Unmanned aerial vehicle combined ranging method based on visual image radial / normal motion

Abstract

The application discloses a kind of unmanned aerial vehicle combined ranging method based on visual image radial / normal motion, belong to aircraft combined navigation and computer vision technical field;The present application is aimed at the flight dynamic geometric relation of unmanned aerial vehicle, the flight line-of-sight motion of unmanned aerial vehicle is decomposed along radial and normal, respectively constructs the distance inference module and its three-redundancy voting mechanism based on normal motion, and the distance inference module and its noise suppression mechanism based on radial motion, according to the ingenious transformation of original distance fusion inference into combined navigation problem, with radial motion inference distance module as main system, with normal motion inference distance module as auxiliary system, and the former is corrected, suppresses drift error.The present application helps to significantly reduce the cost of flight platform, and the present application is based on different ranging principles, and the calculation process is relatively simple, can guarantee high refresh rate and long distance accuracy on fast platform MCU, ensure accuracy while ranging scene is not limited.

Description

Technical Field

[0001] This invention belongs to the fields of aircraft integrated navigation and computer vision technology, specifically a UAV integrated ranging method based on visual image radial / normal motion. Background Technology

[0002] The key to swarm systems lies in low-cost offsetting strategies, and reducing the cost of flight platforms has always been a goal pursued by industry professionals. Traditional methods of reducing costs through hardware simplification or reuse are facing significant limitations due to factors such as Moore's Law reaching its limits. Given the current state of information redundancy on flight platforms, replacing hardware devices with software reuse can significantly reduce costs while maintaining ranging functionality.

[0003] For small unmanned aerial vehicles (UAVs), they mainly consist of two parts: the flight platform and the mission payload. The platform includes components such as the outer shell structure, flight control system, recovery system, radio antenna, and servo motors. The most typical mission payload is the electro-optical pod. Among these devices, the electro-optical pod is usually the most expensive single system. In particular, when the electro-optical pod has laser ranging capabilities, on the one hand, the need for expensive precision optical equipment will further increase its price exponentially; on the other hand, adding a laser ranging pod means greater weight, size, and power consumption, placing a greater burden on the entire system and further increasing additional costs. If redundant information on the aircraft platform can be utilized, and based on a standard electro-optical pod without ranging capabilities, deeply integrated with onboard equipment, the line-of-sight / normal motion can be decomposed to achieve system-level software ranging using a combined navigation approach, which can significantly reduce the cost of the flight platform.

[0004] Currently, the main ranging methods used on small and medium-sized UAV platforms are as follows: 1) Laser ranging: A laser is used as the light source to illuminate the target, and the round-trip time or phase difference is measured. The distance from the aircraft to the target is then obtained by combining this with the speed of light in the medium. This method has high accuracy, but it increases the additional hardware cost, and smoke, dust, and raindrops in the atmosphere can interfere with the accuracy. 2) Monocular visual ranging: Target recognition is performed through image matching, and the target distance is estimated based on the principle of similar triangles using the target's prior size and its size in the image. This method has relatively low accuracy and relies on prior information and updating and maintaining a large sample database. It cannot judge or measure targets outside the database. 3) Binocular visual ranging: The same pixel is accurately matched in the left and right camera images, the parallax between the two images is calculated, and the distance between the pixels is estimated by combining the baseline length. This method has higher accuracy than monocular ranging, but the relative accuracy of long-distance measurement will decrease due to the limitation of the baseline length. At the same time, due to the complexity of the algorithm, it is difficult to implement on an MCU, and it is difficult to ensure a high refresh rate on a fast-flying platform, which limits its application. 4) Triangle method estimation: This method approximates the height difference between the aircraft platform and the takeoff point as the height difference with the target, and estimates the distance based on the geometric relationship of triangles. This method has low accuracy and requires the target and takeoff point to be at roughly the same height. The calculated distance often has limited reference value and has obvious limitations in application scenarios.

