Autonomous obstacle avoidance system of special geophysical prospecting unmanned rotorcraft

Abstract

The invention, which relates to the technical field of the unmanned rotorcraft, discloses an autonomous obstacle avoidance system of a special unmanned rotorcraft for geophysical prospecting, comprising an acquisition and conversion module, an image segmentation module, a detection module and a planning module. According to the invention, the acquisition unit uses a millimeter-wave radar and a binocular vision sensor to acquire data and obtain multi-modal data, and converts radar data into left-eye image coordinates through a coordinate conversion formula, so that multi-source data fusion is realized, and the accuracy and reliability of obstacle detection are improved; a binocular vision image is segmented by using a mean shift algorithm, the image is divided into different areas with similar features, a global energy function is constructed in combination with image segmentation information, the energy function is minimized through multidirectional matching cost aggregation, and an optimal disparity map is selected, so that disparity map noise can be effectively removed, edges are more continuous, and the image quality is improved. More accurate depth information is provided, and a good foundation is laid for subsequent obstacle detection and positioning.

Description

Technical Field

[0001] The present invention relates to the technical field of rotary-wing UAVs, and in particular to an autonomous obstacle avoidance system of a rotary-wing UAV dedicated to geophysical exploration. Background Art

[0002] Rotary-wing UAVs dedicated to geophysical exploration are mainly used for flight operations in front of and behind mountains below 4,000 meters above sea level, where fixed-wing aircraft cannot lower their flight altitude. Terrain-following flight is required to obtain high-precision and high-resolution magnetic field information. Three-dimensional precise navigation technology is an effective technical support for terrain-following flight. When rotary-wing UAVs conduct low-altitude geophysical exploration flights, near-surface obstacles such as high-voltage power towers and signal towers pose a great danger to the flight. Autonomous obstacle avoidance technology can effectively reduce this danger. Robust adaptive control of the suspension for crosswind interference can improve the safety of flight operations.

[0003] At present, existing navigation technology is difficult to meet the needs of three-dimensional precise obstacle avoidance and is prone to collision risks. In addition, the disparity maps generated by traditional stereo matching algorithms have problems such as high noise and discontinuous edges. In particular, mismatches are prone to occur at the boundaries of objects or in occluded areas, resulting in inaccurate depth information. This will cause the drone to misjudge the distance or shape of obstacles, affecting the reliability of obstacle avoidance path planning.

[0004] Therefore, an autonomous obstacle avoidance system for a geophysical exploration-specific rotor UAV is proposed to solve the above problems. Summary of the Invention

[0005] The main purpose of the present invention is to provide an autonomous obstacle avoidance system for a geophysical exploration-specific rotor UAV to solve the problems raised in the above background.

[0006] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is as follows: an autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV, comprising an acquisition and conversion module, an image segmentation module, a detection module, and a planning module;

[0007] The acquisition and conversion module is used to collect data through the millimeter wave radar and the binocular vision sensor, and convert the obstacle position points detected by the millimeter wave radar into left-eye image coordinates;

[0008] The image segmentation module uses the mean shift algorithm to segment the image acquired by binocular vision, calculates the mean shift vector of the pixel points, and calculates the matching cost between pixels at different disparities. It also constructs a global energy function based on the image segmentation information, aggregates the matching costs in multiple directions, minimizes the energy function, compares the energy values ​​corresponding to all disparity maps, and selects the disparity map with the smallest energy value as the final matching disparity result.

[0009] The detection module uses a region growing image segmentation algorithm to determine candidate obstacle regions based on the converted image coordinate positions and the depth and color information of the disparity map, and detects the candidate regions to determine whether they are real obstacles. If they are obstacles, three-dimensional coordinate conversion is performed;

[0010] The planning module is used to perform global path planning based on the three-dimensional coordinates in the NED coordinate system using the A algorithm, and to control the flight route of the drone using the global path.

[0011] Preferably, the acquisition and conversion module includes an acquisition unit and a conversion unit;

[0012] The acquisition unit is used to obtain object depth and angle information through millimeter wave radar, and to obtain color images and distance information through binocular vision sensors.

[0013] Preferably, the conversion unit is used to convert the obstacle position point detected by the millimeter wave radar into the left eye image coordinates, and the formula is as follows:

[0014]

[0015] Among them, K represents the camera matrix, u and v represent the horizontal and vertical coordinates of the obstacle in the left eye image coordinate system, (R|t) represents the camera external matrix, and x i ,y i , z i Represents the x, y, and z coordinate values ​​of the obstacle in the radar coordinate system.

