Lane environment information fitting method and system and storage medium

Abstract

The invention provides a lane environment information fitting method and system and a storage medium. The method comprises the following steps: acquiring basic environment data of a surrounding environment of a target vehicle through a vehicle-mounted millimeter wave radar of the target vehicle; preprocessing the basic environment data to obtain target environment data in standard normal distribution; constructing a BP neural network module based on the neural network model; inputting the target environment data into the BP neural network module, and outputting an environment data point set through the BP neural network module; and fitting lane environment information around the target vehicle according to the environment data point set. The method has the advantages that the lane boundary line can be accurately fitted, and the cost is low.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and specifically relates to a lane environment information fitting method, system and storage medium. Background Art

[0002] In recent years, with the rapid development of artificial intelligence, autonomous driving has become a hot topic in the industry. For autonomous vehicles, when they sense obstacles through sensors, they need to make lane change decisions, and lane change decisions require knowing a lot of specific information, such as which lane the vehicle is currently in, whether there are lanes next to the current lane, whether the current lane allows lane changes, etc. This requires the use of perception technology in autonomous vehicles to perceive environmental information around the lane.

[0003] In terms of perception of autonomous driving technology, the cameras and lidars used in existing technologies can be said to be the core sensors for realizing autonomous driving environmental perception. They each have their own advantages and disadvantages, and these advantages and disadvantages are not limited to the environmental perception around the vehicle body that we are more familiar with. However, cameras and lidars are easily affected by light and weather, resulting in inaccurate information collection, which can easily lead to wrong decisions during autonomous driving. The above problems can be solved by high-precision map technology. High-precision map (HD map) is a map used for autonomous driving, which contains map elements such as road shape, road markings, traffic signs and obstacles. The map accuracy can be down to the centimeter level. Through high-precision maps, the lane environment around the vehicle can be accurately output, thereby assisting the autonomous driving system to make correct decisions. However, the application cost of high-precision maps is very expensive, and if used in batches, it will incur huge costs. Summary of the invention

[0004] The present invention provides a lane environment information fitting method, system and storage medium to solve the problem of high cost of accurately acquiring lane environment information.

[0005] In a first aspect, the present invention provides a lane environment information fitting method, the method comprising the following steps:

[0006] Acquiring basic environmental data of the surrounding environment of the target vehicle through the on-board millimeter-wave radar of the target vehicle;

[0007] Preprocessing the basic environmental data to obtain target environmental data with standard normal distribution;

[0008] Construct BP neural network module based on neural network model;

[0009] Inputting the target environment data into the BP neural network module, and outputting an environment data point set through the BP neural network module;

[0010] Lane environment information around the target vehicle is fitted according to the environmental data point set.

[0011] Optionally, the preprocessing of the basic environment data to obtain target environment data with a standard normal distribution comprises the following steps:

[0012] Parsing the basic environment data according to the communication protocol corresponding to the vehicle-mounted millimeter-wave radar to obtain target structured data;

[0013] Based on the spatial coordinate system where the target vehicle is located, the target structured data is coordinate-converted to obtain the relative position relationship between the physical target in the surrounding environment and the target vehicle;

[0014] Extracting environmental feature data with the entity target as the main body from the target structured data based on the relative position relationship;

[0015] The environmental characteristic data are normalized to obtain target environmental data with standard normal distribution.

[0016] Optionally, the normalization formula used in the normalization process is as follows:

[0017]

[0018] In the formula: x′ represents the target environment data after normalization, x represents the environment characteristic data before normalization, μ represents the data mean of the environment characteristic data, and δ represents the standard deviation of the environment characteristic data.

[0019] Optionally, the BP neural network module includes a three-layer BP network architecture of an input layer, a hidden layer and an output layer, a first weight matrix is ​​provided between the input layer and the hidden layer, and a second weight matrix is ​​provided between the hidden layer and the output layer.

