Intelligent vehicle wireless positioning non-line-of-sight signal discrimination method based on multi-model fusion

Abstract

The invention discloses an intelligent vehicle wireless positioning non-line-of-sight signal discrimination method based on multi-model fusion. Firstly, related sensor signals are collected, secondly, a wireless signal non-line-of-sight discrimination learning model based on a support vector machine is constructed, then, a wireless signal non-line-of-sight discrimination mathematical model based on multi-sensor consistency is constructed, and finally, a wireless signal non-line-of-sight discrimination fusion model based on a D-S evidence theory is constructed. And fusing the output results of the two models, and finally carrying out non-line-of-sight condition judgment on the current signal. According to the non-line-of-sight signal discrimination method disclosed by the invention, the robustness and the adaptive capacity of the wireless signal non-line-of-sight discrimination method in a complex environment are improved.

Description

Technical Field

[0001] The invention relates to the field of navigation and positioning, and in particular to a method for distinguishing non-line-of-sight signals in wireless positioning. Background Art

[0002] As a product of the combination of traditional vehicles and cutting-edge technologies such as artificial intelligence, sensors, and wireless communications, smart cars have more reliable intelligent functions and applications, and have gradually achieved preliminary applications in the fields of smart transportation and dangerous operations, promoting the development and progress of related fields. It is worth noting that the intelligent functions of smart cars such as planning and decision-making, as well as intelligent applications such as active safety, all require accurate and reliable location information as a basis. In general, the Global Navigation Satellite System (GNSS) can already provide smart cars with accurate and reliable location information in most scenarios. However, when the satellite signal is blocked, its positioning performance will be severely reduced, especially in environments where the satellite signal is completely blocked, such as tunnels, the GNSS will be completely ineffective.

[0003] At present, with the further development and application of wireless communication technology, wireless positioning technology based on wireless sensor network (WSN) has gradually matured and can be used as an effective supplement to provide location information for smart cars when satellites fail. Wireless positioning usually includes a transmitting station and a receiving station. The distance or angle between the two is obtained by transmitting and receiving wireless signals, and then the location information is solved by methods such as least squares. Therefore, this requires that the transmitting station and the receiving station are in direct line of sight during the wireless positioning process in order to obtain accurate distance or angle information, which is then used for wireless positioning. However, in actual application environments, there are often obstacles between the transmitting station and the receiving station, resulting in non-line-of-sight propagation, which leads to inaccurate or even failure of wireless positioning. Therefore, it is crucial to accurately identify non-line-of-sight signals for wireless positioning.

[0004] Existing wireless signal non-line-of-sight discrimination methods mainly include the following categories: (1) Methods based on mathematical statistics. This type of method mainly considers that the mathematical statistical results of line-of-sight signals and non-line-of-sight signals are different, such as mathematical characteristics such as mean and variance. Therefore, the relevant mathematical statistical information of the signal is evaluated, and when the change of the statistical information exceeds a certain threshold, it is defined as a non-line-of-sight signal. However, this type of method is more dependent on the prior information conditions of the signal, and the threshold varies slightly in different environments, which is difficult to determine; (2) Learning-based methods. This type of method mainly establishes a machine learning or deep learning model in advance, and then inputs the data such as the distance, strength, channel impulse response of the wireless signal into the established learning network, so as to classify the line-of-sight signal and the non-line-of-sight signal. However, this type of method requires a large amount of data for pre-training, has poor interpretability, and is not very adaptable to the environment.

[0005] In general, the existing wireless signal non-line-of-sight identification methods still have problems such as low robustness and poor environmental adaptability, mainly because: (1) The information source is single. Since the existing methods are mostly based on one or several input features of wireless signals (such as the distance, strength, channel impulse response, etc. of the wireless signal), such features can only reflect the absolute changes of the wireless signal. However, in some cases, the sudden change of the signal is not necessarily caused by non-line-of-sight. For example, the speed of the smart car suddenly increases, causing the distance measurement result of the wireless signal to suddenly increase. This is a normal line-of-sight phenomenon, so it is easy to cause misjudgment; (2) The recognition model is single. The existing methods are mostly based on a single mathematical model or learning model for identification. Since a single model cannot be applied to all situations and has poor robustness, it will always cause misjudgment or missed judgment in complex environments, which can easily lead to inaccurate non-line-of-sight signal identification results. Summary of the invention

