A bayesian fusion mapping method and system based on confidence conduction

Abstract

This invention discloses a Bayesian fusion mapping method and system based on confidence propagation. Addressing the problems of traditional UAV mapping being prone to "map flickering" due to sensor noise interference and the inability of static obstacle avoidance boundaries to adapt to high-speed flight, this invention calculates instantaneous confidence based on measurement geometry and uses it as weights for spatial diffusion filtering. It compares the filtering differences to obtain spatial consistency and dynamic update gain, achieving adaptive cascaded updates of the posterior probability of the raster map. Furthermore, it combines real-time ground speed and heading to extract anisotropic physical occupancy states and constructs a dynamic safety repulsion field coupling kinematic braking and mapping uncertainties. This invention is mainly used for high-precision 3D mapping and high-speed dynamic obstacle avoidance of UAVs in complex and unknown environments, effectively filtering out instantaneous noise and providing dynamic collision avoidance safety margins.

Description

Technical Field

[0001] This invention belongs to the field of UAV environmental perception and autonomous navigation technology, specifically involving a Bayesian fusion mapping method and system based on confidence transfer. Background Technology

[0002] With the development of sensor technology, computer vision technology, artificial intelligence algorithms, and computing hardware, the autonomous operation capabilities of unmanned aerial vehicle (UAV) systems in complex environments have been significantly improved, and their applications now widely cover agricultural plant protection, power line inspection, forest fire fighting, and logistics delivery. In these complex and ever-changing application scenarios, the high-speed autonomous flight capability of UAVs is closely related to their real-time and accurate environmental perception capabilities. To ensure flight safety and mission efficiency, it is urgent to solve the key problem of refined mapping for real-time obstacle avoidance of UAVs in high-speed flight modes.

[0003] The obstacle avoidance process of drones relies heavily on onboard sensors (such as LiDAR and depth cameras) to perceive the surrounding environment and construct environmental maps using this data, thus providing decision-making basis for upper-level motion planning. However, in actual operations, due to the inherent physical characteristics of sensors, changes in ambient lighting, and surface textures of objects, the 3D point cloud data collected by sensors is often accompanied by unavoidable instantaneous noise and measurement errors. Traditional environmental mapping algorithms (such as simple occupancy grid map updates) often use a fixed probability update mechanism when processing this noisy observation data, ignoring the confidence differences in spatial distance and observation perspective. This results in maps prone to obstacle ghosting, isolated noise points, or environmental holes. Such inaccuracies in mapping directly cause "map flickering," which not only seriously affects the search efficiency of subsequent path planning algorithms (such as A*, RRT, or improved Fast-Planner algorithms), but may even lead to drones braking incorrectly, trajectory oscillations, or direct collisions, greatly limiting the drone's ability to fly rapidly in complex environments.

[0004] Furthermore, existing obstacle avoidance systems typically employ fixed probability thresholds and static collision repulsion radii when converting probabilistic maps into physical obstacle avoidance boundaries for UAVs. This fails to effectively incorporate the UAV's current kinematic state (such as long braking distances due to high-speed flight) and the uncertainties in the underlying mapping quality. While this static obstacle avoidance mechanism may be applicable at low speeds, it often fails to provide sufficient safety margins when the UAV is moving at high speeds or when the environmental map quality deteriorates, leading to obstacle avoidance failure.

[0005] Therefore, there is an urgent need for a method that can systematically and dynamically process sensor noise and introduce confidence transfer and kinematic compensation into the mapping update mechanism and physical obstacle avoidance boundary construction, so as to fill the gap in the existing technology for high-speed obstacle avoidance mapping and safe closed-loop control of UAVs, and provide core technical support for the intelligent and high-speed application of UAVs. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in UAV mapping and obstacle avoidance algorithms, such as high sensor noise interference, map flickering due to fixed probability updates, and the inability of static obstacle avoidance boundaries to adapt to high-speed flight and mapping uncertainties. This invention proposes a Bayesian fusion mapping method and system based on confidence propagation. This method scientifically eliminates sensor noise by constructing a cascaded Bayesian fusion framework from instantaneous confidence calculation and spatial diffusion filtering to dynamic gain updates. Furthermore, it combines the real-time kinematic state of the UAV with the mapping quality to dynamically extract the physical occupancy state and construct a nonlinear repulsion radius. This provides accurate environmental maps and reliable trajectory collision detection boundaries for high-speed obstacle avoidance in complex environments, improving the safety, stability, and path planning efficiency of UAV flight.

