A multi-sensor fusion positioning method, system and device for an autonomous vehicle

Abstract

The application discloses a multi-sensor fusion positioning method, system and device for an automatic driving vehicle, which is improved and optimized based on multi-sensor fusion positioning. For multi-source positioning data (GPS positioning and NDT positioning) import, not only the current state quantity of various positioning modes and the change of previous frames of data are considered, but also the overall change trend of previous 3s of data is fully considered, and the trend is analyzed in a fuzzy manner to obtain a comprehensive judgment on whether the current positioning state is reliable. In combination with the reliability score of multi-source positioning and the comprehensive optimization weight coefficient of the current state value of each source, a positioning result is obtained and used for vehicle automatic driving.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and specifically to a multi-sensor fusion localization method, system, and apparatus that employs confidence coefficients and a two-layer fuzzy rule table. Background Technology

[0002] Vehicle positioning has always been crucial in autonomous driving. Ensuring accurate vehicle positioning remains a significant technical challenge. Currently, a widely adopted approach is to utilize multi-sensor fusion to achieve positioning redundancy, thereby improving positioning accuracy and robustness. The most common method is to fuse GPS positioning with SLAM positioning to enhance vehicle positioning performance. Among existing LiDAR SLAM positioning methods, the NDT algorithm is the most widely used and is extensively applied in LiDAR SLAM map registration and positioning. During vehicle operation, Kalman filtering is employed to fuse NDT and GPS positioning methods using multi-sensor positioning, thereby improving positioning performance.

[0003] However, in practical applications, Kalman filtering involves Jacobian matrix operations, which require significant computing power, especially for vehicle controllers and embedded systems.

[0004] Sensor fusion, through the Kalman filter algorithm, is essentially a process of finding the minimum difference to seek the optimal solution. However, it ignores the possibility of errors in the sensor's own parameters, which can lead to false positives and false negatives. That is, the sensor's own parameters may indicate that the positioning is very accurate, but in reality, it is not accurate enough, or the sensor's own parameters may indicate that the positioning is inaccurate, but in reality, it is relatively accurate. Summary of the Invention

[0005] This invention has made certain improvements and optimizations based on multi-sensor fusion positioning. For the import of multi-source positioning data (GPS positioning and NDT positioning), it not only considers the current state of various positioning methods and the changes in the previous few frames of data, but also fully considers the overall trend of data changes in the previous 3 seconds. The trend is analyzed in a fuzzy manner to obtain a comprehensive judgment on whether the positioning status at the current moment is reliable. Then, the reliability score of multi-source positioning and the weight coefficient of each current state value are combined to obtain the positioning result, which can be used for autonomous driving of vehicles.

[0006] To address the problems existing in the prior art, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a multi-sensor fusion localization method for autonomous vehicles, the method comprising the following steps:

[0008] S1: Retrieve the GPS status information and coordinate data information and NDT status information and coordinate data information at time t1-tk obtained by the positioning module.

[0009] S2: Match the GPS status information and coordinate data information at times t1-tk with the GPS confidence fuzzy rule table to obtain the confidence coefficients of the GPS status information and coordinate data information. Match the NDT status information and coordinate data information at times t1-tk with the NDT confidence fuzzy rule table to obtain the confidence coefficients of the NDT status information and coordinate data information.

[0010] S3: Obtain the confidence coefficients of GPS status information and coordinate data, the confidence coefficients of NDT status information and coordinate data, the GPS status information and coordinate data at time tk, and the NDT status information and coordinate data at time tk. Match the above four data items to make a comprehensive judgment on the fuzzy control rule table and output the comprehensive weight coefficient of the positioning data.

[0011] S4: The comprehensive positioning data weighting coefficient is determined by combining the GPS coordinate data and NDT coordinate data at time tk to obtain the final comprehensive positioning information.

[0012] In conjunction with the first aspect, the first possible scenario under any of its possible conditions is that step S3 is:

[0013] The confidence coefficients of GPS status information and coordinate data, and the confidence coefficients of NDT status information and coordinate data are obtained. These are then matched against a two-layer integrated fuzzy control rule table to obtain the confidence coefficients of GPS status information and coordinate data at time tk, as well as the confidence coefficients of NDT status information and coordinate data at time tk.

[0014] The GPS status information and coordinate data at time tk, along with their corresponding confidence coefficients, and the NDT status information and coordinate data at time tk, along with their corresponding confidence coefficients, are matched with a two-layer comprehensive judgment fuzzy control rule table to output the comprehensive weight coefficient of the positioning data.

[0015] In combination with the first aspect or the first situation mentioned above, the second situation under any possible circumstances is as follows: Step S4 is: the comprehensive weighting coefficient of the positioning data is multiplied by the GPS status information and coordinate data information at time tk and the NDT status information and coordinate data information at time tk respectively, and the final comprehensive positioning information is determined based on the calculation results.