[0006] The present invention proposes a non-line-of-sight signal discrimination method for wireless positioning of intelligent vehicles based on multi-model fusion, and the specific features are as follows: (1) Rich feature sources and strong environmental adaptability. The non-line-of-sight signal discrimination method proposed in the present invention, in addition to using the distance measurement of the wireless signal and the corresponding derivation results as input features, also introduces the inertial navigation information of the intelligent vehicle as input, which has low requirements for wireless devices and no requirements for environmental prior information. The input information source is richer and the environmental adaptability is stronger; (2) The judgment model is redundant and the recognition robustness is high. The present invention simultaneously constructs a machine learning model and a mathematical model to make a preliminary judgment on the non-line-of-sight signal, and then uses the DS evidence theory at the decision-making level for effective fusion to finally judge the non-line-of-sight signal, overcoming the problem that a single model is prone to failure and low robustness in a complex environment.

[0007] The present invention firstly collects the sensor signals involved, then constructs a wireless signal non-line-of-sight discrimination learning model based on support vector machine, then constructs a wireless signal non-line-of-sight discrimination mathematical model based on multi-sensor consistency, and finally constructs a wireless signal non-line-of-sight discrimination fusion model based on DS evidence theory to perform non-line-of-sight judgment.

[0008] The idea of ​​the present invention is further described below:

[0009] Step 1: Collect wireless sensor signals and smart car inertial navigation signals

[0010] First, the sensor signals involved are collected. The sensors involved in the present invention include wireless sensors and inertial navigation on the smart car. The wireless sensor mainly collects the distance information between the transmitting station and the receiving station measured by it, and the inertial navigation mainly collects information such as the acceleration and angular velocity of the smart car. The collection process is carried out under line-of-sight conditions, that is, there is no obstruction between the receiving station and the transmitting station, so as to ensure that the obtained wireless signals are all line-of-sight signals.

[0011] Step 2: Build a wireless signal non-line-of-sight discrimination learning model based on support vector machine

[0012] Sub-step 1: Create model input

[0013] First, the present invention constructs a wireless signal non-line-of-sight discrimination learning model based on support vector machine. Different from the existing method that only uses the characteristics of wireless sensor signals as input, the input of the support vector machine model in the present invention uses both wireless sensor signals and inertial navigation signals. The input of the model is mainly composed of the change rate v of the wireless signal distance measurement value at the current k moment wk And the speed value v of the smart car obtained by inertial navigation integration vk Composition, denoted as (v wk ,v vk ). The rate of change of the wireless signal distance value is calculated by the following formula:

[0014]

[0015] where ρ k and ρ k-1 Represent the distance measurement values ​​obtained by the wireless sensor at time k and the previous adjacent time, respectively, and Δt represents the time interval between the adjacent times. The method of using inertial navigation integration to obtain the speed information of the smart car can be found in the reference (Jia Xiaoxue, Zhao Dongqing, Zhang Letian, et al. Visual SLAM algorithm based on adaptive inertial navigation assisted feature matching [J]. Optics and Precision Engineering, 2023, 31(05): 621-630).

[0016] Sub-step 2: Construct training and test sets

[0017] Select some of the data collected in the above steps and add random noise with a mean value of 1 to 2 meters to it, so as to convert the line-of-sight signal into a non-line-of-sight signal. On this basis, construct a training set and a test set. At the same time, set the data label of the signal with added noise to non-line-of-sight, and set the data label of the original signal without added noise to line-of-sight.

[0018] Sub-step 3: Model training

[0019] The radial basis function (RadialBasisFunction) is selected as the kernel function of the support vector machine model in the present invention, and its specific content can be found in the reference (Zhou Zhihua. Machine Learning [M]. Tsinghua University Press, 2016). The training set is input into the model for training, and the model parameters are adjusted using the cross-validation method to optimize the model performance.

[0020] Sub-step 4: Model prediction

[0021] After completing the model training of the support vector machine, it can be used to predict subsequent input data, output the non-line-of-sight situation of the current signal, and the probability p1 of being a non-line-of-sight signal, thereby obtaining the probability q1=1-p1 that the current signal is a line-of-sight signal.

