A Multi-Source Data Selective Fusion Method for Multi-Environment Localization of Unmanned Vehicles

Abstract

The present invention discloses a multi-source data selective fusion method for multi-environment positioning of an unmanned vehicle. Specifically, the method includes the following steps: S1: Information control, S2: Control evaluation, and S3: Evaluation feedback. Compared with the prior art, by designing a brand-new pre-data fusion algorithm, this method collaborates with information control and control evaluation to control and adjust multi-source data at the input end, so as to directionally screen and filter sensor data information with higher credibility for the unmanned vehicle in various different environments, reduce the computational complexity and data redundancy in the data fusion process, and control the computational cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of sensors, and particularly relates to a method for pre-data processing of multi-source sensor fusion. Background Art

[0002] An unmanned vehicle, also known as an autonomous vehicle, mainly relies on sensors to perceive the surrounding environment when applied to a specific driving environment. A single sensor can only provide surrounding environment data of a single form. Therefore, in a specific unmanned vehicle application scenario, a vehicle is often equipped with multiple types and quantities of sensor clusters. Different types of sensors cooperate and work together to provide multi-source environment data for the unmanned vehicle, providing a decision-making basis for tasks such as real-time positioning, pose adjustment, and scenario prediction of the unmanned vehicle.

[0003] Sensors applied to specific unmanned vehicles often include lidar, vision sensors, locators, inertial navigation sensors, etc. The setting method of multiple sensors of different types working together has good versatility and robustness: First, since each sensor detects different target physical quantities, different sensors will cooperate to provide comprehensive and multi-dimensional environmental perception data for the vehicle. Different sensors often complement each other in terms of detection objects, detection accuracy, and detection range. By leveraging the complementary advantages of multiple sensors, more comprehensive environmental perception data can be obtained for the vehicle. Second, since different sensors are often set separately and are not related to each other, the failure conditions and failure modes of different sensors are also to some extent not related to each other. Therefore, during the vehicle's driving process, no matter how the surrounding environment changes, it can always be ensured that at least one sensor maintains normal operation, which will help the vehicle significantly improve its robustness and well adapt to changes in the surrounding environment.

[0004] When using multi-source sensors to help the vehicle obtain multi-source data of the surrounding environment, the vehicle needs to perform data fusion processing on the multi-source data collected by multiple sensors. The essence of data fusion processing is to automatically analyze and synthesize multi-source information and data from multiple sensors using certain technical means, perform computational solutions such as filtering and denoising, backend optimization, etc. on the input multi-source sensor information and data, and give a negative feedback effect to each original data information to complete error correction for the information and data detected by the multi-source sensors. Ultimately, it helps the vehicle achieve scenario prediction, unmanned vehicle position correction, etc.

[0005] When a driverless vehicle travels in an environment, environmental factors will inevitably affect different sensors. Therefore, during the data fusion process, it is necessary to focus on the sensor detection errors caused by environmental factors. In the prior art, error data is often screened out through multiple calculations, and the error data is excluded or replaced by a predetermined algorithm and then transmitted to the next link. This means that the amount of calculation will increase exponentially, which not only requires larger-scale and higher-cost computing resources, but also greatly reduces the computing efficiency of the vehicle, prolongs the response time of the vehicle, and is not conducive to the progress of driverless vehicle technology. Summary of the Invention

[0006] To solve the above problems, the purpose of the present invention is to provide a method that uses multi-source data composed of different sensor data as input. After an information control process, it realizes the screening and filtering of sensor data in multiple different environments, eliminates sensor data that may have large errors in each specific working scenario, and reduces the data volume and computational complexity in the subsequent data fusion process.

[0007] To achieve the above purpose, the technical solution of the present invention is as follows:

[0008] A multi-source data selective fusion method for multi-environment positioning of driverless vehicles, and the method is as follows:

[0009] S1: Information control: Classify the environment, assign a credibility weight to each type of sensor in different environments, and for a specified sensor in a specified working environment, selectively fuse its output data according to its credibility weight;

[0010] S2: Control evaluation: Analyze and evaluate the information control process, and evaluate its reliability, risk rate, and usability;

[0011] S3: Evaluation feedback: Analyze the result of information control according to the control evaluation result, timely adjust the control strategy, and continuously select more reliable sensor data to be sent to the subsequent data fusion link for analysis and fusion;

[0012] S4: Data fusion positioning and positioning evaluation: Fuse multi-source data, perform synchronous positioning on the driverless vehicle, and analyze and evaluate the current positioning of the driverless vehicle.

