A multi-sensor collaborative positioning method for non-line-of-sight scenarios

Abstract

The present invention discloses a multi-sensor collaborative positioning method for non-line-of-sight scenarios. First, a multi-layer perceptron is constructed to suppress non-line-of-sight errors. Then, a weighted least squares solution is performed on the corrected signal. Finally, the multi-sensor information is fused to obtain the position. The multi-sensor collaborative positioning method for non-line-of-sight scenarios disclosed by the present invention effectively overcomes the defect that the existing collaborative positioning method is easily affected by non-line-of-sight errors and fails.

Description

Technical Field

[0001] The present invention belongs to the field of ground unmanned vehicle positioning, and in particular relates to a ground unmanned vehicle collaborative positioning method in a non-line-of-sight scenario. Background Art

[0002] In recent years, ground unmanned vehicles represented by driverless cars and intelligent robots have gradually begun to be used in the fields of intelligent transportation, logistics and transportation, emergency rescue, etc., which has improved the degree of unmanned and intelligent in related fields and further improved efficiency on the basis of ensuring safety. Among them, accurate and reliable location information is the basis for the realization of autonomous functions such as path planning, decision-making and control of ground unmanned vehicles. However, the Global Navigation Satellite System (GNSS), as the most commonly used positioning method, will be inaccurate or fail in semi-occluded or occluded conditions, making it difficult to provide accurate location information. Collaborative positioning based on wireless sensor networks (WSN) such as cellular, Bluetooth, and ultra-wideband, as a novel positioning method, provides a solution for the positioning of ground unmanned vehicles when GNSS fails. The collaborative positioning system is mainly composed of a base station and a receiving station. Among them, the base station is fixed and the location information is known, and the receiving station is fixed on the land unmanned vehicle. The two transmit signals through the above-mentioned wireless sensor network. Based on the distance or angle between the base station and the receiving station measured by multiple groups of wireless sensor networks, the position of the receiving station can be solved using the least squares method (LS) to achieve positioning.

[0003] However, the accuracy of collaborative positioning will be affected by non-line-of-sight (NLOS) propagation. Non-line-of-sight propagation refers to the presence of dynamic and static obstacles in the path of wireless signal propagation, which causes signal occlusion and delay, thereby increasing the ranging error and causing errors in the final positioning solution. Therefore, being able to identify non-line-of-sight signals and then suppressing non-line-of-sight errors is the key to ensuring the effectiveness and reliability of collaborative positioning. Existing technical means have been able to better identify non-line-of-sight signals. However, for the identified non-line-of-sight signals, there is still a lack of certain methods to suppress non-line-of-sight errors. At present, most collaborative positioning adopts the strategy of discarding non-line-of-sight signals to prevent non-line-of-sight noise from polluting the positioning results. However, this will increase the cost of deploying base stations on the one hand, and on the other hand, it will cause the positioning error to increase infinitely in the case of long-term non-line-of-sight conditions. In addition, there are also non-line-of-sight error suppression methods based on mathematical modeling, etc., which attempt to recover real data from non-line-of-sight signals. However, such methods are relatively complex, have a large amount of calculation, and are difficult to meet the real-time requirements of collaborative positioning. In summary, the existing methods have the disadvantages of insufficient reliability and high cost. Summary of the invention

[0004] The present invention proposes a multi-sensor collaborative positioning method that integrates wireless sensor networks and inertial navigation data to ensure reliable positioning in non-line-of-sight scenarios. The specific features are as follows: (1) High reliability. Based on the vehicle's motion state, a relationship between the multi-layer perceptron processing inertial navigation data and the change in wireless signal is established to correct the non-line-of-sight signal, and the weighted least squares method is used to process the corrected value to obtain the final observation signal, which effectively suppresses the error caused by the non-line-of-sight effect. The multi-layer perceptron has a small number of layers and parameters, and can meet the real-time requirements of collaborative positioning. (2) Low cost. The present invention makes full use of the location information and measurement signal information of each base station, without the need to deploy more base stations, and does not require high equipment computing power, effectively reducing the cost of collaborative positioning and being more conducive to the promotion and application of technology.