[0016] As a second aspect, a multi-sensor fusion positioning module for autonomous vehicles, characterized in that the system comprises:

[0017] The positioning module includes a GPS positioning module and an NDT positioning module. The GPS positioning module is used to acquire GPS status information and coordinate data information, and the NDT positioning module is used to acquire NDT status information and coordinate data information.

[0018] The data processing module is used to call and store the GPS status information and coordinate data information and NDT status information and coordinate data information obtained by the positioning module. Based on the stored GPS confidence fuzzy rule table, NDT confidence fuzzy rule table and comprehensive judgment fuzzy control rule table, it calculates and analyzes to obtain the final comprehensive positioning information.

[0019] As a third aspect, the present invention provides a computer-readable storage medium for storing one or more programs, the computer-readable storage medium storing one or more program instructions, which, when executed by a processor, perform any one of the methods described in the first aspect or its first or second case.

[0020] As a fourth aspect, the present invention provides a computer including a central processing unit, a computer-readable storage device, and a data bus. The central processing unit is connected to the computer-readable storage device via the data bus. The central processing unit executes any one of the methods described in the first aspect or its first or second case according to the operation instructions stored in the computer-readable storage device.

[0021] The beneficial effects of this invention are:

[0022] By boldly questioning the reliability of the state variables of GPS and NDT positioning, and by introducing the state information of the previous k time steps and adding a confidence coefficient to the next time step, the misjudgment of false good and false bad positioning states is further eliminated. By using a two-layer fuzzy rule table to quantitatively analyze the good and bad and trend of positioning state information, the accuracy and robustness of positioning are improved, and the erroneous judgment of the good and bad of multi-source positioning states is further avoided. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the method in Example 1;

[0025] Figure 2 This is a system structure diagram of Example 2. Detailed Implementation

[0026] The technical solution of the present invention will be further described in detail below through embodiments and with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention.

[0027] The technical solution of the present invention will now be described in conjunction with the accompanying drawings and embodiments. The meanings of some of the English abbreviations are as follows:

[0028] SLAM – Simultaneous Localization and Mapping

[0029] NDT – Normal Distribution Transform

[0030] GPS - Global Positioning System

[0031] Example 1:

[0032] like Figure 1 As shown in the figure, this embodiment provides a multi-sensor fusion localization method for autonomous vehicles, and the method steps are as follows:

[0033] This invention provides a multi-sensor fusion localization method for autonomous vehicles, the method comprising the following steps:

[0034] S1: The data processing module retrieves the GPS status information and coordinate data information and the NDT status information and coordinate data information at time t1-tk obtained by the positioning module.

[0035] S2: The data processing module matches the GPS status information and coordinate data information at times t1-tk with the GPS confidence fuzzy rule table to obtain the confidence coefficients of the GPS status information and coordinate data information. The data processing module also matches the NDT status information and coordinate data information at times t1-tk with the NDT confidence fuzzy rule table to obtain the confidence coefficients of the NDT status information and coordinate data information.

[0036] Steps S1-S2 introduce the state information from the previous k time steps and add a confidence coefficient to the subsequent time step, further eliminating false positives and negatives in positioning. For example: at time t1, although both NDT and GPS positioning statuses show "good," based on previous data, GPS has been fluctuating between "bad" and "good," while NDT has been fluctuating between "good" and "good." Trend analysis suggests that NDT's current "good" status is more realistic, so the fuzzy rule table will increase the confidence coefficient for NDT positioning. Similarly, if GPS has been fluctuating between "good" and "good," but NDT has been fluctuating between "bad" and "good," trend analysis suggests that GPS's current "good" status is more realistic, so the fuzzy rule table will increase the confidence coefficient for GPS positioning.

[0037] The GPS and NDT confidence fuzzy rule tables described in this invention represent a process of converting precise quantities into fuzzy quantities, known as fuzzification or fuzzy quantization, typically achieved by discretizing precise quantities. Each variable in the fuzzy set is assigned a confidence level (also called a confidence coefficient or confidence parameter). Fuzzy control rules (FAM) (i.e., the confidence fuzzy rule table and the comprehensive judgment fuzzy control rule table described in this invention; the fuzzy rule table is a collection of judgment conditions for the fuzzy control rules) link the fuzzy input and output sets. These fuzzy control rules can be expressed in natural language, usually using if-then statements, thereby establishing the fuzzy control rules and forming the aforementioned confidence fuzzy rule table and comprehensive judgment fuzzy control rule table.