[0013] Further, S1 is specifically as follows:

[0014] S11: Arrange sensors: Arrange n types of sensors on the vehicle, where n≥2 and n is an integer; denote the specified sensor as s i , gather all the sensors on the vehicle, and obtain the total sensor set as S: S = {s1, s2... s n}, and there is:

[0015] S12: Specify the basic scenarios: Denote the situation where only sensor s1 fails and other sensors s2, s3,..., s n are all working properly as the first basic scenario c1; Denote the situation where only sensor s2 fails and other sensors s1, s3,..., s n are all working properly as the second basic scenario c2; And so on, denote the situation where only sensor s i fails and other sensors s1, s2,..., s i-1 , s i+1 ,..., s n are all working properly as the i-th basic scenario c i ; And so on, denote the situation where only sensor s n fails and other sensors s1, s2,..., s n-1 are all working properly as the n-th basic scenario c n ; Collect all the basic scenarios to obtain the total set of basic scenarios C: C = {c1, c2,..., c n}, and there is Since the basic scenarios specify the scenarios where only one sensor fails and other sensors are all working properly, it is easy to calculate that when there are n types of sensors arranged on the vehicle, there should be types of basic scenarios, that is, c1 to c n .

[0016] S13: Define the acquisition degree: Define the acquisition degree D to characterize the acquisition situation of the output data of the corresponding sensor, and it is stipulated that the value of the acquisition degree D is 0 or 1. When the value is 0, the output data of its corresponding sensor is not acquired; when the value is 1, the output data of its corresponding sensor is acquired;

[0017] S14: Calculate the acquisition degree of each sensor in each basic scenario: According to the sensor failure situation in each basic scenario, for the sensors that fail in this basic scenario, do not acquire their sensor data, and the value of their acquisition degree is 0. For the sensors that are working properly, acquire their sensor data, and the value of their acquisition degree is 1;

[0018] Denote the acquisition degree of sensor s i under the i-th basic scenario c j as Collect the acquisition degrees of each sensor under each basic scenario to obtain the total set of basic scenario acquisition degrees D C : There is Therefore, it can be known that among the total set of sensors S, the total set of basic scenarios C, and the total set of basic scenario acquisition degrees D C , their relationship should be as follows in the table:

[0019]

[0020] Furthermore, S1 also includes:

[0021] S15: Divide the working environment: When it is determined that all sensors are working properly, the vehicle is in a normal working environment, denoted as E0;

[0022] According to the total set of basic scenarios C, the abnormal environments are divided into m types, namely abnormal environments E1, E2... E m should be one or more combinations of the above first basic scenario C1, second basic scenario C2... and the nth basic scenario C n ; where and m is an integer;

[0023] Denote the specified working environment as E x ; The total set of working environments E is obtained by aggregating all working environments: E = {E0, E1, E2... E m}}, and and E x ≠ E0;

[0024] The abnormal environment refers to the situation where the driving environment of the vehicle interferes with the sensors arranged on the vehicle, resulting in abnormal output data of the corresponding sensors or the failure of the entire sensor. Therefore, it can be known that considering the influence of external factors on the sensors in the abnormal environment, starting from the basic scenarios, the performance of the sensors in different abnormal environments can be described by a combination of one or more basic scenarios. When the external factors only affect a single sensor, the corresponding affected sensor can be used to find a matching basic scenario to describe the abnormal environment; when the external factors affect multiple sensors simultaneously, after finding the matching basic scenarios for the corresponding affected sensors respectively, the specific acquisition degrees of each sensor in this abnormal environment can be obtained by taking the logical AND of the acquisition degrees of the corresponding sensors in multiple basic scenarios. And precisely because the abnormal environment is described by a combination of one or more basic scenarios, it is easy to infer that the number m of abnormal environments should be at least 1 and at most Among them is a consideration made from the aspect of ensuring that at least two sensors work properly and ensuring that the subsequent data processing system can fuse at least two types of data. When the driving environment of the vehicle is too harsh and has a huge interference on the sensors arranged on the vehicle, resulting in only one sensor being effective or all sensors failing, it should be determined that the external environment at this time is not suitable for the driverless scenario and is not suitable for the vehicle to drive, and the vehicle should be forced to stop until the external environment improves before it can travel again.