[0022] Step 3: Construct a mathematical model for wireless signal non-line-of-sight discrimination based on multi-sensor consistency

[0023] Secondly, the present invention constructs a mathematical model for non-line-of-sight discrimination of wireless signals based on multi-sensor consistency. The multi-sensor consistency in the present invention is mainly constructed through the mathematical relationship between the three sides of the triangle. The distance measurement value ρ of the wireless signal at the current time k k , the distance measurement value ρ of the wireless signal at the previous moment k-1 And the distance Δd traveled by the smart car from the last moment to the current moment can form the three sides of the triangle. Therefore, according to the mathematical relationship:

[0024] |ρ k -ρ k-1 |<Δd#(8)

[0025] Since the driving distance Δd of the smart car is obtained by integrating the inertial navigation in a very short time, it can be considered that its accuracy is relatively high. Therefore, the measurement distance error σ of the wireless signal is mainly considered. WSN Since non-line-of-sight propagation brings positive errors, it will lead to ρ k The result is greater than the actual result, so if:

[0026] |ρ k -ρ k-1 |-Δd>σ WSN #(9)

[0027] At the same time, according to the following formula, the probability p2 of the wireless signal belonging to the non-line-of-sight signal is determined, and then the probability q2=1-p2 of the current signal belonging to the line-of-sight signal is obtained.

[0028]

[0029] Among them, the value of probability p2 is adjusted according to engineering experience and test results.

[0030] Step 4: Construct a wireless signal non-line-of-sight discrimination fusion model based on DS evidence theory

[0031] According to the non-line-of-sight discrimination results and corresponding probabilities obtained in step two and step three, the DS evidence theory is constructed for decision-level fusion, and finally the non-line-of-sight discrimination result of the present invention is obtained. It is recorded that the current signal is judged as a non-line-of-sight signal as event N, and the line-of-sight signal is judged as event L. M1 is the probability distribution function obtained based on the support vector machine, that is, M1(N)=p1, M1(L)=q1, M2 is the probability distribution function obtained by multi-sensor consistency, that is, M2(N)=p2, M2(L)=q2, and M is the orthogonal sum of M1 and M2. Calculate the normalization constant K, the formula is:

[0032]

[0033] Then calculate the optimal probability R of non-line-of-sight signal:

[0034]

[0035] The non-line-of-sight condition of the current signal is determined according to the probability R predicted by the DS evidence theory. In the present invention, it is defined that when R>0.8, the signal is determined to be a non-line-of-sight signal.

[0036] The beneficial effect of the present invention is that the non-line-of-sight signal discrimination method for wireless positioning of intelligent vehicles based on multi-model fusion proposed in the present invention effectively overcomes the problems of poor robustness and insufficient environmental adaptability of existing non-line-of-sight signal discrimination methods, and is conducive to improving the accuracy of non-line-of-sight discrimination of wireless signals in complex environments, thereby ensuring the accuracy and reliability of wireless positioning, thereby promoting the development of wireless positioning technology and its application in related fields such as intelligent vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of the method of the present invention.

[0038] Figure 2 It is a schematic diagram of the mathematical model of wireless signal non-line-of-sight discrimination based on multi-sensor consistency. DETAILED DESCRIPTION

[0039] As a product of the combination of traditional vehicles and cutting-edge technologies such as artificial intelligence, sensors, and wireless communications, smart cars have more reliable intelligent functions and applications, and have gradually achieved preliminary applications in the fields of smart transportation and dangerous operations, promoting the development and progress of related fields. It is worth noting that the intelligent functions of smart cars such as planning and decision-making, as well as intelligent applications such as active safety, all require accurate and reliable location information as a basis. In general, the Global Navigation Satellite System (GNSS) can already provide smart cars with accurate and reliable location information in most scenarios. However, when the satellite signal is blocked, its positioning performance will be severely reduced, especially in environments where the satellite signal is completely blocked, such as tunnels, the GNSS will be completely ineffective.

[0040] At present, with the further development and application of wireless communication technology, wireless positioning technology based on wireless sensor network (WSN) has gradually matured and can be used as an effective supplement to provide location information for smart cars when satellites fail. Wireless positioning usually includes a transmitting station and a receiving station. The distance or angle between the two is obtained by transmitting and receiving wireless signals, and then the location information is solved by methods such as least squares. Therefore, this requires that the transmitting station and the receiving station are in direct line of sight during the wireless positioning process in order to obtain accurate distance or angle information, which is then used for wireless positioning. However, in actual application environments, there are often obstacles between the transmitting station and the receiving station, resulting in non-line-of-sight propagation, which leads to inaccurate or even failure of wireless positioning. Therefore, it is crucial to accurately identify non-line-of-sight signals for wireless positioning.

