajectory features and the historical motion trajectory features, and then performing lane-changing trajectory prediction on the extracted motion trajectory sequence features to obtain the lane-changing trajectory prediction result.

8. A safety driving assistance device for operating vehicles, characterized in that Including: A motion trajectory acquisition module for acquiring the current motion trajectory information of the target commercial vehicle during driving; A lane-changing trajectory prediction module for inputting the current motion trajectory information into the lane-changing trajectory prediction model, extracting the global motion trajectory features of the current motion trajectory information through the lane-changing trajectory prediction model to obtain the current motion trajectory features of the target commercial vehicle, and then performing optimal lane-changing trajectory prediction based on the current motion trajectory features and historical motion trajectory features to obtain the lane-changing trajectory prediction result of the target commercial vehicle output by the lane-changing trajectory prediction model; A safe driving assistance module for feeding back the lane-changing trajectory prediction result to other vehicles around the target commercial vehicle, so that when the other vehicles recognize the lane-changing intention of the target commercial vehicle from the lane-changing trajectory prediction result, a safe driving warning is issued.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the commercial vehicle safe driving assistance method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the commercial vehicle safe driving assistance method according to any one of claims 1 to 7.

Images