[0025] S16: Calculate the acquisition degrees of each sensor in each working environment: According to the sensor failure situations in each basic scenario, for the sensors that fail in this basic scenario, do not collect the data of this sensor, and its acquisition degree takes the value of 0; for the sensors that are working properly, collect the data of this sensor, and its acquisition degree takes the value of 1;

[0026] Denote the working environment as E x Under this condition, the sensor s y has a collection degree of Set D as the collection degree set of all sensors under all working environments E : There is

[0027] Therefore, it can be known that the total sensor set is S, the total working environment set is E, and the collection degree set of sensors is D E The relationship is as follows in the table

[0028]

[0029] Furthermore, S1 also includes

[0030] S17: Define the attention degree: Define the attention degree A to represent the credibility of the output data of the corresponding sensor, and it is stipulated that the value of the attention degree A is in the range of [0, 1]. The larger the value of the attention degree A, the higher the trust degree given to the output data of the corresponding sensor; Keep the sum of the attention degree values of all sensors in the same working environment as 1, and assign the smallest attention degree value to the sensors that fail in the current working environment

[0031] S18: Assign the attention degree: Denote the working environment as E x Under this condition, the sensor s y has an attention degree of Set A as the total attention degree set obtained by the attention degrees of each type of sensor under all working environments E : There is Then there is the total sensor set as S, the total working environment set as E, and the total attention degree set of sensors as A E The relationship is as follows in the table

[0032]

[0033] Furthermore, S1 also includes

[0034] S19: Set the fusion decision threshold: Set the fusion decision threshold Q α ;

[0035] S110: Calculate the actual fusion parameter: Define the fusion parameter as Q, denote the working environment as E x Under this condition, the sensor s y has a fusion parameter of It is stipulated that Set Q as the total fusion parameter set obtained by the fusion parameters of each type of sensor under each type of working environment E :

[0036] For different sensors in different environments, different external factors will affect the corresponding sensors to varying degrees. Based on the sensors that can operate normally in the working scenario, if the acquisition degree of the sensors is maintained at 1, it means that in this working scenario, the external factors have no impact on them and the output data of the sensors can be collected normally. Conversely, it means that in this working scenario, the external factors interfere with the corresponding sensors, resulting in large errors in the output data of the sensors. At this time, the acquisition of the output data of this sensor is stopped, and the data with large errors is prohibited from entering the system to avoid spending a large amount of computing power and time on correcting the sensor data with large errors during subsequent data processing.

[0037] In different working environments, by further assigning attention levels to different sensors, the importance of data can be set for multiple sensors working together in each working environment. Higher attention levels are assigned to the sensor data that is more reliable in the working scenario, and the lowest attention levels are assigned to the sensors that have failed in the working scenario, so as to distinguish the status of different sensors in making the next autonomous driving decision for the vehicle in the specific driving environment, and make more full use of the front-end sensor data to provide more reliable detection data for vehicle autonomous driving.

[0038] S111: Fusion determination: Set fusion conditions And Verify each element in the total set of fusion parameters Q E one by one;

[0039] S112: Selective fusion: If the current element meets the fusion conditions set in S111, the current sensor in the current working environment represented by this element is allowed to output sensor data to the subsequent data fusion link; otherwise, the output data of the current sensor in the current working environment represented by this element is discarded. For a single sensor s y , in the working environment E x , the fusion parameter calculated by will simultaneously express information such as the working environment where the sensor is located, the impact of its working environment on the sensor, and the importance of the output data of the sensor in this working environment. Therefore one item will evaluate the reliability of the current sensor s for its own data; and for the current working environment E y , x for will express the degree of influence of this working scenario on all the sensors set on the vehicle, that is, for the current working environment E x , if there exists It means that in this working environment, some sensors may fail due to external factors, but the remaining sensors that can work normally can provide sufficient and reliable real-time detection data for the vehicle, and the vehicle can travel in this working environment E. x Therefore, it can be known that by setting the fusion conditions and verifying each element in the total set of fusion parameters Q E one by one; the sensor data in different working environments can be comprehensively evaluated from the aspects of a single sensor and the entire working scenario, and sufficiently reliable data can be selected to enter the subsequent data fusion link, ensuring that the sensor data determined by this fusion is reliable and has few errors, saving computing power for the subsequent sensor data fusion process and improving the fusion efficiency of the subsequent data fusion process.

[0040] Furthermore, S2 is specifically as follows:

[0041] S21: Record that there are k interruptions in the entire information control process; make a reliability assessment of the entire information control process and evaluate the normal working continuity of the entire information control process;

[0042] S22: Make a risk rate assessment of the entire information control process and calculate the normal working duration of the vehicle before a certain interruption occurs during the information control process;

[0043] S23: Make an availability assessment of the entire information control process and calculate the effective availability of the data and the time ratio of the effective data during the entire information control process. For the information control process described above, evaluating it from three aspects of reliability, risk rate, and availability can fully evaluate the work of its information control process, and the evaluation results are transmitted to S3, which continuously adjusts according to the evaluation results.