[0038] The fuzzy rules for the GPS and NDT confidence fuzzy rule tables are as follows: The state information and coordinate data of NDT and GPS are divided into five levels: very poor, poor, zero, good, and excellent. Different levels correspond to different confidence coefficients (range: 0-1). Inputting k GPS state information and coordinate data into the GPS confidence fuzzy rule table will output a GPS confidence coefficient of 0-1. Similarly, inputting k NDT state information and coordinate data into the NDT confidence fuzzy rule table will output an NDT confidence coefficient of 0-1. The GPS confidence fuzzy rule table and the NDT confidence fuzzy rule table have similar structures, with a total of k*5*3 judgment conditions. The content of these judgment conditions can be manually defined or learned and defined by artificial intelligence. The output has three scenarios: the NDT or GPS state information and coordinate data are: unreliable (0-0.3), average (0.3-0.7), and very reliable (0.7-1).

[0039] S3: Obtain the confidence coefficients of GPS status information and coordinate data, and the confidence coefficients of NDT status information and coordinate data. Match them with the two-layer comprehensive judgment fuzzy control rule table to obtain the confidence coefficients of GPS status information and coordinate data at time tk, and the confidence coefficients of NDT status information and coordinate data at time tk.

[0040] The GPS status information and coordinate data at time tk, along with their corresponding confidence coefficients, and the NDT status information and coordinate data at time tk, along with their corresponding confidence coefficients, are matched with a two-layer comprehensive judgment fuzzy control rule table to output the comprehensive weight coefficient w1 of the positioning data.

[0041] S4: The comprehensive weighting coefficient w1 of the positioning data is multiplied by the GPS coordinate data information and the NDT coordinate data information at time tk, respectively. The final comprehensive positioning information is determined based on the calculation results. The final positioning coordinates are w1*GPS coordinates + (1-w1)*NDT coordinates.

[0042] The comprehensive judgment fuzzy control rule table contains 625 judgment conditions. The input includes GPS and NDT status information and coordinate data at time tk, along with their corresponding confidence coefficients. Based on the judgment conditions set by humans or artificial intelligence learning, the output consists of five cases of comprehensive weight coefficients for positioning data: w1-0~0.05 (strong negative), w1-0.05~0.4 (negative), w1-0.4~0.6 (zero), w1-0.6~0.95 (positive), and w1-0.95~1 (strong positive).

[0043] In steps S3-S4, the good and bad of the positioning status information and the good and bad trend are quantitatively analyzed by the two-layer comprehensive judgment fuzzy control rule table, which improves the accuracy and robustness of positioning and further avoids the situation of incorrect judgment of the good and bad of multi-source positioning status.

[0044] In existing technologies, for the NDT positioning method, the state variables presented at time t1 are "NDT positioning is good" and "GPS positioning is good," and the GPS positioning state variables are also "GPS positioning is good." Furthermore, the GPS positioning state was also good at the previous time, while the NDT positioning state was bad. According to the Kalman filter algorithm, the calculation result will most likely show that the GPS error is better than the NDT positioning method at this time. Therefore, the GPS positioning result is used as the current positioning at time t1. However, in real vehicle testing, there are situations where, although the GPS shows that its own state is good, the actual positioning state is not good, presenting a false positive state. Therefore, it is necessary to judge and identify such situations.

[0045] The algorithm in this embodiment not only considers the current situation and the previous frame data, but also comprehensively considers the positioning status change values ​​of the previous few frames. Based on the data of the previous few frames, it makes a comprehensive judgment to obtain the confidence coefficient of the current GPS and NDT positioning. Then, based on the reliability coefficients of the two positioning methods and the current state quantity, it outputs the positioning weight coefficient. The current positioning information is obtained by multiplying the weight coefficient by the positioning coordinates.

[0046] For example, near a certain timestamp, although both NDT and GPS status and coordinate data are good, the status and coordinate data for a short period near this time are input into the GPS and NDT confidence fuzzy rule tables respectively. If the GPS jump is large and remains good for a long time, but the NDT jump is small, then according to the GPS and NDT confidence fuzzy rule tables, the output confidence parameters are determined to be 0.3 for GPS and 0.7 for NDT. Combining the status and coordinate data of both GPS and NDT with the comprehensive judgment fuzzy control rule table, the output positioning data comprehensive weight coefficient w1 is determined to be negative. This will cause the car to trust the NDT positioning more, thus obtaining a more reliable final positioning. This scheme avoids choosing the less reliable GPS positioning method when it is impossible to accurately determine which positioning method to use.

[0047] Example 2:

[0048] like Figure 2 As shown, this embodiment provides a multi-sensor fusion positioning module for autonomous vehicles, characterized in that the system includes:

[0049] The positioning module includes a GPS positioning module and an NDT positioning module. The GPS positioning module is used to acquire GPS status information and coordinate data information, and the NDT positioning module is used to acquire NDT status information and coordinate data information.

[0050] The data processing module is used to call and store the GPS status information and coordinate data information and NDT status information and coordinate data information obtained by the positioning module. Based on the stored GPS confidence fuzzy rule table, NDT confidence fuzzy rule table and comprehensive judgment fuzzy control rule table, it calculates and analyzes to obtain the final comprehensive positioning information.

