Unknown environment-oriented intelligent vehicle collaborative positioning method

Abstract

An unknown environment-oriented intelligent vehicle collaborative positioning method, comprising: first, constructing a collaborative positioning system; second, creating a collaborative positioning observation model; then, training the collaborative positioning observation model for predicting the position increment of an intelligent vehicle; and finally, constructing a collaborative fusion positioning model to acquire the position of the intelligent vehicle. The unknown environment-oriented intelligent vehicle collaborative positioning method disclosed in the present invention realizes intelligent vehicle collaborative positioning in unknown environments having missing prior information and limited observation information.

Description

A collaborative positioning method for intelligent vehicles in unknown environments Technical Field

[0001] The present invention belongs to the field of intelligent vehicle collaborative positioning, and in particular relates to an intelligent vehicle collaborative positioning method in an unknown environment. Background Art

[0002] In recent years, thanks to advances in deep learning, sensors, and information processing technologies, smart cars have experienced rapid development and are gradually being applied to fields such as smart transportation and emergency rescue. Obtaining accurate location information is essential for carrying out subsequent operations. While the Global Navigation Satellite System (GNSS) can provide accurate location information in most situations, it can fail in semi- or fully obscured environments such as urban canyons, boulevards, and tunnels. Collaborative positioning based on wireless sensor networks (WSNs) such as ultra-wideband, cellular, and Bluetooth is a solution for satellite failures. Generally speaking, a collaborative positioning system consists of a receiving station and a base station. For smart cars, the receiving station is typically fixed to the car, while the base station is typically fixed to the roadside. Collaborative positioning uses wireless sensors to measure the distance or angle between the car and the receiving station, and then uses triangulation to determine the car's location.

[0003] Therefore, existing collaborative positioning methods usually require the location information of roadside base stations, and four or more base stations are required to solve the three-dimensional position information of the smart car. However, in many cases, the prior location information of the roadside is missing, that is, the environment is unknown. In such an environment, the smart car system does not know the prior location information of the roadside base station, which makes the traditional triangulation positioning method unusable; in addition, the continuous deployment of multiple roadside base stations will also bring additional costs, and at least four roadside base stations are required to achieve three-dimensional positioning, which to a certain extent limits the application of collaborative positioning. Therefore, the existing collaborative positioning faces the following problems: (1) It is impossible to achieve collaborative positioning when the prior location information of the roadside base station is missing; (2) It is impossible to achieve collaborative positioning of the three-dimensional position of the smart car when there are less than four roadside base stations. Therefore, in the above unknown environment, the collaborative positioning of smart cars still faces challenges.

[0004] Summary of the Invention

[0005] The present invention proposes a method for cooperative positioning of intelligent vehicles in unknown environments, with the following specific features: (1) It does not require prior knowledge of the location information of roadside base stations. The method proposed in the present invention can achieve positioning using only the distance observation information provided by the roadside base stations without knowing the location information of the roadside base stations; (2) Only the observation information of one roadside base station is required to achieve three-dimensional positioning of the intelligent vehicle. The method proposed in the present invention does not require the observation information of at least four roadside base stations, but only the distance observation information provided by one roadside base station to achieve positioning.

[0006] The present invention first constructs a collaborative positioning system, then establishes a collaborative positioning observation model, then trains the model and uses it to predict the position increment of the smart car, and finally constructs a collaborative fusion positioning model to obtain the position of the smart car.

[0007] The idea of ​​the present invention is further described below:

[0008] Step 1: Build a collaborative positioning system

[0009] The collaborative positioning system in the present invention mainly includes a smart car and a roadside base station. The smart car is equipped with an inertial navigation system (INS), which can obtain the speed information of the smart car through integration calculation. In addition, the smart car is also fixed with a wireless receiving device for receiving the distance information between the smart car and the base station, and the roadside base station is fixed with a wireless transmitting device for measuring and providing distance information.

[0010] Step 2: Establish collaborative positioning observation model

[0011] Sub-step 1: Build model input

[0012] Taking into account that the position change of the smart car is mainly caused by changes in its own speed, attitude angle, etc., in addition, the position change of the smart car itself can also be reflected in the change in distance measured by the roadside unit. Therefore, a mathematical model of the position change of the smart car is established in the present invention. The model takes the speed change of the smart car, the attitude angle change, and the distance change between the smart car and the roadside base station per unit time as input, and the position change of the smart car per unit time as output.