[0005] The idea of ​​the present invention is further described below:

[0006] Step 1: Non-line-of-sight error suppression

[0007] Using the existing non-line-of-sight signal discrimination method, the wireless signals in the collaborative positioning are divided into line-of-sight signals and non-line-of-sight signals;

[0008] Sub-step 1: Building a multi-layer perceptron

[0009] A multilayer perceptron is a neural network that can handle complex nonlinear regression problems. It is generally composed of input, intermediate layers, and output. The detailed principle can be found in the reference (Zhou Zhihua. Machine Learning [M]. Tsinghua University Press, 2016.). In the actual collaborative positioning process, the changes in each wireless signal are caused by the movement of the land unmanned vehicle, and the movement state of the land unmanned vehicle can be obtained through inertial navigation. A multilayer perceptron is constructed for the wireless signal of each base station, and its input is the speed of the inertial navigation solution. and The subscript k represents the time, the superscripts x and y represent the coordinate axes, and the output is the change in the distance between the land unmanned vehicle and each base station. Among them, the subscript k represents the time, the superscript i represents the base station number, and the middle layer of the multi-layer perceptron is set to two layers to handle nonlinear relationships;

[0010] Sub-step 2: Training or prediction based on signal category

[0011] If the signal is determined to be a line-of-sight signal, the multi-layer perceptron executes the training mode, takes the speed calculated by the inertial navigation as input, and the change calculated by the line-of-sight signal as the true value, and trains the multi-layer perceptron to adjust the parameters; if it is determined to be a non-line-of-sight signal, the multi-layer perceptron enters the prediction mode, takes the speed as input, and outputs the signal change Then predict the actual distance information

[0012] Step 2: Calculate position information

[0013] When all received wireless signals are determined to be line-of-sight signals, the least square method is used to solve the position; when there are non-line-of-sight signals, the weighted least square method is used to solve the position. The detailed methods of least squares and weighted least squares can be found in the reference (Yang Xiaoqian, Tong Hao, Wang Bin. A weighted least squares positioning algorithm based on ranging information [C] / / The 11th China Satellite Navigation Annual Conference.). Note:

[0014]

[0015] in:

[0016]

[0017] In the above formula, is the value of the i-th wireless signal restored by the multi-layer perceptron at time k, is the actual measured signal value, then the weight matrix W is:

[0018] W=R -1 #(3)

[0019] After obtaining the weight matrix, the position information is solved by the weighted least squares method;

[0020] Step 3: Fusion of corrected wireless observations and inertial navigation

[0021] The inertial navigation recursion result is used as the state quantity, and the position information solved in step 2 is used as the observation quantity. The Kalman filter equation is established for fusion calculation to obtain the final position information. For the specific Kalman filter method, please refer to the reference (Qin Yongyuan, Zhang Hongyue, Wang Shuhua. Kalman Filter and Integrated Navigation [M]. Northwestern Polytechnical University Press, 2012). The observation quantity covariance in the Kalman filter is R0. In order to better adapt to the changes in the quality of the wireless signal, the observation covariance is adaptively modified as follows:

[0022] R0=cR0#(4)

[0023] Among them, c is the correction coefficient, which is used to correct the original observation covariance. Its value is related to the number of non-line-of-sight signals in the wireless signal. The more non-line-of-sight signals there are, the larger its value is.

[0024] Compared with the prior art, the beneficial effects of the present invention are: the multi-sensor collaborative positioning method for non-line-of-sight scenarios proposed in the present invention effectively suppresses non-line-of-sight errors, makes up for the defects of insufficient reliability and high cost of the prior art methods, ensures the accuracy and reliability of collaborative positioning, and strongly promotes the further development of collaborative positioning technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a positioning method framework diagram;

[0026] Figure 2 It is a schematic diagram of a multi-layer perceptron. DETAILED DESCRIPTION

[0027] In recent years, ground unmanned vehicles represented by driverless cars and intelligent robots have gradually begun to be used in the fields of intelligent transportation, logistics and transportation, emergency rescue, etc., which has improved the degree of unmanned and intelligent in related fields and further improved efficiency on the basis of ensuring safety. Among them, accurate and reliable location information is the basis for the realization of autonomous functions such as path planning, decision-making and control of ground unmanned vehicles. However, the Global Navigation Satellite System (GNSS), as the most commonly used positioning method, will be inaccurate or fail in semi-occluded or occluded conditions, making it difficult to provide accurate location information. Collaborative positioning based on wireless sensor networks (WSN) such as cellular, Bluetooth, and ultra-wideband, as a novel positioning method, provides a solution for the positioning of ground unmanned vehicles when GNSS fails. The collaborative positioning system is mainly composed of a base station and a receiving station. Among them, the base station is fixed and the location information is known, and the receiving station is fixed on the land unmanned vehicle. The two transmit signals through the above-mentioned wireless sensor network. Based on the distance or angle between the base station and the receiving station measured by multiple groups of wireless sensor networks, the position of the receiving station can be solved using the least squares method (LS) to achieve positioning.