[0016] Preferably, the image segmentation module includes a segmentation mean unit, a matching function unit and an aggregation disparity unit.

[0017] Preferably, the segmentation mean unit is used to segment the image acquired by binocular vision using a mean shift algorithm and calculate the mean shift vector of the pixel points. The calculation formula is as follows:

[0018]

[0019] Among them, M(x a ) represents pixel x a The mean shift vector, x a Represents a pixel in the image, a represents the index of the pixel, and k represents the pixel x a Neighborhood N(x a ) the number of pixels, N(x a ) represents pixel x a Neighborhood, xj represents pixel x a Neighborhood N(x a ) other pixels within, Represents pixel xa The new position after an iterative update.

[0020] Preferably, the matching function unit is used to calculate the matching cost between pixels under different disparities and construct a global energy function in combination with image segmentation information. The calculation steps are as follows:

[0021] Step 1: Calculate the matching cost between pixels at different disparities based on the sum of absolute differences method. The calculation formula is as follows:

[0022]

[0023] Among them, C(p,d) represents the matching cost of pixel p under disparity d, I1 and I2 represent the left and right view images respectively, and N p represents the neighborhood of pixel p, d represents disparity, p represents a pixel in the image, and q represents a pixel in the neighborhood of pixel p;

[0024] Step 2: Construct a global energy function based on image segmentation information. The formula is as follows:

[0025] E(D)=∑ p C(p,D P )+γ1∑ p,q∈N4(P) P1·T[|D p -D q |=1]+γ2∑ p,q∈N4(P) P2·T[|D p -D q |>1];

[0026] Among them, E(D) represents the global energy function, D represents the disparity map, p represents the pixel in the image, and D P represents the disparity of pixel p, C(p,D P ) represents the pixel p in the disparity D P The matching cost under , γ1 and γ2 represent weight coefficients, q represents the pixel in the 4-neighborhood N4(P) of pixel p, N4(P) represents the set of four directly adjacent pixels above, below, left, and right of pixel p, P1 and P2 represent smoothing penalty parameters, and P1 is less than P2, T[|D p -D q |=1] and T[|D p -D q |>1] represents the indicator function.

[0027] Preferably, the aggregation disparity unit is used to select multiple directions for cost aggregation, including horizontal, vertical and diagonal directions; in each direction, a dynamic programming method is used to perform aggregation calculation of the matching cost, combining the aggregation cost of its neighboring pixels, its own matching cost and a smoothing penalty parameter in this direction, where the aggregation cost represents the absolute difference and the calculated matching cost, and iteratively calculates the aggregation cost of the current pixel under different disparities. After completing the aggregation calculation in all directions, the aggregation costs in each direction are summed to obtain the final aggregate matching cost of each pixel;

[0028] Substitute the final aggregate matching cost into the formula of step 2 of the matching function unit, compare the energy values ​​corresponding to all disparity maps, and select the disparity map with the smallest energy value as the final matching disparity result.

[0029] Preferably, the detection module includes a candidate unit and a determination unit;

[0030] The candidate unit adopts a region growing image segmentation algorithm based on the image coordinate position converted by the conversion unit and the disparity map with the minimum energy value, combines the depth information of the disparity map and the color information of the color image to determine the similarity criterion, selects a representative pixel point near the converted image coordinate position as a seed point, and starts from the seed point, continuously merges the neighboring pixels that meet the similarity criterion into the current region, and gradually expands the region range until no neighboring pixels that meet the conditions can be found, thereby determining the obstacle candidate region.

[0031] Preferably, the determination unit is used to judge the candidate obstacle area through a support vector machine. If it is determined to be an obstacle area, the left-eye image coordinates (u, v) of the domain are converted into three-dimensional coordinates in the NED coordinate system in combination with the camera parameters and the conversion relationship between the world coordinate system and the camera coordinate system. If it is determined not to be an obstacle area, the obstacle area information is discarded.

[0032] Preferably, the planning module includes a planning unit and an execution unit;

[0033] The planning unit performs global path planning using the A algorithm based on the three-dimensional coordinates in the NED coordinate system;

[0034] The execution unit controls the flight route of the UAV based on the planned global path.