[0007] To achieve the above objectives, the technical solution provided by this invention is:

[0008] A Bayesian fusion mapping method based on confidence propagation includes the following steps:

[0009] Step 1: Based on the depth measurement data acquired by the UAV's onboard sensors, calculate the instantaneous confidence coefficient of the current measurement point, and take the corresponding grid in the original grid map as the current grid to generate the original logarithmic observation probability of the current grid.

[0010] Step 2: Use the instantaneous confidence coefficient calculated in Step 1 as a spatial weighting factor to perform weighted fusion on the original logarithmic observation probabilities of the current grid and its neighboring grids within the set neighborhood range to obtain the spatially corrected observation value of the current grid.

[0011] Step 3: Based on the difference between the original logarithmic observation probability of the current grid obtained in Step 1 and the spatially corrected observation value of the current grid obtained in Step 2, calculate the spatial consistency score and solve the dynamic update gain accordingly.

[0012] Step 4: Using the spatially corrected observations obtained in Step 2 and the dynamic update gain obtained in Step 3, perform a nonlinear iterative update on the global posterior log probability of each grid cell in the original grid map to obtain the fused mapping result.

[0013] Further, in step 1, the formula for calculating the instantaneous confidence coefficient of the measurement point is:

[0014]

[0015] In the formula, This is the instantaneous confidence coefficient; The measurement point is the Euclidean distance between the point of measurement and the drone; This is the effective maximum range of the sensor; It is the angle between the sensor's line-of-sight vector and the normal vector of the object's surface at the measurement point.

[0016] Further, in step 1, the formula for calculating the original logarithmic observation probability is:

[0017]

[0018] In the formula, For the current grid The original logarithmic observation probability, To correspond to the current grid The instantaneous confidence coefficient of the current measurement point. The reference logarithm occupies the probability constant inherent in the sensor.

[0019] Further, in step 2, the formula for calculating the spatially corrected observation is:

[0020]

[0021] In the formula, For the current grid Spatial correction observations; For the current grid The original logarithmic observation probability; This is the spatial diffusion intensity coefficient; For the current grid The set of neighboring grid cells within the defined neighborhood; For the first The instantaneous confidence coefficient of each neighboring grid cell; For the first The raw logarithmic observation probability of each neighboring grid cell; To prevent constants with a denominator of zero.

[0022] Furthermore, in step 3, the formula for calculating the spatial consistency score is:

[0023]

[0024] In the formula, For the current grid Spatial consistency score, To prevent constants with a denominator of zero;

[0025] The formula for calculating the dynamically updated gain is:

[0026]

[0027] In the formula, For the current grid Dynamically updated gain; This is the preset minimum update rate threshold.

[0028] Furthermore, in step 4, the global posterior log-probability after the nonlinear iterative update is:

[0029]

[0030] In the formula, for Current grid The global posterior log-odds; for Current grid The global posterior logarithmic odds.

[0031] Furthermore, the method also includes:

[0032] Step 5: Convert the global posterior log probability into the physical occupancy state of the current grid required for obstacle avoidance.

[0033]

[0034] in, for Current grid The physical occupancy state, Values Time determination Current grid Occupied by physical obstacles Values Time determination Current grid Not occupied by physical obstacles; for Current grid The global posterior log-odds; Anisotropic velocity adaptive determination threshold, , Set a preset static base occupancy threshold; This represents the maximum downward movement of the threshold. for Timing-rate activation function , for The drone's horizontal ground speed at all times The maximum safe speed limit allowed for drones, This represents the hyperbolic velocity gain coefficient. for Time-of-flight heading alignment function. , for The drone's velocity vector and the drone's centroid pointing to the current grid are constantly changing. The spatial angle between position vectors; For course focus index;

[0035] Step 6, if If the drone continues along the currently planned path, then proceed to step 7;

[0036] Step 7: Calculate the feedforward braking compensation distance of the UAV, and generate the single-point dynamic expansion repulsion radius based on the current physical occupancy state of the grid and the dynamic update gain.

[0037] Step 8: If the distance between the center of the UAV body and the current grid is less than the single-point dynamic expansion repulsion radius obtained in Step 7, the UAV continues to travel along the current planned path; otherwise, proceed to Step 9.

[0038] Step 9: With the goal of avoiding the spherical region centered on the current grid center and with a single-point dynamic expansion repulsion radius as the radius, replan the path for the UAV and continue to travel along the new planned path.