[0028] However, the accuracy of collaborative positioning will be affected by non-line-of-sight (NLOS) propagation. Non-line-of-sight propagation refers to the presence of dynamic and static obstacles in the path of wireless signal propagation, which causes signal occlusion and delay, thereby increasing the ranging error and causing errors in the final positioning solution. Therefore, being able to identify non-line-of-sight signals and then suppressing non-line-of-sight errors is the key to ensuring the effectiveness and reliability of collaborative positioning. Existing technical means have been able to better identify non-line-of-sight signals. However, for the identified non-line-of-sight signals, there is still a lack of certain methods to suppress non-line-of-sight errors. At present, most collaborative positioning adopts the strategy of discarding non-line-of-sight signals to prevent non-line-of-sight noise from polluting the positioning results. However, this will increase the cost of deploying base stations on the one hand, and on the other hand, it will cause the positioning error to increase infinitely in the case of long-term non-line-of-sight conditions. In addition, there are also non-line-of-sight error suppression methods based on mathematical modeling, etc., which attempt to recover real data from non-line-of-sight signals. However, such methods are relatively complex, have a large amount of calculation, and are difficult to meet the real-time requirements of collaborative positioning. In summary, the existing methods have the disadvantages of insufficient reliability and high cost.

[0029] In response to the above problems, the present invention proposes a multi-sensor collaborative positioning method that integrates wireless sensor networks and inertial navigation data to ensure reliable positioning in non-line-of-sight scenarios. The specific features are as follows: (1) High reliability. Based on the vehicle's motion state, a relationship between the multi-layer perceptron processing inertial navigation data and the change in wireless signal is established to correct the non-line-of-sight signal, and the weighted least squares method is used to process the corrected value to obtain the final observation signal, which effectively suppresses the error caused by the non-line-of-sight effect. The multi-layer perceptron has a small number of layers and parameters, and can meet the real-time requirements of collaborative positioning. (2) Low cost. The present invention makes full use of the location information and measurement signal information of each base station, without the need to deploy more base stations, and does not require high computing power for the equipment, effectively reducing the cost of collaborative positioning and being more conducive to the promotion and application of technology.

[0030] The multi-sensor collaborative positioning method for non-line-of-sight scenarios proposed in the present invention effectively suppresses non-line-of-sight errors, makes up for the shortcomings of existing methods such as insufficient reliability and high cost, ensures the accuracy and reliability of collaborative positioning, and strongly promotes the further development of collaborative positioning technology.

[0031] The idea of ​​the present invention is further described below:

[0032] Step 1: Non-line-of-sight error suppression

[0033] Using the existing non-line-of-sight signal discrimination method, the wireless signals in the collaborative positioning are divided into line-of-sight signals and non-line-of-sight signals;

[0034] Sub-step 1: Building a multi-layer perceptron

[0035] A multilayer perceptron is a neural network that can handle complex nonlinear regression problems. It is generally composed of input, intermediate layers, and output. The detailed principle can be found in the reference (Zhou Zhihua. Machine Learning [M]. Tsinghua University Press, 2016.). In the actual collaborative positioning process, the changes in each wireless signal are caused by the movement of the land unmanned vehicle, and the movement state of the land unmanned vehicle can be obtained through inertial navigation. A multilayer perceptron is constructed for the wireless signal of each base station, and its input is the speed of the inertial navigation solution. and The subscript k represents the time, the superscripts x and y represent the coordinate axes, and the output is the change in the distance between the land unmanned vehicle and each base station. Among them, the subscript k represents the time, the superscript i represents the base station number, and the middle layer of the multi-layer perceptron is set to two layers to handle nonlinear relationships;

[0036] Sub-step 2: Training or prediction based on signal category

[0037] If the signal is determined to be a line-of-sight signal, the multi-layer perceptron executes the training mode, takes the speed calculated by the inertial navigation as input, and the change calculated by the line-of-sight signal as the true value, and trains the multi-layer perceptron to adjust the parameters; if it is determined to be a non-line-of-sight signal, the multi-layer perceptron enters the prediction mode, takes the speed as input, and outputs the signal change Then predict the actual distance information