[0041] Existing wireless signal non-line-of-sight discrimination methods mainly include the following categories: (1) Methods based on mathematical statistics. This type of method mainly considers that the mathematical statistical results of line-of-sight signals and non-line-of-sight signals are different, such as mathematical characteristics such as mean and variance. Therefore, the relevant mathematical statistical information of the signal is evaluated, and when the change of the statistical information exceeds a certain threshold, it is defined as a non-line-of-sight signal. However, this type of method is more dependent on the prior information conditions of the signal, and the threshold varies slightly in different environments, which is difficult to determine; (2) Learning-based methods. This type of method mainly establishes a machine learning or deep learning model in advance, and then inputs the data such as the distance, strength, channel impulse response of the wireless signal into the established learning network, so as to classify the line-of-sight signal and the non-line-of-sight signal. However, this type of method requires a large amount of data for pre-training, has poor interpretability, and is not very adaptable to the environment.

[0042] In general, the existing wireless signal non-line-of-sight identification methods still have problems such as low robustness and poor environmental adaptability, mainly because: (1) The information source is single. Since the existing methods are mostly based on one or several input features of wireless signals (such as the distance, strength, channel impulse response, etc. of the wireless signal), such features can only reflect the absolute changes of the wireless signal. However, in some cases, the sudden change of the signal is not necessarily caused by non-line-of-sight. For example, the speed of the smart car suddenly increases, causing the distance measurement result of the wireless signal to suddenly increase. This is a normal line-of-sight phenomenon, so it is easy to cause misjudgment; (2) The recognition model is single. The existing methods are mostly based on a single mathematical model or learning model for identification. Since a single model cannot be applied to all situations and has poor robustness, it will always cause misjudgment or missed judgment in complex environments, which can easily lead to inaccurate non-line-of-sight signal identification results.

[0043] In view of the above problems, the present invention proposes a non-line-of-sight signal discrimination method for wireless positioning of intelligent vehicles based on multi-model fusion, and the specific features are as follows: (1) Rich feature sources and strong environmental adaptability. The non-line-of-sight signal discrimination method proposed in the present invention, in addition to using the distance measurement of the wireless signal and the corresponding derivation results as input features, also introduces the inertial navigation information of the intelligent vehicle as input, which has low requirements for wireless devices and no requirements for environmental prior information. The input information source is richer and the environmental adaptability is stronger; (2) The judgment model is redundant and the recognition robustness is high. The present invention simultaneously constructs a machine learning model and a mathematical model to make a preliminary judgment on the non-line-of-sight signal, and then uses the DS evidence theory at the decision-making level for effective fusion to finally judge the non-line-of-sight signal, overcoming the problem that a single model is prone to failure and low robustness in a complex environment.

[0044] The non-line-of-sight signal discrimination method for wireless positioning of smart cars based on multi-model fusion proposed in the present invention effectively overcomes the problems of poor robustness and insufficient environmental adaptability of existing non-line-of-sight signal discrimination methods, and is conducive to improving the accuracy of non-line-of-sight discrimination of wireless signals in complex environments, thereby ensuring the accuracy and reliability of wireless positioning, thereby promoting the development of wireless positioning technology and its application in related fields such as smart cars.

[0045] The idea of ​​the present invention is further described below:

[0046] Step 1: Collect wireless sensor signals and smart car inertial navigation signals

[0047] First, the sensor signals involved are collected. The sensors involved in the present invention include wireless sensors and inertial navigation on the smart car. The wireless sensor mainly collects the distance information between the transmitting station and the receiving station measured by it, and the inertial navigation mainly collects information such as the acceleration and angular velocity of the smart car. The collection process is carried out under line-of-sight conditions, that is, there is no obstruction between the receiving station and the transmitting station, so as to ensure that the obtained wireless signals are all line-of-sight signals.

[0048] Step 2: Build a wireless signal non-line-of-sight discrimination learning model based on support vector machine

[0049] Sub-step 1: Create model input

[0050] First, the present invention constructs a wireless signal non-line-of-sight discrimination learning model based on support vector machine. Different from the existing method that only uses the characteristics of wireless sensor signals as input, the input of the support vector machine model in the present invention uses both wireless sensor signals and inertial navigation signals. The input of the model is mainly composed of the change rate v of the wireless signal distance measurement value at the current k moment wk And the speed value v of the smart car obtained by inertial navigation integration vk Composition, denoted as (v wk ,v vk ). The rate of change of the wireless signal distance value is calculated by the following formula:

[0051]

[0052] where ρ k and ρ k-1 Represent the distance measurement values ​​obtained by the wireless sensor at time k and the previous adjacent time, respectively, and Δt represents the time interval between the adjacent times. The method of using inertial navigation integration to obtain the speed information of the smart car can be found in the reference (Jia Xiaoxue, Zhao Dongqing, Zhang Letian, et al. Visual SLAM algorithm based on adaptive inertial navigation assisted feature matching [J]. Optics and Precision Engineering, 2023, 31(05): 621-630).