[0035] The present invention has the following beneficial effects:

[0036] 1. In the present invention, the acquisition unit uses millimeter-wave radar and binocular vision sensors to collect data and obtain multimodal data. The radar data is converted into left-eye image coordinates through a coordinate conversion formula to achieve multi-source data fusion and improve the accuracy and reliability of obstacle detection. In the segmentation mean unit, the mean shift algorithm is used to segment the binocular vision image. The mean shift algorithm continuously calculates the mean shift vector of pixel points to move and converge the pixel points toward the center of the density distribution in their neighborhood, dividing the image into different areas with similar features. The global energy function is constructed based on the image segmentation information. The energy function is minimized through multi-directional matching cost aggregation, and the optimal disparity map is selected. This can effectively remove disparity map noise, make edges more continuous, provide more accurate depth information, and lay a good foundation for subsequent obstacle detection and positioning.

[0037] 2. In the present invention, the candidate unit uses a region growing algorithm based on image coordinates and the optimal disparity map, and combines depth and color information to determine the similarity criterion to select seed points and expand the area. Depth similarity is determined based on the disparity range of adjacent pixels, and color similarity is based on the color difference threshold. It can effectively screen out areas that may contain obstacles from complex backgrounds. The determination unit uses a support vector machine to determine the candidate area. If it is determined to be an obstacle area, it can accurately convert the image coordinates into three-dimensional coordinates in the NED coordinate system, providing precise location information for subsequent planning.

[0038] 3. In the present invention, the planning unit uses the A algorithm to perform global path planning based on the three-dimensional coordinates of the obstacles, and the execution unit controls the flight route of the UAV according to the planned path to ensure that the UAV flies along the planned safe path, achieves effective obstacle avoidance, and improves the safety and reliability of the UAV flying in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of an autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to the present invention;

[0040] Figure 2 This is a flow chart of the determination unit of the autonomous obstacle avoidance system of a geophysical exploration-specific rotor UAV of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Implementation 1: Please refer to Figure 1 and Figure 2,The present invention provides a technical solution: an autonomous obstacle avoidance system for a geophysical exploration-specific rotor UAV, comprising an acquisition and conversion module, an image segmentation module, a detection module and a planning module;

[0043] The acquisition and conversion module is used to collect data through the millimeter wave radar and the binocular vision sensor, and convert the obstacle position points detected by the millimeter wave radar into left-eye image coordinates;

[0044] The image segmentation module uses the mean shift algorithm to segment the image acquired by binocular vision, calculates the mean shift vector of the pixel points, and calculates the matching cost between pixels at different disparities. It also constructs a global energy function based on the image segmentation information, aggregates the matching costs in multiple directions, minimizes the energy function, compares the energy values ​​corresponding to all disparity maps, and selects the disparity map with the smallest energy value as the final matching disparity result.

[0045] The detection module uses a region growing image segmentation algorithm to determine candidate obstacle regions based on the converted image coordinate positions and the depth and color information of the disparity map, and detects the candidate regions to determine whether they are real obstacles. If they are obstacles, three-dimensional coordinate conversion is performed;

[0046] The planning module is used to perform global path planning based on the three-dimensional coordinates in the NED coordinate system using the A algorithm, and to control the flight route of the drone using the global path.

[0047] Preferably, the acquisition and conversion module includes an acquisition unit and a conversion unit;

[0048] The acquisition unit is used to obtain object depth and angle information through millimeter wave radar, and to obtain color images and distance information through binocular vision sensors.

[0049] Preferably, the conversion unit is used to convert the obstacle position point detected by the millimeter wave radar into the left eye image coordinates, and the formula is as follows:

[0050]

[0051] Among them, K represents the camera matrix, u and v represent the horizontal and vertical coordinates of the obstacle in the left eye image coordinate system, (R|t) represents the camera external matrix, and x i ,y i , z i Represents the x, y, and z coordinate values ​​of the obstacle in the radar coordinate system.

[0052] Preferably, the image segmentation module includes a segmentation mean unit, a matching function unit and an aggregation disparity unit.

[0053] Preferably, the segmentation mean unit is used to segment the image acquired by binocular vision using a mean shift algorithm and calculate the mean shift vector of the pixel points. The calculation formula is as follows:

[0054]

[0055] Among them, M(x a ) represents pixel x a The mean shift vector, x a Represents a pixel in the image, a represents the index of the pixel, and k represents the pixel x a Neighborhood N(x a ) the number of pixels, N(x a ) represents pixel x a Neighborhood, xj represents pixel x a Neighborhood N(x a ) other pixels within, Represents pixel x a The new position after an iterative update.