[0013] Furthermore, in order to better decouple the position changes in each direction, the present invention constructs a multi-layer perceptron model for prediction of the position changes in the east, north and sky directions. The input of the east position change prediction model is the east velocity change Δv x , heading angle change Δh and distance change Δd, the output is the east position change Δx, and the input of the north position change prediction model is the north velocity change Δv y, heading angle change Δh and distance change Δd, the output is the north position change Δy, and the input of the celestial position change prediction model is the celestial velocity change Δv z , pitch angle change Δp and distance change Δd, and the output is the celestial position change Δz.

[0014] Sub-step 2: Building the model network structure

[0015] Considering that changes in speed, posture, etc. are time series data, a long short-term memory network (LSTM) is first used to process the time series data. In the present invention, the input of each model includes speed changes, posture angle changes, and distance changes. The time length of the input sequence is 10, that is, the input layer dimension is 30, the number of LSTM layers is set to 1, and the number of hidden nodes is 64. Then, the fully connected layer is used to continue to regress the position increment, and the output of LSTM is used as the input of the fully connected layer. In the present invention, there are two fully connected layers, each with 11 nodes, and the final output layer dimension is 1, which is the predicted position increment.

[0016] Step 3: Collect data to train the model

[0017] Data collected from the collaborative positioning system primarily includes the changes in the smart vehicle's eastbound, northbound, and celestial speeds per unit time, the changes in the smart vehicle's attitude angle, and the changes in the distance between the smart vehicle and the base station. The smart vehicle's speed and attitude angle changes can be provided by the INS, assuming the smart vehicle's initial absolute position and attitude are known. Distance changes are provided by the roadside base station. After data collection, the model established in the above steps is trained using the mean square error (MSE) as the loss function. Training stops when the number of training attempts reaches a certain threshold or when the error falls below a certain threshold.

[0018] Step 4: Use the trained model to obtain collaborative observations

[0019] Based on the model trained in the above steps, during actual operation of the smart car, the vehicle's speed and posture changes, as well as the distance changes obtained from the roadside base station, are input into the trained model to predict the smart car's three-dimensional position changes Δx, Δy, and Δz. The collaborative observation information of the smart car's position is then obtained based on the smart car's previous position information.

[0020] In the above formula, the subscript represents time. Finally, when the location of the roadside base station is unknown and there is only one base station, the collaborative observation of the smart car's position is obtained.

[0021] Step 5: Build a collaborative fusion positioning model

[0022] A collaborative fusion positioning model based on Kalman filtering is constructed to fuse the collaborative positioning observation information obtained in the above steps with the information provided by the intelligent vehicle's INS. The INS information is used for time updates, while the collaborative positioning observation information is used for measurement updates. A more detailed description of the Kalman filter-based fusion model can be found in the reference (Qin Yongyuan, Zhang Hongyue, and Wang Shuhua. Kalman Filter and Integrated Navigation [M]. Northwestern Polytechnical University Press, 2012).

[0023] The beneficial effects of the present invention are as follows: the collaborative positioning method of intelligent vehicles for unknown environments proposed in the present invention realizes the collaborative positioning of intelligent vehicles in unknown environments where the position information of roadside base stations is unknown and observation information is insufficient, which effectively promotes the further development of collaborative positioning technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] FIG1 is a flow chart of the method of the present invention.

[0025] Figure 2 is a schematic diagram of the collaborative positioning observation model. DETAILED DESCRIPTION

[0026] In recent years, thanks to advances in deep learning, sensors, and information processing technologies, smart cars have experienced rapid development and are gradually being applied to fields such as smart transportation and emergency rescue. Obtaining accurate location information is essential for carrying out subsequent operations. While the Global Navigation Satellite System (GNSS) can provide accurate location information in most situations, it can fail in semi- or fully obscured environments such as urban canyons, boulevards, and tunnels. Collaborative positioning based on wireless sensor networks (WSNs) such as ultra-wideband, cellular, and Bluetooth is a solution for satellite failures. Generally speaking, a collaborative positioning system consists of a receiving station and a base station. For smart cars, the receiving station is typically fixed to the car, while the base station is typically fixed to the roadside. Collaborative positioning uses wireless sensors to measure the distance or angle between the car and the receiving station, and then uses triangulation to determine the car's location.