[0020] Optionally, the step of inputting the target environment data into the BP neural network module and outputting the environment data point set through the BP neural network module comprises the following steps:

[0021] Initializing the first weight matrix and the second weight matrix;

[0022] Inputting the target environment data into the hidden layer in the BP neural network module through the input layer, and outputting first output data through the hidden layer;

[0023] Inputting the first output data as input data into the output layer, and outputting second output data through the output layer;

[0024] Calculating an output error based on the second output data, and back-propagating the output error to adjust the weights of the first weight matrix and the second weight matrix;

[0025] Determining whether the output error is less than a preset error threshold;

[0026] If the output error is greater than or equal to the error threshold, then repeat the above data processing steps;

[0027] If the output error is less than the error threshold, the second output data is output as an environmental data point set.

[0028] Optionally, the expression formula of the first output data is as follows:

[0029]

[0030] Where: j = 1, 2, ..., m, j represents the jth neuron in the hidden layer, y j represents the first output data output by the jth neuron in the hidden layer, V=(v 1 , v 2 , ..., v n ) T represents the first weight matrix, represents the weight vector corresponding to the jth neuron in the hidden layer, X=(x 1 , x 2 ,...x n ) T represents the target environment data, and f(·) represents the activation function;

[0031] The expression formula of the second output data is as follows:

[0032]

[0033] Where: k = 1, 2, ..., m, k represents the kth neuron in the output layer, o k represents the second output data output by the kth neuron in the output layer, W=(w 1 , w 2 , ..., w n ) T represents the second weight matrix, Represents the weight vector corresponding to the kth neuron in the output layer.

[0034] Optionally, the calculating the output error based on the second output data, and back-propagating the output error to adjust the weights of the first weight matrix and the second weight matrix comprises the following steps:

[0035] calculating an output error based on the second output data;

[0036] Back-propagating the output error to the output layer and the hidden layer, and respectively calculating a first error signal of the output layer and a second error signal of the hidden layer;

[0037] The weights of the second weight matrix are adjusted according to the first error signal, and the weights of the first weight matrix are adjusted according to the second error signal.

[0038] Optionally, the calculation formula of the output error is as follows:

[0039]

[0040] Where: E p represents the output error of the pth target environment data, l represents the lth layer of neurons in the output layer, represents the expected output value of the p-th target environment data corresponding to the k-th neuron of the output layer, represents the second output data corresponding to the p-th target environment data;

[0041] The calculation formula of the first error signal is as follows:

[0042]

[0043] Where: k = 1, 2, ..., l, represents the first error signal, d k Indicates the expected output value;

[0044] The calculation formula of the second error signal is as follows:

[0045]

[0046] Where: j = 1, 2, ..., m, represents the second error signal, w jk A weight vector representing the kth neuron in the output layer and the jth neuron in the hidden layer;

[0047] The formula for adjusting the weights of the second weight matrix according to the first error signal is as follows:

[0048]

[0049] The formula for adjusting the weights of the first weight matrix according to the second error signal is as follows:

[0050]

[0051] Where: η represents the learning rate.

[0052] In a second aspect, the present invention further provides a lane environment information fitting system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the lane environment information fitting method as described in the first aspect when executing the computer program.

[0053] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the lane environment information fitting method described in the first aspect are implemented.

[0054] The beneficial effects of the present invention are:

[0055] The present invention realizes lane line fitting through processing and treatment based on the static target position information output by the millimeter-wave radar sensor, and does not fit lane lines based on the dynamic target information output after the radar is fused with other sensors. According to the installation position of the millimeter-wave radar and the coordinate system, a coordinate conversion is performed based on the coordinate system of the autonomous driving vehicle, so as to obtain the relative position of the autonomous driving vehicle and the static targets on both sides of the lane, and the position of the vehicle is obtained according to the interaction of the current lane line detected by the on-board visual camera; in addition, according to the positional relationship between the static targets on both sides of the lane, a linear curve is fitted to obtain the lane type and direction. The present invention makes up for the problem that lane environment information cannot be output without a high-precision map, and the cost of millimeter-wave radar is relatively low. While replacing the high-precision map to output lane environment information, a lot of costs can also be saved. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of the lane environment information fitting method in the present invention.