[0044] Furthermore, S21 is specifically as follows:

[0045] Define the reliability of the information control process within the time period of Δt as K(Δt): K(Δt) = e-λΔt. For the entire information control process, calculate the average reliability value K(Δt) avg : and the root mean square value of reliability K(Δt) rms :

[0046] Furthermore, S22 is specifically as follows:

[0047] Record the running time before the σth interruption occurs in the zth work as Then there is: Calculate the risk rate P of the entire information control process risk as:

[0048] Further, S23 specifically includes:

[0049] S231: Perform multiple samplings in the data stream in two ways: at a fixed time interval ΔT and a random time interval Ran(T). Each sampling is recorded as a data point.

[0050] S232: Denote the amount of information data in a certain data point as ζ, and the type of information data as Set the target amount of information data ζ set , the target number of information data types as

[0051] S234: Classify and count the data points: F1 represents the number of data points that do not meet the data selection requirements, and its expression is: 1 ≤ ζ < ζ set < n; F2 represents the number of data points that do not meet the integrity requirements, and its expression is: the amount of information data in the data point F3 represents the number of data points with interrupted data continuity, and its expression is: ζ = 0, F represents the total number of data points sampled and statistically analyzed within the total time of the whole process;

[0052] S235: Calculate the effective availability of data in the entire information control process:

[0053] S236: Denote the time t when the data in the data stream is effectively available 有效 occupying the entire information control time T 全部 , when ζ = ζ set , , statistically analyze the time t of the effective data 有效 , and calculate the time ratio ω of the effective data: The availability evaluation is a judgment of functional availability. It is carried out when the information control meets the purpose of the scheme, can normally implement the function and play a role, and is also based on the selection requirements, integrity requirements, and continuity requirements. The original multi-source data is selected to form a filtered data stream. Analyzing the availability from the two perspectives of the effective availability of data in the information control process and the time ratio of the effective data can provide result feedback for the information control process, so that the vehicle can adjust the control strategy in a timely manner.

[0054] The advantages of the present invention are as follows: Compared with the prior art, in the present invention, by designing a brand-new pre-data fusion algorithm, with the joint cooperation of information control and control evaluation, the control and adjustment of multi-source data at the input end are realized, so as to directionally screen and filter more reliable sensor data information for the unmanned vehicle in various different environments, reduce the computational complexity and data redundancy in the data fusion process, and control the computational cost. Description of the Drawings

[0055] Figure 1 It is a flowchart of the multi-source data selective fusion method for multi-environment positioning of an autonomous vehicle provided in the specific implementation manner. Specific Implementation Manner

[0056] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] To achieve the above objectives, the technical solution of the present invention is as follows:

[0058] Please refer to Figure 1 .

[0059] A multi-source data selective fusion method for multi-environment positioning of an autonomous vehicle, and the method is as follows:

[0060] S1: Information control: Classify the environment, assign credibility weights to each sensor in different environments, and for a specified sensor in a specified working environment, selectively fuse its output data according to its credibility weight;

[0061] S2: Control evaluation: Analyze and evaluate the information control process, and evaluate its reliability, risk rate and usability;

[0062] S3: Evaluation feedback: Analyze the result of the information control according to the control evaluation result, adjust the control strategy in a timely manner, and continuously select more reliable sensor data to be sent to the subsequent data fusion link for analysis and fusion.

[0063] Further, in this specific implementation manner, S1 is specifically as follows:

[0064] S11: Arrange sensors: Arrange four sensors on the vehicle, namely an inertial navigation system, a GNSS satellite locator, a camera, and a lidar; that is, in this specific implementation manner, n = 4.

[0065] S12: Specify the basic scenarios: Denote the scenario where only the inertial navigation system fails and other sensors work normally as the first basic scenario c1, denote the scenario where only the GNSS satellite locator fails and other sensors work normally as the second basic scenario c2, denote the situation where only the camera fails and other sensors work normally as the third basic scenario c3, and the situation where only the lidar fails and other sensors work normally as the fourth basic scenario c4.