[0027] Therefore, existing collaborative positioning methods usually require the location information of roadside base stations, and four or more base stations are required to solve the three-dimensional position information of the smart car. However, in many cases, the prior location information of the roadside is missing, that is, the environment is unknown. In such an environment, the smart car system does not know the prior location information of the roadside base station, which makes the traditional triangulation positioning method unusable; in addition, the continuous deployment of multiple roadside base stations will also bring additional costs, and at least four roadside base stations are required to achieve three-dimensional positioning, which to a certain extent limits the application of collaborative positioning. Therefore, the existing collaborative positioning faces the following problems: (1) It is impossible to achieve collaborative positioning when the prior location information of the roadside base station is missing; (2) It is impossible to achieve collaborative positioning of the three-dimensional position of the smart car when there are less than four roadside base stations. Therefore, in the above unknown environment, the collaborative positioning of smart cars still faces challenges.

[0028] To address the above issues, the present invention proposes a collaborative positioning method for intelligent vehicles in unknown environments, with the following specific features: (1) It does not require the prior location information of the roadside base station. The method proposed in the present invention can achieve positioning using only the distance observation information provided by the roadside base station without knowing the location information of the roadside base station; (2) Only the observation information of one roadside base station is required to achieve three-dimensional positioning of the intelligent vehicle. The method proposed in the present invention does not require the observation information of at least four roadside base stations, but only the distance observation information provided by one roadside base station to achieve positioning.

[0029] The collaborative positioning method for intelligent vehicles in unknown environments proposed in the present invention realizes the collaborative positioning of intelligent vehicles in unknown environments where the position information of roadside base stations is unknown and observation information is insufficient, which effectively promotes the further development of collaborative positioning technology.

[0030] The idea of ​​the present invention is further described below:

[0031] Step 1: Build a collaborative positioning system

[0032] The collaborative positioning system in the present invention mainly includes a smart car and a roadside base station. The smart car is equipped with an inertial navigation system (INS), which can obtain the speed information of the smart car through integration calculation. In addition, the smart car is also fixed with a wireless receiving device for receiving the distance information between the smart car and the base station, and the roadside base station is fixed with a wireless transmitting device for measuring and providing distance information.

[0033] Step 2: Establish collaborative positioning observation model

[0034] Sub-step 1: Build model input

[0035] Taking into account that the position change of the smart car is mainly caused by changes in its own speed, attitude angle, etc., in addition, the position change of the smart car itself can also be reflected in the change in distance measured by the roadside unit. Therefore, a mathematical model of the position change of the smart car is established in the present invention. The model takes the speed change of the smart car, the attitude angle change, and the distance change between the smart car and the roadside base station per unit time as input, and the position change of the smart car per unit time as output.

[0036] Furthermore, in order to better decouple the position changes in each direction, the present invention constructs a multi-layer perceptron model for prediction of the position changes in the east, north and sky directions. The input of the east position change prediction model is the east velocity change Δv x , heading angle change Δh and distance change Δd, the output is the east position change Δx, and the input of the north position change prediction model is the north velocity change Δv y , heading angle change Δh and distance change Δd, the output is the north position change Δy, and the input of the celestial position change prediction model is the celestial velocity change Δv z , pitch angle change Δp and distance change Δd, and the output is the celestial position change Δz.

[0037] Sub-step 2: Building the model network structure

[0038] Considering that changes in speed, posture, etc. are time series data, a long short-term memory network (LSTM) is first used to process the time series data. In the present invention, the input of each model includes speed changes, posture angle changes, and distance changes. The time length of the input sequence is 10, that is, the input layer dimension is 30, the number of LSTM layers is set to 1, and the number of hidden nodes is 64. Then, the fully connected layer is used to continue to regress the position increment, and the output of LSTM is used as the input of the fully connected layer. In the present invention, there are two fully connected layers, each with 11 nodes, and the final output layer dimension is 1, which is the predicted position increment.

[0039] Step 3: Collect data to train the model

[0040] Data collected from the collaborative positioning system primarily includes the changes in the smart vehicle's eastbound, northbound, and celestial speeds per unit time, the changes in the smart vehicle's attitude angle, and the changes in the distance between the smart vehicle and the base station. The smart vehicle's speed and attitude angle changes can be provided by the INS, assuming the smart vehicle's initial absolute position and attitude are known. Distance changes are provided by the roadside base station. After data collection, the model established in the above steps is trained using the mean square error (MSE) as the loss function. Training stops when the number of training attempts reaches a certain threshold or when the error falls below a certain threshold.

[0041] Step 4: Use the trained model to obtain collaborative observations

[0042] Based on the model trained in the above steps, during actual operation of the smart car, the vehicle's speed and posture changes, as well as the distance changes obtained from the roadside base station, are input into the trained model to predict the smart car's three-dimensional position changes Δx, Δy, and Δz. The collaborative observation information of the smart car's position is then obtained based on the smart car's previous position information.