[0057] Figure 2 It is a system structure diagram of the lane environment information fitting system in the present invention. DETAILED DESCRIPTION

[0058] The invention discloses a lane environment information fitting method.

[0059] Reference Figure 1 , the lane environment information fitting method specifically includes the following steps:

[0060] S101. Obtain basic environmental data of the target vehicle's surroundings through the target vehicle's on-board millimeter-wave radar.

[0061] The vehicle-mounted millimeter-wave radar emits millimeter-wave signals and receives echo signals reflected by other vehicles, obstacles, road boundaries and other objects in the target vehicle's surrounding environment. By analyzing the time delay, intensity and frequency of the echo signal, basic environmental data such as the position, speed and shape of objects around the target vehicle can be obtained.

[0062] S102. Preprocess basic environmental data to obtain target environmental data with standard normal distribution.

[0063] The basic environment data is preprocessed to make it meet the requirements of standard normal distribution. Some statistical methods and algorithms, such as mean normalization and variance normalization, can be used to standardize the data.

[0064] S103. Construct a BP neural network module based on the neural network model.

[0065] Among them, the BP neural network is a multi-layer feedforward neural network that performs error correction through the error back propagation algorithm. Its most core feature is that the signal is forward propagated, while the error is backward propagated. During the forward propagation process, the input signal is processed layer by layer through the input layer and the hidden layer. When it reaches the output layer, if the result does not meet the expected requirements, it enters the back propagation process, returns the error signal along the original path, and modifies the weights of each layer. The BP neural network includes an input layer, a hidden layer, and an output layer, among which there can be multiple hidden layers. The number of nodes in the input layer and the output layer is fixed (the number of variables of the input sample and the number of output labels, respectively), but the number of nodes in the hidden layer is not fixed.

[0066] Taking a single hidden layer BP neural network as an example, assuming that the input layer has n neurons, the hidden layer has q neurons, the output layer has m neurons, the weight matrix between the input layer and the hidden layer is V, the weight matrix between the hidden layer and the output layer is W, and the input variable is X, then in the forward propagation process, the output of the hidden layer is:

[0067]

[0068] The output of the output layer is:

[0069]

[0070] S104. Input the target environment data into the BP neural network module, and output the environment data point set through the BP neural network module.

[0071] The pre-processed target environment data is input into the constructed BP neural network module, and trained by the back propagation algorithm to obtain an environmental data point set as output. These environmental data point sets can represent the lane boundary lines around the target vehicle.

[0072] S105. Fit the lane environment information around the target vehicle according to the environmental data point set.

[0073] Among them, according to the environmental data point set, the lane boundary line of the target vehicle is fitted and estimated by using a fitting algorithm such as the least square method or spline interpolation, or a curve fitting method. Thus, the lane boundary line of the target vehicle is obtained. This can help the vehicle perform functions such as lane keeping and automatic driving.

[0074] The implementation principle of this embodiment is:

[0075] The present invention realizes lane line fitting through processing and treatment based on the static target position information output by the millimeter-wave radar sensor, and does not fit lane lines based on the dynamic target information output after the radar is fused with other sensors. According to the installation position of the millimeter-wave radar and the coordinate system, a coordinate conversion is performed based on the coordinate system of the autonomous driving vehicle, so as to obtain the relative position of the autonomous driving vehicle and the static targets on both sides of the lane, and the position of the vehicle is obtained according to the interaction of the current lane line detected by the on-board visual camera; in addition, according to the positional relationship between the static targets on both sides of the lane, a linear curve is fitted to obtain the lane type and direction. The present invention makes up for the problem that lane environment information cannot be output without a high-precision map, and the cost of millimeter-wave radar is relatively low. While replacing the high-precision map to output lane environment information, a lot of costs can also be saved.