[0039] Furthermore, in step 7, the feedforward braking compensation distance for:

[0040]

[0041] In the formula, This is the maximum reverse braking acceleration of the drone. This is the aerodynamic damping attenuation coefficient.

[0042] Furthermore, in step 7, the single-point dynamic expansion repulsion radius... for:

[0043]

[0044] In the formula: This refers to the basic physical envelope radius of the drone; This is the feedforward braking compensation distance; It is a dimensionless spatial uncertainty multiplier. , Amplify the weights for uncertainty. For the current grid Dynamically updated gain; To prevent real numbers with a denominator of zero.

[0045] This invention also proposes a system that applies the Bayesian fusion mapping method based on confidence propagation as described above, comprising:

[0046] The raw logarithmic observation probability generation module is used to calculate the instantaneous confidence coefficient of the current measurement point based on the depth measurement data obtained by the UAV's onboard sensor, and take the corresponding grid of the current measurement point in the original grid map as the current grid to generate the raw logarithmic observation probability of the current grid.

[0047] The spatial correction observation acquisition module is used to use the instantaneous confidence coefficient as a spatial weighting factor to perform weighted fusion of the original logarithmic observation probabilities of the current grid and its neighboring grids within a set neighborhood range to obtain the spatial correction observation value of the current grid.

[0048] The dynamic update gain solution module is used to calculate the spatial consistency score based on the difference between the original logarithmic observation probability and the spatially corrected observation value of the current raster, and to solve for the dynamic update gain accordingly.

[0049] The fusion mapping module is used for spatial correction of observations and dynamic gain updates. It performs nonlinear iterative updates on the global posterior log probability of each grid cell in the original raster map to obtain the fusion mapping results.

[0050] Beneficial effects:

[0051] 1) This invention constructs an instantaneous confidence evaluation model based on measurement geometry and dynamically adjusts the posterior update gain using spatial diffusion filtering and spatial consistency scores, achieving cascaded adaptive fusion of sensor observation data. This method completely changes the static assumptions of "fixed observation probability" and "independent grid updates" in traditional Bayesian mapping. It can isolate and filter discrete noise and observational abrupt changes caused by sensor physical limitations or environmental light and shadow interference in real time, thereby significantly suppressing "map flickering" and obstacle ghosting phenomena in complex and unknown environments. This greatly improves the accuracy and convergence stability of 3D environment modeling, providing a high-confidence perception base for high-speed autonomous flight of UAVs.

[0052] 2) The dynamic safety field construction method based on composite kinematics and uncertainty provided by this invention deeply couples the quality of the underlying probabilistic mapping with the upper-level engineering obstacle avoidance strategy. By introducing an anisotropic velocity adaptive extraction mechanism and a multiplicative coupling nonlinear repulsion radius formula, a dynamic closed-loop transformation from "probabilistic state" to "physical safety boundary" is achieved. This method overcomes the limitation of traditional static expansion radius, which cannot simultaneously account for high-speed maneuverability and mapping quality degradation. It can exponentially expand or smoothly contract the safety boundary in a feedforward manner based on the aircraft's real-time ground speed, heading field of view, and the convergence state of the local map. This greatly enhances the UAV's high-speed collision avoidance sensitivity under strong sensor noise interference, while ensuring its physical passability in known narrow spaces. This provides key engineering technology support for building a UAV obstacle avoidance system with self-state awareness and extreme environment adaptability. Attached Figure Description

[0053] Figure 1 This is the overall flowchart of the present invention.

[0054] Figure 2 This is a flowchart of the physical occupancy state extraction and dynamic security field construction process. Detailed Implementation

[0055] The present invention will now be described in detail with reference to the accompanying drawings and exemplary embodiments thereof. It should be noted that the following detailed description of the present invention is for illustrative purposes only and is not intended to limit the scope of the invention.

[0056] Reference Figure 1 The overall process of the Bayesian fusion mapping method based on confidence propagation, as an exemplary embodiment of the present invention, includes the following steps:

[0057] Step S1, Instantaneous confidence coefficient and original observation value calculation: The airborne sensor acquires environmental depth measurement data, and the system calculates the instantaneous confidence coefficient of the current measurement point in real time. The measurement point is then mapped to the corresponding grid cell in the original grid map as the current grid cell. And combined with the baseline probability constant Generate the original logarithmic observation probability .

[0058] Step S2, Spatial diffusion filtering based on confidence weights: The instantaneous confidence coefficients output in step S1 are used as spatial weighting factors for the current raster. and its neighboring areas The original logarithmic observation probabilities within the range are weighted and fused, and the calculation formula is as follows: Thus, spatially corrected observations are obtained. This step utilizes spatial correlation to effectively suppress random sensor noise.