[0066] S13: Define the acquisition degree: Define the acquisition degree D to characterize the acquisition of the output data of the corresponding sensor. It is stipulated that the value of the acquisition degree D is 0 or 1. When the value is 0, the output data of the corresponding sensor is not acquired. When the value is 1, the output data of the corresponding sensor is acquired;

[0067] S14: Calculate the acquisition degree of each sensor in each basic scenario: According to the sensor failure situation in each basic scenario, for the sensors that fail in this basic scenario, their sensor data is not acquired, and the acquisition degree takes the value of 0. For the sensors that are working normally, their sensor data is acquired, and the acquisition degree takes the value of 1; Therefore, it can be known that the relationship among the sensor, the basic scenario, and the acquisition degree of the basic scenario should be as follows in the table:

[0068]

[0069] Furthermore, S1 also includes:

[0070] S15: Divide the working environment: When it is determined that all sensors are working normally, the vehicle is in the normal working environment, denoted as E0;

[0071] According to the total set C of basic scenarios, the abnormal environment is divided into the GNSS satellite locator signal weak or signal loss environment, the weak texture environment, the weak vision or night environment, the bad weather environment, and the simulated inertial navigation failure environment; that is, in this specific embodiment, m = 5. The total set E of all working environments is obtained: E = {E0, E1, E2, E3, E4, E5}.

[0072] S16: Calculate the acquisition degrees of sensors in each working environment: According to the sensor failure situations in each basic scenario, for the sensors that fail in the basic scenario, do not collect the data of these sensors, and their acquisition degrees are set to 0. For the sensors that are working normally, collect the data of these sensors, and their acquisition degrees are set to 1. It can be easily inferred that in the normal working environment, the inertial navigation, GNSS satellite locator, camera, and lidar are all working normally, so their respective acquisition degrees are all 1. In the environment where the GNSS satellite locator has weak signal or signal loss, the signal of the GNSS satellite locator is affected. That is, this working environment is essentially the second basic scenario c2. At this time, the acquisition degree of the GNSS satellite locator should be set to 0 (but considering the important role of the GNSS satellite locator in vehicle positioning, even if the signal of the GNSS satellite locator is weak, as long as its signal exists, the signal of the GNSS satellite locator should also be collected. Therefore, for a specific GNSS satellite locator, its acquisition degree in the environment where the GNSS satellite locator has weak signal or signal loss should be 1 / 0). In the weak texture environment, the camera signal may be deviated, but considering the role of the image signal in vehicle positioning, the camera signal should be collected. Therefore, in the weak texture environment, the acquisition degrees of all sensors are set to 1. In the weak vision or night environment, the camera is seriously interfered by external factors and cannot image normally. That is, this working environment is essentially the third basic scenario c3. At this time, the acquisition degree of the camera should be set to 0, and the acquisition degrees of other sensors remain 1. In the bad weather environment, both the camera and the lidar are greatly affected by the external environment, and their output data may have large errors. That is, this working environment is essentially the superposition of the third basic scenario c3 and the fourth basic scenario c4. Therefore, in the bad weather environment, the acquisition degrees of the camera and the lidar should be set to 0, and the acquisition degrees of other sensors remain 1. In the simulated inertial navigation failure environment, the output data of the inertial navigation is abnormal. That is, this working environment is essentially the first basic scenario c2. At this time, the acquisition degree of the inertial navigation should be set to 0, and the acquisition degrees of other sensors remain 1. Therefore, it can be known that the total set of sensors is S, the total set of working environments is E, and the set of acquisition degrees of sensors is D E The relationship of

[0073]

[0074] Furthermore, S1 also includes:

[0075] S17: Define the attention degree: Define the attention degree A to represent the credibility of the output data of the corresponding sensor, and it is stipulated that the value of the attention degree A is in the range of [0, 1]. The larger the value of the attention degree A, the higher the trust degree given to the output data of the corresponding sensor. Keep the sum of the attention degree values of all sensors in the same working environment as 1, and assign the smallest attention degree value to the sensors that fail in the current working environment;

[0076] S18: Allocate attention: Allocate attention to each sensor in all working environments. In normal working environments, all sensors work properly. At this time, according to experience, the maximum attention should be allocated to the GNSS satellite locator. In an environment where the GNSS satellite locator has weak or lost signals, the inertial navigation is taken as the sensor focus and the maximum attention value is allocated to it. In a weak texture environment, the camera signal is interfered to a certain extent, so the minimum attention value is allocated. In a weak vision or night environment, the camera is severely affected and can hardly image. At this time, the minimum attention is allocated to the camera. Considering the weak vision or night environment, the lidar is also affected to a certain extent. Therefore, in this working environment, a relatively small attention value needs to be allocated to the lidar. In a harsh weather environment, affected by wind, frost, rain and snow, the camera and lidar are almost completely ineffective. At this time, the minimum attention is allocated to the camera and lidar. In a simulated inertial navigation failure environment, the inertial navigation signal completely disappears. At this time, the GNSS satellite locator should be taken as the sensor focus and the minimum attention is allocated to the inertial navigation.