[0076] In one implementation manner, step S102 specifically includes the following steps:

[0077] Analyze basic environmental data according to the communication protocol corresponding to the vehicle-mounted millimeter-wave radar to obtain target structured data;

[0078] Based on the spatial coordinate system where the target vehicle is located, the target structured data is transformed to obtain the relative position relationship between the physical target and the target vehicle in the surrounding environment;

[0079] Extracting environmental feature data with entity targets as the main body from target structured data based on relative position relationship;

[0080] The environmental characteristic data are normalized to obtain the target environmental data with standard normal distribution.

[0081] In this embodiment, according to the CAN communication protocol of the vehicle-mounted millimeter wave radar, the received raw data is parsed and processed to extract the target structured data. These data may include information such as the position, speed, and size of the physical targets around the vehicle. It is assumed that the spatial coordinate system where the target vehicle is located is a rectangular coordinate system with the vehicle as the origin. By relatively calculating the object position information in the target structured data and the position information of the target vehicle, the position relationship of each target object relative to the target vehicle, such as distance and azimuth, can be obtained. The target structured data around the target vehicle may include information such as the position, speed, and size of the physical targets around the vehicle. Therefore, environmental feature data such as the type (vehicle or pedestrian), shape (rectangular or circular) and motion state (stationary or moving) of each target object can be extracted through the relative position relationship. The extracted environmental feature data is normalized to meet the requirements of the standard normal distribution. Some statistical methods and algorithms, such as mean normalization and variance normalization, can be used to standardize the data.

[0082] In this embodiment, the corresponding features of the target are normalized and the data is converted to [0, 1] to avoid excessive network prediction errors caused by excessive data gaps. Zero-mean normalization can be used, which is a method of converting data to a mean of 0 and a standard deviation of 1. The normalization formula used in the normalization process is as follows:

[0083]

[0084] Where: x′ represents the target environment data after normalization, x represents the environment characteristic data before normalization, μ represents the data mean of the environment characteristic data, and δ represents the standard deviation of the environment characteristic data.

[0085] In one embodiment, the BP neural network module constructed based on the neural network model includes a three-layer BP network architecture of an input layer, a hidden layer and an output layer, a first weight matrix is ​​provided between the input layer and the hidden layer, and a second weight matrix is ​​provided between the hidden layer and the output layer.

[0086] In one implementation manner, step S104 specifically includes the following steps:

[0087] Initializing a first weight matrix and a second weight matrix;

[0088] Inputting the target environment data into the hidden layer in the BP neural network module through the input layer, and outputting the first output data through the hidden layer;

[0089] Inputting the first output data as input data into the output layer, and outputting the second output data through the output layer;

[0090] Calculate an output error based on the second output data, and back-propagate the output error to adjust the weights of the first weight matrix and the second weight matrix;

[0091] Determine whether the output error is less than a preset error threshold;

[0092] If the output error is greater than or equal to the error threshold, the above data processing steps are repeated;

[0093] If the output error is less than the error threshold, the second output data is output as an environmental data point set.

[0094] In this embodiment, the expression formula of the first output data is as follows:

[0095]

[0096] Where: j = 1, 2, ..., m, j represents the jth neuron in the hidden layer, y j represents the first output data output by the jth neuron in the hidden layer, V = (v 1 , v 2 , ..., v n ) T represents the first weight matrix, represents the weight vector corresponding to the jth neuron in the hidden layer, X = (x 1 , x 2 ,...x p ) T represents the target environment data, f(·) represents the activation function;

[0097] The expression formula of the second output data is as follows:

[0098]

[0099] Where: k = 1, 2, ..., m, k represents the kth neuron in the output layer, o k represents the second output data output by the kth neuron in the output layer, W = (w 1 , w 2 , ..., w n ) T represents the second weight matrix, Represents the weight vector corresponding to the kth neuron in the output layer.