[0005] In summary, developing system-level ranging capabilities for small flight platforms using a novel approach, achieving line-of-sight / normal motion decomposition through device fusion, and designing a combined filtering module for distance inference can significantly reduce cost requirements while largely eliminating scene limitations and reliance on prior information. This approach has significant research potential and application value. Summary of the Invention

[0006] The purpose of this invention is to provide a combined UAV ranging method based on visual image radial / normal motion, and to provide a new software ranging approach that utilizes platform redundancy information, thereby solving the technical problems of high hardware cost of laser ranging, limited application scenarios of single / binocular visual ranging and triangulation ranging methods, and reliance on prior information.

[0007] To achieve the above objectives and solve the above technical problems, the technical solution of the present invention is as follows:

[0008] A combined UAV ranging method based on visual image trajectory / normal motion includes the following steps:

[0009] Step 1: The photoelectric pod continuously identifies the selected target image through pixel matching and drives the pod motor to change the frame angle to lock it in the center of the image, so as to ensure that the spatial angle of the pod axis is equal to the aircraft-target line of sight angle during the ranging period.

[0010] Step 2: The photoelectric pod will measure the line-of-sight pitch rate ω.θ,L , line-of-sight angular velocity ω ψ,L Pitch frame angle θ F azimuth frame angle ψ F Send to the flight control MCU via serial communication;

[0011] Step 3: The GPS / INS system transmits the measured platform pitch angle θ and azimuth attitude angle ψ to the flight control MCU via serial communication, and compares them with the real-time received frame angle θ. F and azimuth frame angle ψ F Superimpose them separately:

[0012]

[0013] Step 4: The differential positioning system sends the measured precise relative position information of the platform to the flight control MCU via serial communication, and calculates the UAV's horizontal ground speed v in real time through differential estimation. d Climbing speed v h azimuth angle ψ v ;

[0014] Step 5: The distance inference module based on normal motion performs multi-mode calculation and selects the appropriate model through a triple redundancy voting mechanism, such that when L... n,1 ,L n,2 ,L n,3 When there are significant differences in the calculation results, the more accurate distance estimation value is selected by using the principle of "three judgments and two selections".

[0015] The normal calculation channel 1 for the distance to be measured is:

[0016]

[0017] Channel 2 for calculating the normal of the distance to be measured is:

[0018]

[0019] Channel 3 for calculating the normal of the distance to be measured is:

[0020]

[0021] Where μ = v d cosθ L cos(ψ v -ψ L )+v h sinθ L ;

[0022] The voting principle is as follows:

[0023] Principle 1: Perform a three-judgment and two-processing on the distance calculated for the three channels of normal motion. Calculate the difference between the distance of each channel and the distance of the other two channels in turn. If the difference is greater than 5 times the difference between the distances of the other two channels, it indicates that the channel is seriously deviated from the other two channels and is discarded. Otherwise, it is retained.

[0024] Principle 2: Take the average distance of the remaining channels. For a unified formula, denote the set of discarded channel numbers as Ω. When no channels are discarded, Ω is an empty set, and its average value is...

[0025]

[0026] Step 6: The distance inference module based on radial motion performs integral calculations, utilizing the inherent properties of integration to suppress white noise.

[0027] The radial calculation formula for the distance to be measured is:

[0028]

[0029] Where L r0 The initial value can be any value, when ω θ,L ,ω ψ,L When the absolute value is large, it can be initialized to L to speed up the convergence. n0 ;

[0030] Step 7: Simultaneously input the distances inferred from normal and radial motion in steps 5 and 6 into the combined filtering module with a switch-switching mechanism. The switch-switching mechanism is used to filter out bad values ​​in the distance inferred from normal motion; The upper bound of the absolute value of the denominator in formulas (2)-(4) is represented. When σ≤ε, the smaller value ε is the set sensitivity constant. The switch is open and normally kept closed.

[0031] The radial motion inference system is the main system to achieve smooth output, while the normal motion inference system is the auxiliary system to suppress drift error. The radial / normal motion combined filter with a switching mechanism can be either an open-loop or closed-loop architecture.