[0077] Therefore, there is a relationship among the total set of sensors S, the total set of working environments E, and the total set of attention of sensors A E as follows in the table:

[0078]

[0079] Furthermore, S1 also includes:

[0080] S19: Set the fusion judgment threshold: Set the fusion judgment threshold Q α = 0.8;

[0081] S110: Calculate the actual fusion parameter: Define the fusion parameter as Q. Denote the fusion parameter of sensor s x under the working environment E y as It is stipulated that: The fusion parameter total set Q is obtained by aggregating the fusion parameters of each sensor in each working environment E :

[0082] For different sensors in different environments, different external factors will affect the corresponding sensors to different degrees. According to the sensors that can work properly in this working scenario, keeping the sensor acquisition degree at 1 means that in this working scenario, the external factors have not affected it and the output data of this sensor can be normally acquired. Otherwise, it means that in this working scenario, the external factors interfere with the corresponding sensor, resulting in a large error in the sensor output data. At this time, the acquisition of the output data of this sensor is stopped, and the data with large errors is prohibited from entering the system to avoid spending a large amount of computing power and time on correcting the sensor data with large errors during subsequent data processing.

[0083] In different working environments, we can further allocate attention to different sensors. Then, we can set data importance for multiple sensors working together in each working environment, allocate higher attention to sensor data that is more reliable in the working scenario, and allocate the lowest attention to sensors that have failed in the working scenario. In this way, we can distinguish the status of different sensors in making the next autonomous driving decision for the vehicle in a specific driving environment, make more effective use of front-end sensor data, and provide more reliable detection data for vehicle autonomous driving.

[0084] S111: Fusion determination: Setting fusion conditions and For the total set of fusion parameters Q E Verify each element one by one;

[0085] S112: Selective fusion: If the current element meets the fusion condition set in S111, the current sensor in the current working environment represented by the element is allowed to output sensor data to the subsequent data fusion link; otherwise, the output data of the current sensor in the current working environment represented by the element is abandoned. The calculated fusion parameters are It will simultaneously express the working environment of the sensor, the impact of its working environment on the sensor, the importance of the output data of the sensor in the working environment, etc. One will assess the trustworthiness of current sensors for their own data; and for the current working environment, It will express the impact of this working scenario on all sensors installed on the vehicle, that is, for the current working environment E x If there is This means that in this working environment, some sensors may fail due to external factors, but the remaining sensors that can work normally can provide the vehicle with sufficiently reliable real-time detection data, and the vehicle can operate in this working environment. x Therefore, it can be seen that setting the fusion condition and For the total set of fusion parameters Q E Each element in the system is verified one by one; the sensor data in different working environments can be comprehensively evaluated from the level of a single sensor and the entire working scenario, and sufficiently reliable data can be selected to enter the subsequent data fusion link to ensure that the sensor data determined by the fusion are reliable and have few errors, which provides computing power for the subsequent sensor data fusion process and improves the fusion efficiency of the subsequent data fusion process.

[0086] Furthermore, S2 is specifically:

[0087] S21: Denote that there are k interruptions in the entire information control process; conduct a reliability assessment on the entire information control process to evaluate the normal working continuity of the entire information control process;

[0088] S22: Conduct a risk rate assessment on the entire information control process and calculate the normal working duration of the vehicle before a certain interruption occurs during the information control process;

[0089] S23: Conduct an availability assessment on the entire information control process and calculate the effective availability of the data and the time proportion of the effective data during the entire information control process. For the information control process described above, evaluate it from three aspects: reliability, risk rate, and availability, which can fully evaluate the work of its information control process. The evaluation results are transmitted to S3, and S3 continuously adjusts according to the evaluation results

[0090] Furthermore, S21 is specifically as follows:

[0091] Define the reliability of the information control process within the time period Δt as K(Δt): K(Δt) = e-λΔt. For the entire information control process, calculate the average reliability value K(Δt) of its entire time avg : and the root mean square value of reliability K(Δt) of the entire time rms :

[0092] Furthermore, S22 is specifically as follows:

[0093] Denote the running time before the σth interruption occurs in the zth work as Then there is: Calculate the risk rate P of the entire information control process risk as:

[0094] Furthermore, S23 is specifically as follows:

[0095] S231: Conduct multiple samplings in two ways: at a fixed time interval ΔT and a random time interval Ran(T) in the data stream, and each sampling is recorded as a data point;

[0096] S232: Denote the amount of information data in a certain data point as ζ, and the types of information data as Set the target amount of information data ζ set and the target number of information data types as

[0097] S234: Classify and count the data points: F1 represents the number of data points that do not meet the data selection requirements, and its expression form is: 1 ≤ ζ < ζ set<n; F2 represents the number of data points that do not meet the integrity requirements, and its expression form is: the amount of information data in the data points F3 represents the number of data points with data continuity interruption, and its expression form is: ζ = 0, F represents the total number of data points sampled and statistically counted within the total time of the whole process;

[0098] S235: Calculate the effective availability of data in the entire information control process:

[0099] It should be noted that the more sampling times, the more accurate the availability evaluation, but the increase in workload also needs to be considered. Therefore, in the sampling process at a fixed time interval, the following regulations are stipulated in this specific implementation manner:

[0100] When the data flow time t > 1h, the sampling time interval Δt ≤ 1s;

[0101] When the data flow time 30min ≤ t ≤ 1h, the sampling time interval Δt ≤ 0.5s;

[0102] When the data flow time 0 < t < 30min, the sampling time interval Δt ≤ 0.1s.

[0103] S236: Denote the time t when the data in the data stream is effectively available 有效 occupying the entire information control time T 全部 When ζ = ζ set 、 , count the time t of the effective data 有效 and calculate the time occupancy ratio ω of the effective data:

[0104] The availability evaluation is a judgment of functional availability, which is carried out when the information control meets the purpose of the scheme, can normally realize the function and play a role, and is also based on the selection requirements, integrity requirements and continuity requirements. The original multi-source data is selected to form a screened data stream, and the availability is analyzed from two perspectives: the effective availability of data in the information control process and the time occupancy ratio of the effective data, which can provide result feedback for the information control process so that the vehicle can timely adjust the control strategy.

[0105] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-source data selective fusion method for multi-environment positioning of unmanned vehicles, characterized in that The method is as follows: S1: Information control: Classify the environment, assign credibility weights to each sensor in different environments, and for a specified sensor in a specified working environment, selectively fuse its output data according to its credibility weight. S2: Control evaluation: Analyze and evaluate the information control process, and evaluate its reliability, risk rate, and availability. S3: Evaluation feedback: Analyze and fuse the results of information control according to the control evaluation results. Specifically, S1 is as follows: S11: Arranging sensors: Arrange n types of sensors on the vehicle, where n ≥ 2 and n is an integer; Denote the specified sensor as s i , Aggregate all the sensors on the vehicle to obtain the total sensor set S: S = {s1, s2... s n}, There is: S12: Define the basic scenarios: Denote the situation where only sensor s1 fails and other sensors s2, s3, …, s n are all working properly as the first basic scenario c1; Denote the situation where only sensor s2 fails and other sensors s1, s3, …, s n are all working properly as the second basic scenario c2; And so on, denote the situation where only sensor s i fails and other sensors s1, s2, …, s i-1 , s i+1 , …, s n are all working properly as the i-th basic scenario c i ; And so on, denote the situation where only sensor s n fails and other sensors s1, s2, …, s n-1 are all working properly as the n-th basic scenario c n ; Combine all the basic scenarios to obtain the total set of basic scenarios C: C = {c1, c2, …, c n}, where there is S13: Define the acquisition degree: Define the acquisition degree D to represent the acquisition situation of the output data of the corresponding sensor, and it is stipulated that the value of the acquisition degree D is 0 or 1. When the value is 0, the output data of the corresponding sensor is not acquired. When the value is 1, the output data of the corresponding sensor is acquired. S14: Calculate the acquisition degree of each sensor in each basic scenario: According to the sensor failure situation in each basic scenario, for the sensors that fail in this basic scenario, do not acquire their sensor data, and the acquisition degree takes the value of 0. For the sensors that are working normally, acquire their sensor data, and the acquisition degree takes the value of 1. Denote the i-th basic scenario as c i Under this condition, the sensor s j has a collection degree of Collect the collection degrees of each sensor under each basic scenario to obtain the total set D of basic scenario collection degrees C : There is S15: Divide the working environment: When it is determined that all sensors are working normally, the vehicle is in the normal working environment, and the normal working environment is denoted as E0. The abnormal environments are divided into m types according to the total set C of basic scenarios, where and m is an integer; the abnormal environments E1, E2... E m should be one or more combinations of the above-mentioned first basic scenario C1, second basic scenario C2... and nth basic scenario C n ; Record the specified working environment as E x ; The total set of working environments E is obtained by aggregating all working environments: E = {E0, E1, E2... E m}, where and E x ≠ E0; S16: Calculate the acquisition degree of each sensor in each working environment: According to the sensor failure situation in each basic scenario, for the sensors that fail in this basic scenario, do not acquire the data of this sensor, and the acquisition degree takes the value of 0. For the sensors that are working normally, acquire the data of this sensor, and the acquisition degree takes the value of 1. Record the working environment E x Under the condition, sensor s y The acquisition degree is Collect the acquisition degree set D of all sensors under all working environments E : There is S17: Define the attention degree: Define the attention degree A to represent the credibility degree of the output data of the corresponding sensor, and it is stipulated that the value of the attention degree A is in the range of [0, 1]. The larger the value of the attention degree A, the higher the trust degree given to the output data of the corresponding sensor. Keep the sum of the attention degree values of all sensors in the same working environment as 1, and assign the smallest attention degree value to the sensors that fail in the current working environment. S18: Allocate attention: Denote the working environment as E x Under the condition of y the attention of sensor s Collect the attention of each type of sensor under all working environments to obtain the total attention set A E : There is S19: Set the fusion determination threshold: Set the fusion determination threshold Q α ; S110: Calculate the actual fusion parameter: Define the fusion parameter as Q, and denote the working environment as E x Under the condition of y , the fusion parameter of the sensor s is stipulated as Collect the fusion parameters of each sensor in each working environment to obtain the total set of fusion parameters Q E : S111: Fusion determination: Set fusion conditions and verify each element in the total set Q of fusion parameters E one by one; S112: Selective fusion: If the current element meets the fusion conditions set in S111, then allow the current sensor in the current working environment represented by this element to output sensor data to the subsequent data fusion link. Conversely, discard the output data of the current sensor in the current working environment represented by this element.