[0100] In this embodiment, a three-layer BP network architecture is adopted, which includes an input layer, a hidden layer, and an output layer. The input vector is X=(x 1 , x 2 ,...x p ) T ; The hidden layer output is set to vector Y = (y1 ,y 2 , ..., y n ) T ; The vector of the output layer is set to O = (o 1 , o 2 , ..., o n ) T ; Expected output vector d = (d 1 , d 2 , ..., d n ) T ; The first weight matrix between the input layer and the hidden layer is V = (v 1 , v 2 , ..., v n ) T Represents that the column vector v j represents the weight vector corresponding to the jth neuron in the hidden layer. The second weight matrix from the hidden layer to the output layer is W = (w 1 , w 2 , ..., w n ) T Represents that the column vector w k Represents the weight vector corresponding to the kth neuron in the output layer. The activation function of each layer of neurons uses the sigmoid function, and the expression of the sigmoid function is as follows:

[0101]

[0102] The specific implementation steps of the BP network architecture are as follows: first, define and initialize the number of nodes in the input layer, hidden layer and output layer, the weight matrix of each layer, the bias and activate all nodes of the neural network, and add a bias node to each of the first two layers; secondly, define the signal forward propagation method, take the input data and process it in the hidden layer, multiply the neurons corresponding to the input layer weight W to obtain the neurons corresponding to the hidden layer, and use the sigmoid function to activate the neurons in the hidden layer and save them as the input data of the output layer. The processing steps in the output layer are the same as before, and finally return the activated neurons.

[0103] In this embodiment, the step of calculating the output error based on the second output data and back-propagating the output error to adjust the weights of the first weight matrix and the second weight matrix specifically includes the following steps:

[0104] calculating an output error based on the second output data;

[0105] The output error is back-propagated to the output layer and the hidden layer, and a first error signal of the output layer and a second error signal of the hidden layer are calculated respectively;

[0106] The weights of the second weight matrix are adjusted according to the first error signal, and the weights of the first weight matrix are adjusted according to the second error signal.

[0107] In this embodiment, the back propagation of the error is to add the weight of the error term of a neuron in the lth layer and the error term of the neuron in the l+1th layer connected to the neuron, and then multiply it by the gradient of the activation function of the neuron. In the parameter learning process, the cross entropy loss function is used. For the sample (x, y), the loss function expression is as follows:

[0108]

[0109] Where y represents the neuron vector of the corresponding layer, is the output of the neuron in the corresponding layer. The error term is expressed as the partial derivative of the objective function with respect to the input of the neuron in the lth layer. The input of the neuron in the lth layer is represented by z (l) The error term is represented by δ (l) The error term formula is expressed as

[0110]

[0111] The learning rate is represented by η, and the range of η needs to be controlled in (0, 1*e -5 ), otherwise it will lead to too fast convergence time and too large weight jumps, resulting in oscillation. In addition, the regularization coefficient λ needs to be increased during the weight update process to prevent overfitting. The final weight update steps are:

[0112] w (l) ←w (l) -η(δ (l) (η (l-1) ) T +λw (l) )

[0113] b (l) ←b (l) -ηδ (l)

[0114] In this embodiment, the calculation formula of the output error is as follows:

[0115]

[0116] Where: E p represents the output error of the pth target environment data, l represents the lth layer of neurons in the output layer, represents the expected output value of the p-th target environment data corresponding to the k-th neuron in the output layer, represents the second output data corresponding to the p-th target environment data;

[0117] The calculation formula of the first error signal is as follows:

[0118]

[0119] Where: k = 1, 2, ..., l, represents the first error signal, d k Indicates the expected output value;

[0120] The calculation formula of the second error signal is as follows:

[0121]

[0122] Where: j = 1, 2, ..., m, represents the second error signal, w jk Represents the weight vector corresponding to the k-th neuron in the output layer and the j-th neuron in the hidden layer;

[0123] The formula for adjusting the weights of the second weight matrix according to the first error signal is as follows:

[0124]

[0125] The formula for adjusting the weights of the first weight matrix according to the second error signal is as follows:

[0126]

[0127] Where: η represents the learning rate.