[0032] Furthermore, the UAV combined ranging method is implemented by a combined ranging system, including a flight control computer, a GPS / INS system, a differential positioning system, and an electro-optical pod;

[0033] The flight control computer communicates with other devices via serial port to relay device information and schedule and coordinate device behavior.

[0034] The GPS / INS system sends the platform attitude angle to the flight control computer, which then forwards the platform attitude angle to the electro-optical pod and superimposes it with the pod frame angle to obtain the line-of-sight angle.

[0035] The differential positioning system sends the precise position information relative to the base station to the flight control computer, corrects the absolute position of the platform, calculates the speed and direction of the aircraft, and transmits it to the on-chip radial / normal motion distance inference module of the flight control computer.

[0036] The optoelectronic pod sends the measured line-of-sight rotation rate to the flight control computer, which, together with the line-of-sight angle obtained from the fusion calculation, transmits it to the on-chip radial / normal motion distance inference module of the flight control computer. After summarizing all the information, the radial / normal motion distance inference module uses the radial motion calculation submodule as the main system and the normal motion calculation submodule as the auxiliary system and performs triple redundancy voting to combine and filter the two distance calculation systems and output the final inferred distance.

[0037] Furthermore, in the calculation of formulas (2), (3), and (4), in order to avoid the occurrence of oddities in the calculation, when the denominator is very close to zero, the calculation distance is taken as a sufficiently large reasonable value.

[0038] Furthermore, the open-loop architecture first needs to determine whether σ≤ε. If not, the difference between the radial motion prediction distance and the normal motion prediction distance is input into the Kalman filter, and the radial motion prediction distance is then subtracted from the filter output to obtain the fused distance.

[0039] Furthermore, the closed-loop architecture first checks if σ≤ε. If not, the filtered output is sent to the radial motion estimation distance module for closed-loop correction, and the corrected radial motion estimation distance is used as the fused distance.

[0040] The beneficial effects of this invention are:

[0041] 1. This invention, based on a conventional electro-optical pod without ranging capabilities, achieves system-level software ranging through onboard equipment interaction and fusion, and line-of-sight / normal motion decomposition, using a combined navigation approach. This significantly reduces the cost of the flight platform. For example, a typical small electro-optical pod without laser ranging equipment costs around 40,000 RMB, while one with laser ranging equipment costs over 80,000 RMB, roughly doubling the cost. This invention, based on a conventional pod without ranging equipment, decomposes line-of-sight / normal motion and utilizes redundant velocity information on the aircraft platform to achieve system-level software ranging using a combined navigation approach, significantly reducing the cost of the flight platform.

[0042] 2. Furthermore, compared to monocular / binocular visual ranging and triangulation methods, this invention can eliminate scene limitations and reliance on prior information to a greater extent, and has significant research prospects and application value. Monocular visual ranging relies on prior target size information and updates and maintains a large sample database, making it impossible to judge and measure targets outside the database. This invention does not rely on target size information. Binocular visual ranging has obvious scene limitations, mainly manifested in the relative decrease in accuracy at long distances. At the same time, the algorithm is complex and difficult to implement on MCUs, making it difficult to guarantee a high refresh rate on fast flight platforms. This invention is based on a different ranging principle, with a relatively simple algorithm that can guarantee a high refresh rate and long-distance accuracy on fast platform MCUs. The triangulation method has low accuracy and requires the target to be at approximately the same altitude as the takeoff point. This invention does not have this scene limitation. Attached Figure Description

[0043] Figure 1 A schematic diagram illustrating the system composition and ranging working principle of this invention;

[0044] Figure 2 Schematic diagram of triple redundancy voting for predicting normal motion distance in this invention;

[0045] Figure 3 A schematic diagram of the open-loop architecture of radial / normal motion combined filtering with a switch switching mechanism of the present invention;

[0046] Figure 4 A schematic diagram of the radial / normal motion combined filtering closed-loop architecture with a switch switching mechanism of the present invention.