[0059] Step S3, Spatial Consistency and Dynamic Update Gain Calculation: The system compares the original logarithmic observation probabilities from Step S1 with the spatially corrected observations from Step S2 to calculate the spatial consistency score. Based on this score, the formula is used... Solve for the dynamic update gain to adaptively adjust the step size of map updates.

[0060] Step S4, cascaded update of the raster map: using the spatially corrected observations from step S2 and the dynamic update gain from step S3, a nonlinear iterative update of the global posterior log odds is performed. This process achieves stable convergence of the map state over time.

[0061] Reference Figure 2 The physical occupancy state extraction and dynamic security field construction logic flow of an exemplary embodiment of the present invention includes the following specific execution steps:

[0062] Step S1, input the global posterior logarithmic probability, real-time ground speed, and heading: the system collects the output of step S4 in real time. and the current horizontal ground speed of the drone Angle with heading As input parameters.

[0063] Step S2, calculate the velocity activation function and the field-of-view heading alignment function: The logic unit calculates the velocity activation function. To perceive dynamic risks; and simultaneously calculate the field-of-view heading alignment function. This allows for focused attention on obstacles ahead of the flight path.

[0064] Step S3, solve for the anisotropic velocity adaptive decision threshold: based on the basic threshold With the maximum downward range Solution This threshold decreases nonlinearly with increasing flight speed, thereby improving obstacle avoidance sensitivity at high speeds.

[0065] Step S4: Perform a conditional judgment on whether the occupancy probability is greater than the judgment threshold to obtain the physical occupancy status of the grid.

[0066] .

[0067] Step S5, State Extraction and Classification: If the grid is determined to be occupied by a physical obstacle ( ), triggering a safety field calculation and executing step S6; if not, it is determined that the grid is not occupied by a physical obstacle ( The drone continued its journey.

[0068] Step S6, Input the basic physical envelope and feedforward braking compensation distance: The system inputs the physical envelope radius. And calculate the feedforward braking compensation distance. .

[0069] Step S7, introduce spatial uncertainty multipliers for exponential penalty amplification: calculate multipliers using map update gain. When the map has not fully converged or is flickering, this multiplier will significantly amplify the obstacle avoidance boundary.

[0070] Step S8: Generate and output the dynamic expansion repulsion radius of a single obstacle point: using the formula... Generate the final exclusion boundary and use it as the sphere exclusion constraint for path replanning.

[0071] The invention will be further illustrated below with examples.

[0072] A quadcopter UAV needs to perform rapid reconnaissance missions in an area filled with unknown forests and building debris. The lighting conditions in the flight area are complex, and the depth camera is prone to flight point noise. The requirement is to dynamically suppress mapping noise and generate safe obstacle avoidance boundaries based on the method of this invention.

[0073] In this embodiment, the core parameters and initial state are set as follows:

[0074] Sensor maximum range The basic occupancy probability constant .

[0075] UAV physical envelope radius Maximum safe speed limit Maximum braking acceleration .

[0076] Static base occupancy threshold Maximum downward range .

[0077] Scenario A: Approaching a potential obstacle directly in front at high speed (dynamic threshold and repulsion radius expansion)

[0078] At this time, the drone's flight speed Directly in front (spatial angle) )distance The sensor returns an observation point. The line of sight is perpendicular to the object's surface. ).

[0079] Bottom-level mapping derivation:

[0080] Instantaneous confidence .

[0081] Assuming the area was previously affected by noise, the dynamic gain is derived from the spatial consistency score. (In a semi-converged state).

[0082] After step S4 iteration, the standard occupancy probability of the grid converges to If the probability of this point is not met in a traditional static algorithm (where the fixed threshold is usually 0.65), it will be considered a "safe no-go zone," causing the drone to crash directly at a speed of 4 m / s.

[0083] This invention relates to engineering implementation calculations and decision-making:

[0084] (1) Velocity activation function: .

[0085] (2) Heading alignment function: .

[0086] (3) Adaptive threshold calculation: .

[0087] (4) Judgment logic: Current occupancy probability Established. The system proactively extracts this point as a "physical obstacle".

[0088] (5) Dynamic security field generation:

[0089] Feedforward braking compensation distance: .

[0090] Uncertain multipliers: .

[0091] Final single-point repulsion radius: .