[0053] Sub-step 2: Construct training and test sets

[0054] Select some of the data collected in the above steps and add random noise with a mean value of 1 to 2 meters to it, so as to convert the line-of-sight signal into a non-line-of-sight signal. On this basis, construct a training set and a test set. At the same time, set the data label of the signal with added noise to non-line-of-sight, and set the data label of the original signal without added noise to line-of-sight.

[0055] Sub-step 3: Model training

[0056] The radial basis function (RadialBasisFunction) is selected as the kernel function of the support vector machine model in the present invention, and its specific content can be found in the reference (Zhou Zhihua. Machine Learning [M]. Tsinghua University Press, 2016). The training set is input into the model for training, and the model parameters are adjusted using the cross-validation method to optimize the model performance.

[0057] Sub-step 4: Model prediction

[0058] After completing the model training of the support vector machine, it can be used to predict subsequent input data, output the non-line-of-sight situation of the current signal, and the probability p1 of being a non-line-of-sight signal, thereby obtaining the probability q1=1-p1 that the current signal is a line-of-sight signal.

[0059] Step 3: Construct a mathematical model for wireless signal non-line-of-sight discrimination based on multi-sensor consistency

[0060] Secondly, the present invention constructs a mathematical model for non-line-of-sight discrimination of wireless signals based on multi-sensor consistency. The multi-sensor consistency in the present invention is mainly constructed through the mathematical relationship between the three sides of the triangle. The distance measurement value ρ of the wireless signal at the current time k k , the distance measurement value ρ of the wireless signal at the previous moment k-1 And the distance Δd traveled by the smart car from the last moment to the current moment can form the three sides of the triangle. Therefore, according to the mathematical relationship:

[0061] |ρ k -ρ k-1 |<Δd#(14)

[0062] Since the driving distance Δd of the smart car is obtained by integrating the inertial navigation in a very short time, it can be considered that its accuracy is relatively high. Therefore, the measurement distance error σ of the wireless signal is mainly considered. WSN Since non-line-of-sight propagation brings positive errors, it will lead to ρ k The result is greater than the actual result, so if:

[0063] |ρ k -ρ k-1 |-Δd>σ WSN #(15)

[0064] At the same time, according to the following formula, the probability p2 of the wireless signal belonging to the non-line-of-sight signal is determined, and then the probability q2=1-p2 of the current signal belonging to the line-of-sight signal is obtained.

[0065]

[0066] Among them, the value of probability p2 is adjusted according to engineering experience and test results.

[0067] Step 4: Construct a wireless signal non-line-of-sight discrimination fusion model based on DS evidence theory

[0068] According to the non-line-of-sight discrimination results and corresponding probabilities obtained in step two and step three, the DS evidence theory is constructed for decision-level fusion, and finally the non-line-of-sight discrimination result of the present invention is obtained. It is recorded that the current signal is judged as a non-line-of-sight signal as event N, and the line-of-sight signal is judged as event L. M1 is the probability distribution function obtained based on the support vector machine, that is, M1(N)=p1, M1(L)=q1, M2 is the probability distribution function obtained by multi-sensor consistency, that is, M2(N)=p2, M2(L)=q2, and M is the orthogonal sum of M1 and M2. Calculate the normalization constant K, the formula is:

[0069]

[0070] Then calculate the optimal probability R of non-line-of-sight signal:

[0071]

[0072] The non-line-of-sight condition of the current signal is determined according to the probability R predicted by the DS evidence theory. In the present invention, it is defined that when R>0.8, the signal is determined to be a non-line-of-sight signal.