[0047] Figure 5 A schematic diagram of the simulation results of this invention. Detailed Implementation

[0048] The design concept of this invention is to address the issue of ordinary optoelectronic pods without ranging capabilities by utilizing redundant information on the aircraft platform to design a fusion inference mechanism to achieve system-level software ranging functionality. This satisfies the hardware ranging requirements of traditional equipment, significantly reducing the cost of the flight platform while ensuring complete functionality. This invention addresses the dynamic geometry of UAV flight by decomposing the UAV's line-of-sight motion radially and normally. It constructs a distance inference module based on normal motion and its triple-redundancy voting mechanism, as well as a distance inference module based on radial motion and its noise suppression mechanism. Based on this, the original distance fusion inference is cleverly transformed into a combined navigation problem. The radial motion distance inference module serves as the main system, and the normal motion distance inference module serves as an auxiliary system, with the former being corrected to suppress drift errors.

[0049] To enable those skilled in the art to better understand the technical solution of this invention, the method will be further described in detail below with reference to the system structure framework and specific embodiments. The low-cost combined ranging system based on visual image path / normal motion consists of the following components: Figure 1 As shown:

[0050] A low-cost combined ranging system based on visual image-based radial / normal motion includes a flight control computer, an electro-optical pod, a GPS / INS system, and a differential positioning system. The flight control computer is the core component, communicating with other devices via serial ports to relay information and coordinate device behavior. The GPS / INS system sends the platform's attitude angles to the flight control computer, which then forwards these angles to the electro-optical pod and superimposes them with the pod's frame angles to obtain the line-of-sight angle. The differential positioning system sends precise position information relative to the base station to the flight control computer, enabling accuracy correction of the platform's absolute position and calculating the aircraft's speed and direction, which is then transmitted to the flight control computer's on-chip radial / normal motion distance inference module. The electro-optical pod sends the measured line-of-sight rotation rate to the flight control computer, which, along with the fused line-of-sight angle, transmits it to the flight control computer's on-chip radial / normal motion distance inference module. After summarizing all the information, the radial / normal motion distance inference module uses the radial motion calculation submodule as the main system and the normal motion calculation submodule as the auxiliary system and performs triple redundancy voting. It then performs combined filtering on the two distance calculation systems and outputs the final inferred distance.

[0051] The UAV combined ranging method based on visual image trajectory / normal motion includes the following steps:

[0052] Step 1: The photoelectric pod continuously identifies the selected target image through pixel matching and drives the pod motor to change the frame angle to lock it in the center of the image, so as to ensure that the spatial angle of the pod axis is equal to the aircraft-target line of sight angle during the ranging period.

[0053] Step 2: The photoelectric pod will measure the line-of-sight pitch rate ω. θ,L and line-of-sight azimuth rate ω ψ,L and pitch frame angle θ F and azimuth frame angle ψ F Send to the flight control MCU via serial communication;

[0054] Step 3: The GPS / INS system transmits the measured platform pitch and azimuth attitude angles θ and ψ to the flight control MCU via serial communication, and superimposes them with the real-time received pitch and azimuth frame angles respectively.

[0055]

[0056] Step 4: The differential positioning system sends the measured precise relative position information of the platform to the flight control MCU via serial communication, and calculates the UAV's horizontal ground speed v in real time through differential estimation. d Climbing speed v h azimuth angle ψv ;

[0057] Step 5: The distance inference module based on normal motion performs multi-mode calculation and selects the appropriate model through a triple redundancy voting mechanism, such that when L... n,1 ,L n,2 ,L n,3 When there are significant differences in the calculation results, a more accurate distance estimation value can be selected by using the principle of "three judgments and two selections".

[0058] The normal calculation channel 1 for the distance to be measured is:

[0059]

[0060] Channel 2 for calculating the normal of the distance to be measured is:

[0061]

[0062] Channel 3 for calculating the normal of the distance to be measured is:

[0063]

[0064] Where μ = v d cosθ L cos(ψ v -ψ L )+v h sinθ L For the sake of convenience, we introduce it;

[0065] In all the above operations, to avoid computational oddities, when the denominator is very close to zero, we take a sufficiently large reasonable value for the computational distance, such as the historical maximum value.