[0056] Preferably, the matching function unit is used to calculate the matching cost between pixels under different disparities and construct a global energy function in combination with image segmentation information. The calculation steps are as follows:

[0057] Step 1: Calculate the matching cost between pixels at different disparities based on the sum of absolute differences method. The calculation formula is as follows:

[0058]

[0059] Among them, C(p,d) represents the matching cost of pixel p under disparity d, I1 and I2 represent the left and right view images respectively, and N p represents the neighborhood of pixel p, d represents disparity, p represents a pixel in the image, and q represents a pixel in the neighborhood of pixel p;

[0060] Step 2: Construct a global energy function based on image segmentation information. The formula is as follows:

[0061] E(D)=∑ p C(p,D P )+γ1∑ p,q∈N4(P) P1·T[|D p -D q |=1]+γ2∑ p,q∈N4(P) P2·T[|D p -D q |>1];

[0062] Among them, E(D) represents the global energy function, D represents the disparity map, p represents the pixel in the image, and DP represents the disparity of pixel p, C(p,D P ) represents the pixel p in the disparity D P The matching cost under , γ1 and γ2 represent weight coefficients, q represents the pixel in the 4-neighborhood N4(P) of pixel p, N4(P) represents the set of four directly adjacent pixels above, below, left, and right of pixel p, P1 and P2 represent smoothing penalty parameters, and P1 is less than P2, T[|D p -D q |=1] and T[|D p -D q |>1] represents the indicator function.

[0063] Preferably, the aggregation disparity unit is used to select multiple directions for cost aggregation, including horizontal, vertical and diagonal directions; in each direction, a dynamic programming method is used to perform aggregation calculation of the matching cost, combining the aggregation cost of its neighboring pixels, its own matching cost and a smoothing penalty parameter in this direction, where the aggregation cost represents the absolute difference and the calculated matching cost, and iteratively calculates the aggregation cost of the current pixel under different disparities. After completing the aggregation calculation in all directions, the aggregation costs in each direction are summed to obtain the final aggregate matching cost of each pixel;

[0064] Substitute the final aggregate matching cost into the formula of step 2 of the matching function unit, compare the energy values ​​corresponding to all disparity maps, and select the disparity map with the smallest energy value as the final matching disparity result.

[0065] In this embodiment, millimeter-wave radar and binocular vision sensors are used to collect data, obtain multimodal data, and convert radar data into left-eye image coordinates through a coordinate conversion formula to achieve multi-source data fusion. The mean shift algorithm is used to segment the binocular vision image, and the global energy function is constructed in combination with the image segmentation information. The energy function is minimized through multi-directional matching cost aggregation, and the optimal disparity map is selected.

[0066] In this embodiment, the acquisition unit uses millimeter-wave radar and binocular vision sensors to collect data, obtain multimodal data, and converts the radar data into left-eye image coordinates through a coordinate conversion formula to achieve multi-source data fusion. By complementing the advantages of different sensors, the problems of missed and false detection of obstacles in complex environments can be reduced, and the accuracy and reliability of obstacle detection can be improved. In the segmentation mean unit, the mean shift algorithm is used to segment the binocular vision image. The mean shift algorithm continuously calculates the mean shift vector of the pixel points, so that the pixel points move toward the center of the density distribution in their neighborhood and converge, dividing the image into different areas with similar features. The global energy function is constructed in combination with the image segmentation information, and the energy function is minimized through multi-directional matching cost aggregation. The optimal disparity map is selected, which can effectively remove the noise of the disparity map, make the edges more continuous, provide more accurate depth information, and lay a good foundation for subsequent obstacle detection and positioning.

[0067] Implementation 2: Please refer to Figure 1 and Figure 2 , the present invention provides a technical solution: preferably, the detection module includes a candidate unit and a determination unit;

[0068] The candidate unit adopts a region growing image segmentation algorithm based on the image coordinate position converted by the conversion unit and the disparity map with the minimum energy value, combines the depth information of the disparity map and the color information of the color image to determine the similarity criterion, selects a representative pixel point near the converted image coordinate position as a seed point, and starts from the seed point, continuously merges the neighboring pixels that meet the similarity criterion into the current region, and gradually expands the region range until no neighboring pixels that meet the conditions can be found, thereby determining the obstacle candidate region.

[0069] Specifically, based on depth similarity, adjacent pixels are considered similar if their disparity is within a certain range; based on color similarity, adjacent pixels are considered similar if their color difference is less than a certain threshold.

[0070] Preferably, the determination unit is used to judge the candidate obstacle area through a support vector machine. If it is determined to be an obstacle area, the left-eye image coordinates (u, v) of the domain are converted into three-dimensional coordinates in the NED coordinate system in combination with the camera parameters and the conversion relationship between the world coordinate system and the camera coordinate system. If it is determined not to be an obstacle area, the obstacle area information is discarded.