[0043] In the above formula, the subscript represents time. Finally, when the location of the roadside base station is unknown and there is only one base station, the collaborative observation of the smart car's position is obtained.

[0044] Step 5: Build a collaborative fusion positioning model

[0045] A collaborative fusion positioning model based on Kalman filtering is constructed to fuse the collaborative positioning observation information obtained in the above steps with the information provided by the intelligent vehicle's INS. The INS information is used for time updates, while the collaborative positioning observation information is used for measurement updates. A more detailed description of the Kalman filter-based fusion model can be found in the reference (Qin Yongyuan, Zhang Hongyue, and Wang Shuhua. Kalman Filter and Integrated Navigation [M]. Northwestern Polytechnical University Press, 2012).

Claims

1. A method for cooperative positioning of intelligent vehicles in an unknown environment, characterized in that: First, a collaborative positioning system was built. Second, a collaborative positioning observation model was established. Then, the model was trained and used to predict the position increment of the smart car. Finally, a collaborative fusion positioning model was built to obtain the position of the smart car. The specific steps include: Step 1: Build a collaborative positioning system The collaborative positioning system includes a smart car and a roadside base station. The smart car is equipped with an inertial navigation system, which obtains the speed information of the smart car through integration calculation. In addition, the smart car is also fixed with a wireless receiving device for receiving the distance information between the smart car and the base station, and the roadside base station is fixed with a wireless transmitting device for measuring and providing distance information. Step 2: Establish collaborative positioning observation model Sub-step 1: Build model input Considering that the position change of the smart car is caused by changes in its own speed, attitude angle, etc., in addition, the position change of the smart car itself can also be reflected in the change in the distance measured by the roadside unit. Therefore, a mathematical model of the position change of the smart car is established. The model takes the speed change of the smart car, the attitude angle change, and the distance change between the smart car and the roadside base station per unit time as input, and the position change of the smart car per unit time as output. In order to better decouple the position changes in each direction, multi-layer perceptron models are constructed for prediction of the position changes in the east, north and sky directions. The input of the east position change prediction model is the east velocity change Δv x , heading angle change Δh and distance change Δd, the output is the east position change Δx, and the input of the north position change prediction model is the north velocity change Δv y , heading angle change Δh and distance change Δd, the output is the north position change Δy, and the input of the celestial position change prediction model is the celestial velocity change Δv z , pitch angle change Δp and distance change Δd, the output is the celestial position change Δz; Sub-step 2: Building the model network structure Considering that speed and posture changes are time series data, long short-term memory network LSTM is first used to process time series data. In the present invention, the input of each model includes speed change, posture angle change and distance change. The time length of the input sequence is 10, that is, the input layer dimension is 30, the number of LSTM layers is set to 1, and the number of hidden nodes is 64. Then, the fully connected layer is used to continue to regress the position increment. The output of LSTM is used as the input of the fully connected layer. The fully connected layer has two layers, each with 11 nodes, and the final output layer dimension is 1, which is the predicted position increment. Step 3: Collect data to train the model Collect data from the collaborative positioning system, including the eastward speed, northward speed and The change value of the celestial speed, the change value of the attitude angle of the smart car, and the change value of the distance between the smart car and the base station. Among them, the change values ​​of the speed and attitude angle of the smart car are provided by INS, provided that the initial absolute position and absolute attitude of the smart car are known, and the distance change is provided by the roadside base station. After collecting data, the model established in the above steps is trained, and the loss function is the mean square error. When the number of training times reaches a certain threshold or the error is less than a certain threshold, the training stops; Step 4: Use the trained model to obtain collaborative observations Based on the model trained in the above steps, when the smart car is actually running, the speed and posture changes of the smart car, as well as the distance changes obtained from the roadside base station, are input into the above trained model to predict the three-dimensional position changes Δx, Δy, Δz of the smart car. Then, based on the position information of the smart car at the previous moment, the collaborative observation information of the smart car position is finally obtained. In the above formula, the subscript represents time. Finally, when the location of the roadside base station is unknown and there is only one base station, the collaborative observation of the position of the smart car is obtained; Step 5: Build a collaborative fusion positioning model A collaborative fusion positioning model based on Kalman filtering is constructed to fuse the collaborative positioning observation information obtained in the above steps with the information provided by the smart car INS, wherein the information provided by the INS is used for time update, and the collaborative positioning observation information is used for measurement update.

Images