[0066] The voting principle is as follows:

[0067] Principle 1: Perform a three-judgment and two-processing on the distance calculated for the three channels of normal motion. Calculate the difference between the distance of each channel and the distance of the other two channels in turn. If the difference is greater than 5 times the difference between the distances of the other two channels, it indicates that the channel is seriously deviated from the other two channels and is discarded. Otherwise, it is retained.

[0068] Principle 2: Take the average distance of the remaining channels. For a unified formula, denote the set of discarded channel numbers as Ω (Ω is empty when no channels are discarded), and its average value is...

[0069]

[0070] The specific operating procedures are as follows: Figure 2 As shown.

[0071] Step 6: The distance inference module based on radial motion performs integral calculations, utilizing the natural properties of integration to achieve the suppression effect of white noise.

[0072] The radial calculation formula for the distance to be measured is:

[0073]

[0074] Where L r0 The initial value can be any value, when ω θ,L ,ω ψ,L When the absolute value is large, it can be initialized to L to speed up the convergence. n0 .

[0075] Step 7: Simultaneously input the distances inferred based on radial and normal motion into the combined filtering module with a switching mechanism. The switching mechanism is used to filter out bad values ​​in the distances inferred from the normal motion. The upper bound of the absolute value of the denominator in formulas (2)-(4) is given. When σ≤ε, the small value ε (e.g., 0.5) is the set sensitivity constant, the switch is open, and it is normally closed. Radial motion inference is the main system to achieve smooth output, and normal motion inference is the auxiliary system to suppress drift error. The radial / normal motion combined filter with switch switching mechanism can be either open-loop or closed-loop. The open-loop architecture needs to first judge σ≤ε. If it is not true, the difference between the radial motion inference distance and the normal motion inference distance is input to the Kalman filter, and the radial motion inference distance is then subtracted from the filter output to obtain the fused distance. The closed-loop architecture also needs to perform a similar comparison and filtering process. The difference is that the filter output is sent to the radial motion inference distance module for closed-loop correction, and the corrected radial motion inference distance is used as the fused distance. The above open-loop and closed-loop architectures with switch switching mechanism are respectively as follows. Figure 3 and Figure 4 As shown.

[0076] In summary, existing monocular vision ranging technologies require target recognition through image matching, heavily relying on prior target size information and maintaining a large sample database, making it impossible to judge and measure targets outside the database. Binocular vision ranging, on the other hand, suffers from a relative decrease in accuracy when performing long-distance measurements, with significant scene limitations. Furthermore, the complex algorithms of binocular vision ranging are difficult to implement on MCUs, making it challenging to guarantee high refresh rates on fast flight platforms. This invention decomposes the UAV's flight line-of-sight motion along the radial and normal directions, constructing a distance inference module based on normal motion and its triple redundancy voting mechanism, as well as a distance inference module based on radial motion and its noise suppression mechanism. The algorithms are relatively simple and can guarantee high refresh rates and long-distance accuracy on fast platform MCUs. The triangle method has low accuracy and requires the target to be at approximately the same height as the takeoff point, a limitation not present in this invention.

[0077] This invention was verified through simulation using a terminal guidance model of an aircraft built in the MATLAB / SIMULINK environment. The combined ranging module received velocity and attitude information from the flight control system and line-of-sight information from the electro-optical pod (with approximately 10% white noise added as a superposition) to calculate the target distance. Figure 5 Simulation results show that the calculated distance converges quickly and remains close to the actual distance, demonstrating the effectiveness of this ranging method.