[0092] Decision result: The system outputs a value of up to [value missing] to the trajectory planner. The repulsion field means that if the distance between the center of the current UAV body and the center of the grid is less than 5.5m, the upper-level path planning system will force the high-speed flying UAV to change lanes in advance, thus avoiding the "inability to brake" accident caused by the incomplete convergence of the mapping.

[0093] Scenario B: Slow-speed travel through a narrow corridor (to prevent accidental braking due to noise)

[0094] The drone entered the narrow corridor and slowed down. At this time, the side ( The reflection creates a false noise.

[0095] This invention's engineering implementation calculations and decision-making:

[0096] (1) Velocity activation function: .

[0097] (2) Heading alignment function: .

[0098] (3) Adaptive threshold calculation: (Maintain a high threshold).

[0099] (4) Dynamic security field generation:

[0100] Feedforward braking compensation distance Extremely small.

[0101] Final single-point repulsion radius: .

[0102] Decision result: The system maintained a very small and close-fitting repulsion radius, eliminated lateral noise interference, and ensured that the drone could pass smoothly through narrow spaces without in-situ oscillation or accidental braking.

[0103] Throughout the entire process described above, the assessment of "measurement confidence level" was achieved. "State-consistent adaptive graph building" "Physical state feedforward determination" The autonomous intelligent operation of "nonlinear repulsion field output" significantly improves the robustness and intelligence of UAV obstacle avoidance missions in complex environments.

[0104] This invention also proposes a system for a Bayesian fusion mapping method based on confidence propagation, comprising:

[0105] The raw logarithmic observation probability generation module is used to calculate the instantaneous confidence coefficient of the current measurement point based on the depth measurement data obtained by the UAV's onboard sensor, and take the corresponding grid of the current measurement point in the original grid map as the current grid to generate the raw logarithmic observation probability of the current grid.

[0106] The spatial correction observation acquisition module is used to use the instantaneous confidence coefficient as a spatial weighting factor to perform weighted fusion of the original logarithmic observation probabilities of the current grid and its neighboring grids within a set neighborhood range to obtain the spatial correction observation value of the current grid.

[0107] The dynamic update gain solution module is used to calculate the spatial consistency score based on the difference between the original logarithmic observation probability and the spatially corrected observation value of the current raster, and to solve for the dynamic update gain accordingly.

[0108] The fusion mapping module is used for spatial correction of observations and dynamic gain updates. It performs nonlinear iterative updates on the global posterior log probability of each grid cell in the original raster map to obtain the fusion mapping results.

[0109] Finally, it should be noted that the above examples are only for clearly illustrating the technical solutions and effects of the present invention and should not be construed as limiting the present invention. Under the framework of the present invention, any technical solutions obtained by adaptively adjusting or combining parameters and strategies for different mission scenarios, flight platforms, and constraints should be considered to fall within the protection scope of the present invention.

Claims

1. A confidence propagation based Bayesian fusion mapping method, characterized in that, Includes the following steps: Step 1: Based on the depth measurement data acquired by the UAV's onboard sensors, calculate the instantaneous confidence coefficient of the current measurement point, and take the corresponding grid in the original grid map as the current grid to generate the original logarithmic observation probability of the current grid. Step 2: Use the instantaneous confidence coefficient calculated in Step 1 as a spatial weighting factor to perform weighted fusion on the original logarithmic observation probabilities of the current grid and its neighboring grids within the set neighborhood range to obtain the spatially corrected observation value of the current grid. Step 3: Based on the difference between the original logarithmic observation probability of the current grid obtained in Step 1 and the spatially corrected observation value of the current grid obtained in Step 2, calculate the spatial consistency score and solve the dynamic update gain accordingly. Step 4: Using the spatially corrected observations obtained in Step 2 and the dynamic update gain obtained in Step 3, perform a nonlinear iterative update on the global posterior log probability of each grid cell in the original grid map to obtain the fused mapping result.

2. The confidence-conduct-based Bayesian fusion mapping method according to claim 1, wherein, In step 1, the formula for calculating the instantaneous confidence coefficient of the measurement point is: ， wherein is the instantaneous confidence coefficient; is the Euclidean distance between the measurement point and the UAV; is the effective maximum range of the sensor; is the angle between the sensor observation line-of-sight vector and the object surface normal vector at the measurement point.

3. The confidence-conduct-based Bayesian fusion mapping method according to claim 2, characterized in that, In step 1, the formula for calculating the original logarithmic observation probability is: ， where, is the original log-observation probability of the current grid cell , is the instantaneous confidence coefficient of the current measurement point corresponding to the current grid cell , is the sensor-specific reference log-occupancy probability constant.