[0051] Example 3:

[0052] This embodiment provides a computer-readable storage medium for storing one or more programs, characterized in that the computer-readable storage medium stores one or more program instructions, which, when executed by a processor, complete the method steps described in Embodiments 1, 2, and other embodiments not described that can implement the method of the present invention.

[0053] Example 4:

[0054] This embodiment provides a computer, which includes a central processing unit (CPU), a computer-readable storage device (CROM), and a data bus. The CPU is connected to the CROM via the data bus, and the CPU executes the method steps of Embodiment 1, Embodiment 2, and other embodiments not described that can implement the method of this invention according to the operation instructions stored in the CROM.

[0055] It should be understood that the above embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading this invention, those skilled in the art can make various alterations and modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A multi-sensor fusion positioning method for an autonomous vehicle, characterized in that, The method steps are as follows: S1. The GPS state information and coordinate data information at t1-tk time obtained by the positioning module are matched with the GPS confidence fuzzy rule table to obtain the confidence coefficient of the GPS state information and coordinate data information, and the NDT state information and coordinate data information at t1-tk time are matched with the NDT confidence fuzzy rule table to obtain the confidence coefficient of the NDT state information and coordinate data information; S2. The GPS state information and coordinate data information at t1-tk time obtained by the positioning module are matched with the GPS confidence fuzzy rule table to obtain the confidence coefficient of the GPS state information and coordinate data information, and the NDT state information and coordinate data information at t1-tk time are matched with the NDT confidence fuzzy rule table to obtain the confidence coefficient of the NDT state information and coordinate data information; S3. The GPS state information and coordinate data information confidence coefficient, NDT state information and coordinate data information confidence coefficient, GPS state information and coordinate data information at tk time, and NDT state information and coordinate data information at tk time are matched with the comprehensive judgment fuzzy control rule table to output the positioning data comprehensive weight coefficient; S4. The positioning data comprehensive weight coefficient is combined with the GPS state information and coordinate data information at tk time and the NDT state information and coordinate data information at tk time to determine the final comprehensive positioning information; The step S3 is: The GPS state information and coordinate data information confidence coefficient and the NDT state information and coordinate data information confidence coefficient are obtained by matching the double-layer comprehensive judgment fuzzy control rule table, and the confidence coefficient of the GPS state information and coordinate data information at tk time and the confidence coefficient of the NDT state information and coordinate data information at tk time are obtained. The GPS state information and coordinate data information at tk time and its corresponding confidence coefficient, and the NDT state information and coordinate data information at tk time and its corresponding confidence coefficient are matched with the double-layer comprehensive judgment fuzzy control rule table to output the positioning data comprehensive weight coefficient. 2.The multi-sensor fusion positioning method for an autonomous vehicle of claim 1, wherein, The step S4 is: the positioning data comprehensive weight coefficient is multiplied by the GPS coordinate data information at tk time and the NDT coordinate data information at tk time, respectively, and the final comprehensive positioning information is determined according to the operation result.

3. A multi-sensor fusion positioning system for an autonomous vehicle, characterized by, The multi-sensor fusion positioning method for realizing the automatic driving vehicle of any one of claims 1-2, the system comprises: A positioning module comprising a GPS positioning module and an NDT positioning module, the GPS positioning module is used for obtaining GPS state information and coordinate data information, and the NDT positioning module is used for obtaining NDT state information and coordinate data information; The data processing module is used for calling and storing the GPS state information and coordinate data information and the NDT state information and coordinate data information acquired by the positioning module, obtaining final comprehensive positioning information through operation analysis according to the stored GPS confidence fuzzy rule table, NDT confidence fuzzy rule table and comprehensive judgment fuzzy control rule table, acquiring GPS state information and coordinate data information confidence coefficients, NDT state information and coordinate data information confidence coefficients, matching the double-layer comprehensive judgment fuzzy control rule table, obtaining the GPS state information and coordinate data information confidence coefficients at the time t and the NDT state information and coordinate data information confidence coefficients at the time t, matching the double-layer comprehensive judgment fuzzy control rule table with the GPS state information and coordinate data information at the time t and the corresponding confidence coefficients, the NDT state information and coordinate data information at the time t and the corresponding confidence coefficients, and outputting the positioning data comprehensive weight coefficients.

4. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method comprising: The computer readable storage medium stores one or more program instructions, and the program instructions are executed by the processor to complete the method in any one of claims 1 to 2.

5. A computer comprising a central processing unit, a computer readable memory, a data bus, characterized in that: The central processor is connected with the computer readable storage medium through a data bus, and the central processor executes the method in any one of claims 1 to 2 according to the operation instructions stored in the computer readable storage medium.

Images