2. The multi-source data selective fusion method for multi-environment positioning of driverless vehicles according to claim 1, characterized in that, Specifically, S2 is as follows: S21: Record that there are k interruptions in the entire information control process; Make a reliability assessment of the entire information control process and evaluate the normal working continuity of the entire information control process. S22: Make a risk rate assessment of the entire information control process and calculate the normal working duration of the vehicle before a certain interruption occurs in the information control process. S23: Make an availability assessment of the entire information control process and calculate the effective availability of the data in the entire information control process and the time ratio of the effective data.

3. The multi-source data selective fusion method for multi-environment positioning of driverless vehicles according to claim 2, wherein Specifically, S21 is as follows: Define the reliability of the information control process within the time period Δt as K(Δt): K(Δt) = e -λΔt , for the entire information control process, calculate the average reliability value K(Δt) over all its time avg : and the root mean square value of reliability K(Δt) over all its time rms :

4. The multi-source data selective fusion method for multi-environment positioning of driverless vehicles according to claim 2, wherein, Specifically, S22 is as follows: Let the running time before the $\sigma$-th interruption occurs in the $z$-th operation be Then we have: Calculate the risk rate $P$ of the entire information control process risk as:

5. The multi-source data selective fusion method for multi-environment positioning of driverless vehicles according to claim 2, wherein Specifically, S23 is as follows: S231: Perform multiple samplings in the data stream in two ways: at a fixed time interval ΔT and a random time interval Ran(T). Each sampling is recorded as a data point. S232: Denote the amount of information data in a certain data point as ζ, and the types of information data as Set the target amount of information data ζ set , and the target number of information data types as S234: Classify and statistically analyze data points: F1 represents the number of data points that do not meet the data selection requirements, and its expression is: 1 ≤ ζ < ζ set < n; F2 represents the number of data points that do not meet the integrity requirements, and its expression is: the amount of information data in the data points F3 represents the number of data points where data continuity is interrupted, and its expression is: ζ = 0, F represents the total number of data points sampled and statistically analyzed within the total time of the whole process; S235: Calculate the effective availability of data during the entire information control process: S236: Record the time t when the data in the data stream is valid and available 有效 Accounting for the entire information control time T 全部 , when ζ = ζ set , , count the time t of the valid data 有效 , calculate the time ratio ω of the valid data:

Images