[0071] Preferably, the planning module includes a planning unit and an execution unit;

[0072] The planning unit performs global path planning using the A algorithm based on the three-dimensional coordinates in the NED coordinate system;

[0073] The execution unit controls the flight route of the UAV based on the planned global path.

[0074] In this embodiment, the candidate unit uses a region growing algorithm based on image coordinates and an optimal disparity map, combining depth and color information to determine similarity criteria to select seed points and expand regions. Depth similarity is determined based on the disparity range of adjacent pixels, and color similarity is based on a color difference threshold. This effectively screens out areas that may contain obstacles from complex backgrounds. The determination unit uses a support vector machine to determine candidate regions. If an obstacle area is determined, the image coordinates are accurately converted into three-dimensional coordinates in the NED coordinate system, providing precise location information for subsequent planning. The planning unit uses the A algorithm for global path planning based on the obstacle's three-dimensional coordinates. The execution unit controls the drone's flight path based on the planned path, ensuring that the drone flies along a planned safe path, achieving effective obstacle avoidance and improving the safety and reliability of drone flight in complex environments.

[0075] In the present invention, an autonomous obstacle avoidance system for a geophysical exploration-specific rotor UAV is provided. The acquisition unit utilizes millimeter-wave radar and binocular vision sensors to collect data, obtains multimodal data, and converts the radar data into left-eye image coordinates through a coordinate conversion formula to achieve multi-source data fusion. By complementing the advantages of different sensors, the problems of missed and false detection of obstacles in complex environments can be reduced, and the accuracy and reliability of obstacle detection can be improved. In the segmentation mean unit, a mean shift algorithm is used to segment the binocular vision image. The mean shift algorithm continuously calculates the mean shift vector of pixel points, so that the pixel points move toward the center of the density distribution in their neighborhood and converge, and divides the image into different areas with similar features. A global energy function is constructed in combination with image segmentation information, and the energy function is minimized through multi-directional matching cost aggregation. The optimal disparity map is selected, which can effectively remove disparity map noise, make edges more continuous, provide more accurate depth information, and lay a good foundation for subsequent obstacle detection and positioning.

[0076] Based on image coordinates and the optimal disparity map, the candidate unit uses a region growing algorithm, combining depth and color information to determine similarity criteria to select seed points and expand regions. Depth similarity is determined based on the disparity range of adjacent pixels, and color similarity is based on a color difference threshold. This effectively screens out areas that may contain obstacles from complex backgrounds. The determination unit uses a support vector machine to determine candidate regions. If an obstacle area is determined, the image coordinates are accurately converted into three-dimensional coordinates in the NED coordinate system, providing precise location information for subsequent planning. The planning unit uses the A algorithm for global path planning based on the obstacle's three-dimensional coordinates. The execution unit controls the drone's flight path based on the planned path, ensuring that the drone flies along a planned safe path, achieving effective obstacle avoidance and improving the safety and reliability of drone flights in complex environments.

[0077] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0078] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV, characterized in that: It includes acquisition and conversion module, image segmentation module, detection module and planning module; The acquisition and conversion module is used to collect data through the millimeter wave radar and the binocular vision sensor, and convert the obstacle position points detected by the millimeter wave radar into left-eye image coordinates; The image segmentation module uses the mean shift algorithm to segment the image acquired by binocular vision, calculates the mean shift vector of the pixel points, and calculates the matching cost between pixels at different disparities. It also constructs a global energy function based on the image segmentation information, aggregates the matching costs in multiple directions, minimizes the energy function, compares the energy values ​​corresponding to all disparity maps, and selects the disparity map with the smallest energy value as the final matching disparity result. The detection module uses a region growing image segmentation algorithm to determine candidate obstacle regions based on the converted image coordinate positions and the depth and color information of the disparity map, and detects the candidate regions to determine whether they are real obstacles. If they are obstacles, three-dimensional coordinate conversion is performed; The planning module is used to perform global path planning based on the three-dimensional coordinates in the NED coordinate system using the A algorithm, and to control the flight route of the drone using the global path.

2. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 1, characterized in that: The acquisition and conversion module includes an acquisition unit and a conversion unit; The acquisition unit is used to obtain object depth and angle information through millimeter wave radar, and to obtain color images and distance information through binocular vision sensors.

3. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 2, characterized in that: The conversion unit is used to convert the obstacle position points detected by the millimeter wave radar into the left eye image coordinates. The formula is as follows: Among them, K represents the camera matrix, u and v represent the horizontal and vertical coordinates of the obstacle in the left eye image coordinate system, (R|t) represents the camera external matrix, and x i ,y i , z i Represents the x, y, and z coordinate values ​​of the obstacle in the radar coordinate system.

4. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 1, characterized in that: The image segmentation module includes a segmentation mean unit, a matching function unit and an aggregation disparity unit.

5. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 4, characterized in that: The segmentation mean unit is used to segment the image acquired by binocular vision using the mean shift algorithm and calculate the mean shift vector of the pixel points. The calculation formula is as follows: Among them, M(x a ) represents pixel x a The mean shift vector, x a Represents a pixel in the image, a represents the index of the pixel, and k represents the pixel x a Neighborhood N(x a ) the number of pixels, N(x a ) represents pixel x a Neighborhood, xj represents pixel x a Neighborhood N(x a ) other pixels within, Represents pixel x a The new position after an iterative update.

6. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 5, characterized in that: The matching function unit is used to calculate the matching cost between pixels under different disparities and construct a global energy function in combination with image segmentation information. The calculation steps are as follows: Step 1: Calculate the matching cost between pixels at different disparities based on the sum of absolute differences method. The calculation formula is as follows: Among them, C(p,d) represents the matching cost of pixel p under disparity d, I1 and I2 represent the left and right view images respectively, and N p represents the neighborhood of pixel p, d represents disparity, p represents a pixel in the image, and q represents a pixel in the neighborhood of pixel p; Step 2: Construct a global energy function based on image segmentation information. The formula is as follows: E(D)=∑ p C(p,D P )+γ1∑ p,q∈N4(P) P1·T[|D p -D q |=1]+γ2∑ p,q∈N4(P) P2·T[|D p -D q |>1]; Among them, E(D) represents the global energy function, D represents the disparity map, p represents the pixel in the image, and D P represents the disparity of pixel p, C(p,D P ) represents the pixel p in the disparity D P The matching cost under , γ1 and γ2 represent weight coefficients, q represents the pixel in the 4-neighborhood N4(P) of pixel p, N4(P) represents the set of four directly adjacent pixels above, below, left, and right of pixel p, P1 and P2 represent smoothing penalty parameters, and P1 is less than P2, T[|D p -D q |=1] and T[|D p -D q |>1] represents the indicator function.

7. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 6, characterized in that: The aggregation disparity unit is used to select multiple directions for cost aggregation, including horizontal, vertical and diagonal directions. In each direction, a dynamic programming method is used to perform aggregation calculation of the matching cost. In this direction, the aggregation cost of its neighboring pixels, the matching cost itself and a smoothing penalty parameter are combined. The aggregation cost represents the absolute difference and the calculated matching cost. The aggregation cost of the current pixel under different disparities is iteratively calculated. After the aggregation calculation in all directions is completed, the aggregation costs in each direction are summed to obtain the final aggregate matching cost of each pixel. Substitute the final aggregate matching cost into the formula of step 2 of the matching function unit, compare the energy values ​​corresponding to all disparity maps, and select the disparity map with the smallest energy value as the final matching disparity result.

8. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 1, characterized in that: The detection module includes a candidate unit and a determination unit; The candidate unit adopts a region growing image segmentation algorithm based on the image coordinate position converted by the conversion unit and the disparity map with the minimum energy value, combines the depth information of the disparity map and the color information of the color image to determine the similarity criterion, selects a representative pixel point near the converted image coordinate position as a seed point, and starts from the seed point, continuously merges the neighboring pixels that meet the similarity criterion into the current region, and gradually expands the region range until no neighboring pixels that meet the conditions can be found, thereby determining the obstacle candidate region.

9. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 8, characterized in that: The determination unit is used to judge the candidate obstacle area through a support vector machine. If it is determined to be an obstacle area, the left-eye image coordinates (u, v) are converted into three-dimensional coordinates in the NED coordinate system in combination with the camera parameters and the conversion relationship between the world coordinate system and the camera coordinate system. If it is determined not to be an obstacle area, the obstacle area information is discarded.

10. The autonomous obstacle avoidance system for a geophysical exploration-specific rotary-wing UAV according to claim 1, characterized in that: The planning module includes a planning unit and an execution unit; The planning unit performs global path planning using the A algorithm based on the three-dimensional coordinates in the NED coordinate system; The execution unit controls the flight route of the UAV based on the planned global path.

Images