Claims

1. A method for distinguishing non-line-of-sight signals for wireless positioning of intelligent vehicles based on multi-model fusion, characterized in that: First, the sensor signals involved are collected. Secondly, a wireless signal non-line-of-sight discrimination learning model based on support vector machine is constructed. Then, a wireless signal non-line-of-sight discrimination mathematical model based on multi-sensor consistency is constructed. Finally, a wireless signal non-line-of-sight discrimination fusion model based on DS evidence theory is constructed. The specific steps include: Step 1: Collect wireless sensor signals and smart car inertial navigation signals First, the sensor signals involved are collected. The sensors involved include wireless sensors and inertial navigation on the smart car. The wireless sensor is used to collect the distance information between the transmitting station and the receiving station measured by it, and the inertial navigation is used to collect the acceleration and angular velocity information of the smart car. The collection process is carried out under line-of-sight conditions, that is, there is no obstruction between the receiving station and the transmitting station, so as to ensure that the wireless signals obtained are all line-of-sight signals. Step 2: Build a wireless signal non-line-of-sight discrimination learning model based on support vector machine Sub-step 1: Create model input First, a wireless signal non-line-of-sight discrimination learning model based on support vector machine is constructed, whose input is the change rate v of the wireless signal distance measurement value at the current k moment wk And the speed value v of the smart car obtained by inertial navigation integration vk Composition, denoted as (v wk ,v vk ), the rate of change of the wireless signal distance value is calculated by the following formula: where ρ k and ρ k-1 They represent the distance measurement values ​​obtained by the wireless sensor at time k and the previous adjacent time, Δt represents the time interval between the adjacent times. Sub-step 2: Construct training and test sets Select part of the data collected in the above steps, add random noise with a mean value of 1 to 2 meters, so as to change the line-of-sight signal into a non-line-of-sight signal. On this basis, construct a training set and a test set. At the same time, set the data label of the signal with added noise to non-line-of-sight, and set the data label of the original signal without added noise to line-of-sight. Sub-step 3: Model training The radial basis function is selected as the kernel function of the support vector machine model, the training set is input into the model for training, and the model parameters are adjusted using the cross-validation method to optimize the model performance. Sub-step 4: Model prediction After the support vector machine model training is completed, it is used to predict the subsequent input data, output the non-line-of-sight situation of the current signal, and the probability p1 of the non-line-of-sight signal, thereby obtaining the probability q1=1-p1 that the current signal belongs to the line-of-sight signal. Step 3: Construct a mathematical model for wireless signal non-line-of-sight discrimination based on multi-sensor consistency Secondly, a mathematical model for non-line-of-sight discrimination of wireless signals based on multi-sensor consistency is constructed. The multi-sensor consistency is constructed through the mathematical relationship between the three sides of the triangle. The distance measurement value ρ of the wireless signal at the current k moment k , the distance measurement value ρ of the wireless signal at the previous moment k-1 And the driving distance Δd of the smart car from the previous moment to the current moment, constitute the three sides of the triangle. Therefore, according to the mathematical relationship: r k -r k-1 <Δd#(2) Since the driving distance Δd of the smart car is obtained by integrating the inertial navigation in a very short time, it is believed that its accuracy is high, so the measurement distance error σ of the wireless signal is considered WSN , since non-line-of-sight propagation brings positive errors, it will lead to ρ k The result is greater than the actual result, so if: r k -r k-1 -Δd>σ WSN #(3) The wireless signal at the current moment is considered to be a non-line-of-sight signal, otherwise it is determined to be a line-of-sight signal. At the same time, according to the following formula, the probability p2 of the non-line-of-sight signal is determined, and then the probability q2=1-p2 of the current signal being a line-of-sight signal is obtained. The value of probability p2 is adjusted according to engineering experience and test results. Step 4: Construct a wireless signal non-line-of-sight discrimination fusion model based on DS evidence theory According to the non-line-of-sight discrimination results and corresponding probabilities obtained in step 2 and step 3, the DS evidence theory is constructed for decision-level fusion, and finally the non-line-of-sight discrimination result is obtained. It is recorded that the current signal is judged as a non-line-of-sight signal as event N, and the line-of-sight signal is judged as event L. M1 is the probability distribution function obtained based on the support vector machine, that is, M1(N)=p1, M1(L)=q1, M2 is the probability distribution function obtained by multi-sensor consistency, that is, M2(N)=p2, M2(L=q2, M is the orthogonal sum of M1 and M2, and the normalization constant K is calculated. The formula is: Then calculate the optimal probability R of non-line-of-sight signal: The non-line-of-sight situation of the current signal is determined according to the probability R predicted by the DS evidence theory. It is defined that when R>0.8, the signal is determined to be a non-line-of-sight signal.

Images