[0128] The present invention also discloses a lane environment information fitting system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the lane environment information fitting method described in any one of the above-mentioned embodiments is implemented.

[0129] In one embodiment, referring to Figure 2 , the lane environment information fitting system specifically includes:

[0130] A data acquisition module, used to acquire basic environmental data of the target vehicle's surrounding environment through the target vehicle's on-board millimeter-wave radar;

[0131] A data preprocessing module is used to preprocess basic environmental data to obtain target environmental data with standard normal distribution;

[0132] A neural network module is used to construct a BP neural network module based on a neural network model, input target environment data into the BP neural network module, and output an environment data point set through the BP neural network module;

[0133] The data fitting module is used to fit the lane environment information around the target vehicle based on the environmental data point set.

[0134] The implementation principle of this embodiment is:

[0135] By calling the program and executing the steps of the above modules, the lane line can be fitted through processing and handling based on the static target position information output by the millimeter-wave radar sensor, rather than fitting the lane line based on the dynamic target information output after the radar is fused with other sensors. According to the installation position of the millimeter-wave radar and the coordinate system, a coordinate conversion is performed based on the coordinate system of the autonomous driving vehicle to obtain the relative position of the autonomous driving vehicle and the static targets on both sides of the lane, and the position of the vehicle is obtained according to the interaction of the current lane line detected by the on-board visual camera; in addition, the linear curve is fitted according to the positional relationship between the static targets on both sides of the lane, which is the lane type and direction. The present invention makes up for the problem that lane environment information cannot be output without a high-precision map, and the cost of the millimeter-wave radar is relatively low. While replacing the high-precision map to output lane environment information, a lot of costs can also be saved.

[0136] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the lane environment information fitting method described in any of the above-mentioned embodiments are implemented.

[0137] The implementation principle of this embodiment is:

[0138] By calling the program, the lane line can be fitted through processing and treatment based on the static target position information output by the millimeter-wave radar sensor, rather than fitting the lane line based on the dynamic target information output after the radar is fused with other sensors. According to the installation position of the millimeter-wave radar and the coordinate system, the coordinate conversion is performed based on the coordinate system of the autonomous driving vehicle to obtain the relative position of the autonomous driving vehicle and the static targets on both sides of the lane, and the position of the vehicle is obtained according to the current lane line detected by the on-board visual camera; in addition, the linear curve is fitted according to the positional relationship between the static targets on both sides of the lane, which is the lane type and direction. The present invention makes up for the problem that lane environment information cannot be output without a high-precision map, and the cost of millimeter-wave radar is relatively low. While replacing the high-precision map to output lane environment information, it can also save a lot of costs.

[0139] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as above, which are not provided in detail for the sake of simplicity.

[0140] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A lane environment information fitting method, It is characterized in that The steps include: Acquiring basic environmental data of the surrounding environment of the target vehicle through the on-board millimeter-wave radar of the target vehicle; Preprocessing the basic environmental data to obtain target environmental data with standard normal distribution; Construct BP neural network module based on neural network model; Inputting the target environment data into the BP neural network module, and outputting an environment data point set through the BP neural network module; Lane environment information around the target vehicle is fitted according to the environmental data point set.

2. The lane environment information fitting method according to claim 1, It is characterized in that The preprocessing of the basic environment data to obtain target environment data with standard normal distribution comprises the following steps: Parsing the basic environment data according to the communication protocol corresponding to the vehicle-mounted millimeter-wave radar to obtain target structured data; Based on the spatial coordinate system where the target vehicle is located, the target structured data is coordinate-converted to obtain the relative position relationship between the physical target in the surrounding environment and the target vehicle; Extracting environmental feature data with the entity target as the main body from the target structured data based on the relative position relationship; The environmental characteristic data are normalized to obtain target environmental data with standard normal distribution.