[0038] Step 2: Calculate position information

[0039] When all received wireless signals are determined to be line-of-sight signals, the least square method is used to solve the position; when there are non-line-of-sight signals, the weighted least square method is used to solve the position. The detailed methods of least squares and weighted least squares can be found in the reference (Yang Xiaoqian, Tong Hao, Wang Bin. A weighted least squares positioning algorithm based on ranging information [C] / / The 11th China Satellite Navigation Annual Conference.). Note:

[0040]

[0041] in:

[0042]

[0043] In the above formula, is the value of the i-th wireless signal restored by the multi-layer perceptron at time k, is the actual measured signal value, then the weight matrix W is:

[0044] W=R -1 #(3)

[0045] After obtaining the weight matrix, the position information is solved by the weighted least squares method;

[0046] Step 3: Fusion of corrected wireless observations and inertial navigation

[0047] The inertial navigation recursion result is used as the state quantity, and the position information solved in step 2 is used as the observation quantity. The Kalman filter equation is established for fusion calculation to obtain the final position information. For the specific Kalman filter method, please refer to the reference (Qin Yongyuan, Zhang Hongyue, Wang Shuhua. Kalman Filter and Integrated Navigation [M]. Northwestern Polytechnical University Press, 2012). The observation quantity covariance in the Kalman filter is R0. In order to better adapt to the changes in the quality of the wireless signal, the observation covariance is adaptively modified as follows:

[0048] R0=cR0#(4)

[0049] Among them, c is the correction coefficient, which is used to correct the original observation covariance. Its value is related to the number of non-line-of-sight signals in the wireless signal. The more non-line-of-sight signals there are, the larger its value is.

Claims

1. A multi-sensor collaborative positioning method for non-line-of-sight scenarios, characterized in that: The integration of wireless sensor networks and inertial navigation data to ensure reliable positioning in non-line-of-sight scenarios includes: Step 1: Non-line-of-sight error suppression Using the existing non-line-of-sight signal discrimination method, the wireless signals in the collaborative positioning are divided into line-of-sight signals and non-line-of-sight signals; Sub-step 1: Building a multi-layer perceptron A multilayer perceptron is a neural network that processes complex nonlinear regression problems. It is generally composed of input, intermediate layers, and output. In the actual collaborative positioning process, the changes in each wireless signal are caused by the movement of the land unmanned vehicle, and the movement state of the land unmanned vehicle is obtained through inertial navigation. A multilayer perceptron is constructed for the wireless signal of each base station, and its input is the speed of the inertial navigation solution. and The subscript k represents the time, the superscripts x and y represent the coordinate axes, and the output is the change in the distance between the land unmanned vehicle and each base station. Among them, the subscript k represents the time, the superscript i represents the base station number, and the middle layer of the multi-layer perceptron is set to two layers to handle nonlinear relationships; Sub-step 2: Training or prediction based on signal category If the signal is determined to be a line-of-sight signal, the multi-layer perceptron executes the training mode, takes the speed calculated by the inertial navigation as input, and the change calculated by the line-of-sight signal as the true value, and trains the multi-layer perceptron to adjust the parameters; if it is determined to be a non-line-of-sight signal, the multi-layer perceptron enters the prediction mode, takes the speed as input, and outputs the signal change Then predict the actual distance information Step 2: Calculate position information When all received wireless signals are determined to be line-of-sight signals, the least squares method is used to solve the position; when there are non-line-of-sight signals, the weighted least squares method is used to solve the position, record: in: In the above formula, is the value of the i-th wireless signal restored by the multi-layer perceptron at time k, is the actual measured signal value, then the weight matrix W is: W=R -1 #(3) After obtaining the weight matrix, the position information is solved by the weighted least squares method; Step 3: Fusion of corrected wireless observations and inertial navigation The inertial navigation recursion result is used as the state quantity, and the position information solved in step 2 is used as the observation quantity. The Kalman filter equation is established for fusion calculation to obtain the final position information. The observation quantity covariance in the Kalman filter is recorded as R0. In order to better adapt to the changes in wireless signal quality, the observation covariance is adaptively modified as follows: R0=cR0#(4) Among them, c is the correction coefficient, which is used to correct the original observation covariance. Its value is related to the number of non-line-of-sight signals in the wireless signal. The more the number of non-line-of-sight signals, the larger its value.

Images