[0078] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A combined unmanned aerial vehicle (UAV) ranging method based on visual image path / normal motion, characterized in that, Includes the following steps: Step 1: The photoelectric pod continuously identifies the selected target image through pixel matching and drives the pod motor to change the frame angle to lock it in the center of the image, so as to ensure that the spatial angle of the pod axis is equal to the aircraft-target line of sight angle during the ranging period. Step 2: The photoelectric pod will measure the line-of-sight pitch rate ω. θ,L , line-of-sight angular velocity ω ψ,L Pitch frame angle θ F azimuth frame angle ψ F Send to the flight control MCU via serial communication; Step 3: The GPS / INS system transmits the measured platform pitch angle θ and azimuth attitude angle ψ to the flight control MCU via serial communication, and compares them with the real-time received frame angle θ. F and azimuth frame angle ψ F Superimpose them separately: Step 4: The differential positioning system sends the measured precise relative position information of the platform to the flight control MCU via serial communication, and calculates the UAV's horizontal ground speed v in real time through differential estimation. d Climbing speed v h azimuth angle ψ v ; Step 5: The distance inference module based on normal motion performs multi-mode calculation and selects the appropriate model through a triple redundancy voting mechanism, such that when L... n,1 ,L n,2 ,L n,3 When there are significant differences in the calculation results, the more accurate distance estimation value is selected by using the principle of "three judgments and two selections". The normal calculation channel 1 for the distance to be measured is: Channel 2 for calculating the normal of the distance to be measured is: Channel 3 for calculating the normal of the distance to be measured is: where μ = v d cosθ L cos(ψ v -ψ L ) + v h sinθ L ; Step 6: The distance inference module based on radial motion performs integral calculations, utilizing the inherent properties of integration to suppress white noise. The radial calculation formula for the distance to be measured is: Where L r0 The initial value can be any value, when ω θ,L ,ω ψ,L When the absolute value is large, it can be initialized to L to speed up the convergence. n0 ; Step 7: Simultaneously input the distances inferred from normal and radial motion in steps 5 and 6 into the combined filtering module with a switch-switching mechanism. The switch-switching mechanism is used to filter out bad values ​​in the distance inferred from normal motion; The upper bound of the absolute value of the denominator in formulas (2)-(4) is represented. When σ≤ε, the smaller value ε is the set sensitivity constant. The switch is open and normally kept closed. The radial motion inference system is the main system to achieve smooth output, while the normal motion inference system is the auxiliary system to suppress drift error. The radial / normal motion combined filter with a switching mechanism can be either an open-loop or closed-loop architecture.

2. The UAV combined ranging method based on visual image path / normal motion according to claim 1, characterized in that, The aforementioned UAV combined ranging method is implemented by a combined ranging system, including a flight control computer, a GPS / INS system, a differential positioning system, and an electro-optical pod. The flight control computer communicates with other devices via serial port to relay device information and schedule and coordinate device behavior. The GPS / INS system sends the platform attitude angle to the flight control computer, which then forwards the platform attitude angle to the electro-optical pod and superimposes it with the pod frame angle to obtain the line-of-sight angle. The differential positioning system sends the precise position information relative to the base station to the flight control computer, corrects the absolute position of the platform, calculates the speed and direction of the aircraft, and transmits it to the on-chip radial / normal motion distance inference module of the flight control computer. The optoelectronic pod sends the measured line-of-sight rotation rate to the flight control computer, which, together with the line-of-sight angle obtained from the fusion calculation, transmits it to the on-chip radial / normal motion distance inference module of the flight control computer. After summarizing all the information, the radial / normal motion distance inference module uses the radial motion calculation submodule as the main system and the normal motion calculation submodule as the auxiliary system and performs triple redundancy voting to combine and filter the two distance calculation systems and output the final inferred distance.

3. A UAV combined ranging method based on visual image path / normal motion according to claim 1 or 2, characterized in that, In the calculation of formulas (2), (3), and (4), in order to avoid the occurrence of oddities in the calculation, when the denominator is very close to zero, the calculation distance is taken as a sufficiently large reasonable value.

4. A UAV combined ranging method based on visual image path / normal motion according to claim 1 or 2, characterized in that, The open-loop architecture first needs to determine whether σ≤ε. If not, the difference between the radial motion prediction distance and the normal motion prediction distance is input into the Kalman filter, and the radial motion prediction distance is then subtracted from the filter output to obtain the fused distance.

5. A combined UAV ranging method based on visual image path / normal motion according to claim 1 or 2, characterized in that, The closed-loop architecture first checks if σ≤ε. If not, the filtered output is sent to the radial motion estimation distance module for closed-loop correction, and the corrected radial motion estimation distance is used as the fused distance.

Images