3. The lane environment information fitting method according to claim 2, Features ,The normalization formula used in the normalization process is as follows: In the formula: x′ represents the target environment data after normalization, x represents the environment characteristic data before normalization, μ represents the data mean of the environment characteristic data, and δ represents the standard deviation of the environment characteristic data.

4. The lane environment information fitting method according to claim 1, It is characterized in that The BP neural network module includes a three-layer BP network architecture of an input layer, a hidden layer and an output layer. There is a first weight matrix between the input layer and the hidden layer, and there is a second weight matrix between the hidden layer and the output layer.

5. The lane environment information fitting method according to claim 4, It is characterized in that The step of inputting the target environment data into the BP neural network module and outputting the environment data point set through the BP neural network module comprises the following steps: Initializing the first weight matrix and the second weight matrix; Inputting the target environment data into the hidden layer in the BP neural network module through the input layer, and outputting first output data through the hidden layer; Inputting the first output data as input data into the output layer, and outputting second output data through the output layer; Calculating an output error based on the second output data, and back-propagating the output error to adjust the weights of the first weight matrix and the second weight matrix; Determining whether the output error is less than a preset error threshold; If the output error is greater than or equal to the error threshold, then repeat the above data processing steps; If the output error is less than the error threshold, the second output data is output as an environmental data point set.

6. The lane environment information fitting method according to claim 5, It is characterized in that The expression formula of the first output data is as follows: Where: j = 1, 2, ..., m, j represents the jth neuron in the hidden layer, y j represents the first output data output by the jth neuron in the hidden layer, V=(v 1 , v 2 , ..., v n ) T represents the first weight matrix, represents the weight vector corresponding to the jth neuron in the hidden layer, X=(x 1 , x 2 ,...x p ) T represents the target environment data, and f(·) represents the activation function; The expression formula of the second output data is as follows: Where: k = 1, 2, ..., m, k represents the kth neuron in the output layer, o k represents the second output data output by the kth neuron in the output layer, W=(w 1 , w 2 , ..., w n ) T represents the second weight matrix, Represents the weight vector corresponding to the kth neuron in the output layer.

7. The lane environment information fitting method according to claim 6, It is characterized in that The step of calculating the output error based on the second output data and back-propagating the output error to adjust the weights of the first weight matrix and the second weight matrix comprises the following steps: calculating an output error based on the second output data; Back-propagating the output error to the output layer and the hidden layer, and respectively calculating a first error signal of the output layer and a second error signal of the hidden layer; The weights of the second weight matrix are adjusted according to the first error signal, and the weights of the first weight matrix are adjusted according to the second error signal.

8. The lane environment information fitting method according to claim 7, It is characterized in that The calculation formula of the output error is as follows: Where: E p represents the output error of the pth target environment data, l represents the lth layer of neurons in the output layer, represents the expected output value of the p-th target environment data corresponding to the k-th neuron of the output layer, represents the second output data corresponding to the p-th target environment data; The calculation formula of the first error signal is as follows: Where: k = 1, 2, ..., l, represents the first error signal, d k Indicates the expected output value; The calculation formula of the second error signal is as follows: Where: j = 1, 2, ..., m, represents the second error signal, w jk A weight vector representing the kth neuron in the output layer and the jth neuron in the hidden layer; The formula for adjusting the weights of the second weight matrix according to the first error signal is as follows: The formula for adjusting the weights of the first weight matrix according to the second error signal is as follows: Where: η represents the learning rate.

9. A lane environment information fitting system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the lane environment information fitting method as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the lane environment information fitting method according to any one of claims 1 to 8 are implemented.

Images