4. The confidence-conduct-based Bayesian fusion mapping method according to claim 3, characterized in that, In step 2, the formula for calculating the spatially corrected observation is: ， wherein is the spatially corrected observation value of the current grid ; is the original log-observation probability of the current grid ; is the spatial diffusion intensity coefficient; is the set of neighbor grids within the set neighborhood range of the current grid ; is the instantaneous confidence coefficient of the th neighbor grid; is the original log-observation probability of the th neighbor grid; is a constant to prevent the denominator from being zero.

5. The confidence-conduct-based Bayesian fusion mapping method according to claim 4, characterized in that, In step 3, the formula for calculating the spatial consistency score is: ， wherein is the spatial consistency score for the current grid is the spatial consistency score for the current grid is a constant to prevent the denominator from being zero; The formula for calculating the dynamically updated gain is: ， In the formula, is the dynamic update gain of the current grid ; is a preset minimum update rate threshold.

6. The confidence-conduct-based Bayesian fusion mapping method according to claim 5, characterized in that, In step 4, the global posterior log-probability after the nonlinear iterative update is: ， In the formula, for Current grid The global posterior log-odds; for Current grid The global posterior logarithmic odds.

7. The Bayesian fusion mapping method based on confidence propagation according to claim 1, characterized in that, Also includes: Step 5: Convert the global posterior log probability into the physical occupancy state of the current grid required for obstacle avoidance. ， in, for Current grid The physical occupancy state, Values Time determination Current grid Occupied by physical obstacles Values Time determination Current grid Not occupied by physical obstacles; for Current grid The global posterior log-odds; Anisotropic velocity adaptive determination threshold, , Set a preset static base occupancy threshold; This represents the maximum downward movement of the threshold. for Timing-rate activation function , for The drone's horizontal ground speed at all times The maximum safe speed limit allowed for drones, This represents the hyperbolic velocity gain coefficient. for Time-of-flight field-of-view alignment function, , for The drone's velocity vector and the drone's centroid pointing to the current grid are constantly changing. The spatial angle between position vectors; For course focus index; Step 6, if If the drone continues along the currently planned path, then proceed to step 7; Step 7: Calculate the feedforward braking compensation distance of the UAV, and generate the single-point dynamic expansion repulsion radius based on the current physical occupancy state of the grid and the dynamic update gain. Step 8: If the distance between the center of the UAV body and the current grid is less than the single-point dynamic expansion repulsion radius obtained in Step 7, the UAV continues to travel along the current planned path; otherwise, proceed to Step 9. Step 9: With the goal of avoiding the spherical region centered on the current grid center and with a single-point dynamic expansion repulsion radius as the radius, replan the path for the UAV and continue to travel along the new planned path.

8. The Bayesian fusion mapping method based on confidence propagation according to claim 7, characterized in that, In step 7, the feedforward braking compensation distance for: ， In the formula, This is the maximum reverse braking acceleration of the drone. This is the aerodynamic damping attenuation coefficient.

9. The Bayesian fusion mapping method based on confidence propagation according to claim 7, characterized in that, In step 7, the single-point dynamic expansion repulsion radius for: ， In the formula: This refers to the basic physical envelope radius of the drone; This is the feedforward braking compensation distance; It is a dimensionless spatial uncertainty multiplier. , Amplify the weights for uncertainty. For the current grid Dynamically updated gain; To prevent real numbers with a denominator of zero.

10. A system applying the Bayesian fusion mapping method based on confidence propagation as described in any one of claims 1 to 9, characterized in that, include: The raw logarithmic observation probability generation module is used to calculate the instantaneous confidence coefficient of the current measurement point based on the depth measurement data obtained by the UAV's onboard sensor, and take the corresponding grid of the current measurement point in the original grid map as the current grid to generate the raw logarithmic observation probability of the current grid. The spatial correction observation acquisition module is used to use the instantaneous confidence coefficient as a spatial weighting factor to perform weighted fusion of the original logarithmic observation probabilities of the current grid and its neighboring grids within a set neighborhood range to obtain the spatial correction observation value of the current grid. The dynamic update gain solution module is used to calculate the spatial consistency score based on the difference between the original logarithmic observation probability and the spatially corrected observation value of the current raster, and to solve for the dynamic update gain accordingly. The fusion mapping module is used for spatial correction of observations and dynamic gain updates. It performs nonlinear iterative updates on the global posterior log probability of each grid cell in the original raster map to obtain the fusion mapping results.

Images