Flight agent environment perception and high-precision map construction method in complex dynamic scene

Abstract

The invention discloses a flight agent environment perception and high-precision map construction method in a complex dynamic scene. The method comprises the following steps: firstly, constructing an OFDM signal transmitting and receiving model; two-dimensional frequency domain decomposition is adopted to realize time delay-Doppler joint estimation, and a dynamic target distance is solved in combination with a light velocity parameter; the radar point cloud is processed through an adaptive threshold clustering algorithm, and a vectorized line feature map is generated through iterative endpoint fitting; and finally, creatively fusing a world model knowledge base to construct a confidence competition mechanism, dynamically distinguishing static / moving targets through geometric matching and a free space conflict method, and realizing pose correction and global map consistency maintenance by adopting least square optimization. According to the method, the limitation that a traditional communication and perception fusion system is high in hardware redundancy and weak in dynamic perception capability is broken through, the dynamic target distance measurement precision is remarkably improved and the calculation complexity is reduced through signal processing optimization and priori knowledge fusion, and the problems of real-time navigation and high-precision mapping in complex dynamic scenes such as high-voltage line routing inspection are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous navigation and environmental perception of unmanned aerial vehicles (UAVs), and in particular relates to a method for environmental perception and high-precision map construction of a flying intelligent body in complex dynamic scenes. Background Art

[0002] With the widespread use of drones, autonomous navigation and environmental perception in complex and dynamic scenarios have become a key research area for high-voltage power line inspection. However, high-voltage power lines are often deployed in environments with complex terrain, dense obstacles, and dynamic interference. This poses severe challenges to drones' real-time perception accuracy, mapping efficiency, and endurance.

[0003] Traditional UAV environmental perception systems typically use an independently configured architecture of lidar, visual sensors, and communication modules, resulting in a multi-hardware redundant design. This discrete architecture results in low system integration and increased overall weight. Especially on small UAV platforms with limited payload capacity, hardware size and power consumption limitations significantly restrict the device's endurance. Existing communication and perception fusion (ISAC) technology achieves simultaneous signal transmission and environmental perception in a single system by sharing spectrum and hardware resources, significantly reducing equipment costs and power consumption. However, ISAC technology still faces a key bottleneck in practical applications: existing methods lack an efficient data processing framework. This problem seriously restricts the practical application of ISAC technology in the UAV field, especially in scenarios such as high-voltage line inspections, where real-time and robustness requirements are stringent. Existing technologies make it difficult to ensure safe navigation and efficient task execution of UAVs.

[0004] At the same time, existing ISAC technology primarily focuses on static or low-speed dynamic environments, with limited ability to perceive high-speed moving targets. This makes it difficult to optimize the map-building process using prior knowledge of the environment. Therefore, a method for environmental perception and map-building for intelligent flying agents that balances hardware efficiency, data processing accuracy, and dynamic adaptability is needed to overcome technical bottlenecks in complex dynamic scenarios. Summary of the Invention

[0005] Purpose of the Invention: This invention aims to provide a method for environmental perception and high-precision mapping for flying agents in complex dynamic scenarios. By integrating a delay-Doppler joint estimation mechanism with prior knowledge, this method addresses the challenges of existing ISAC systems in dynamic scenarios, such as high hardware redundancy, insufficient radar point cloud parsing efficiency, low high-speed target perception accuracy, and insufficient real-time and robust mapping due to a lack of environmental prior constraints. This method enables autonomous navigation of flying agents in high-voltage line inspection scenarios.

[0006] Technical solution: The present invention provides a method for environmental perception and high-precision map construction of a flying intelligent agent in a complex dynamic scene, comprising the following steps:

[0007] Step 1: Construct an OFDM signal transmission and reception model and generate a time domain signal matrix through orthogonal frequency division multiplexing modulation;

[0008] Step 2: Based on the time domain signal matrix, two-dimensional frequency domain decomposition is used to realize the delay-Doppler joint estimation, and the speed of light parameter is combined to solve the dynamic target distance;

[0009] Step 3: Process the radar point cloud using an adaptive threshold clustering algorithm and generate a vectorized line feature map using iterative endpoint fitting.

[0010] Step 4: Integrate the vectorized line feature map and the world model knowledge base to build a confidence competition mechanism, dynamically distinguish static / moving targets through geometric matching and free space conflict method, and use least squares optimization to achieve the pose correction and global positioning of the flying agent. Figure 1 Consistency maintenance.

[0011] Furthermore, step 1 specifically includes the following steps:

[0012] Step 1.1: Design the transmit signal frame based on the OFDM radar system, define the subcarrier spacing, guard interval length, and symbol duration, and generate a time-domain signal matrix containing orthogonal frequency division multiplexing modulation. The transmitted OFDM signal matrix is represented by an M×N matrix as follows:

[0013]

[0014] Among them, N rows represent the frequency components of N orthogonal subcarriers, M columns represent M consecutive OFDM symbol periods, and element a k,l is a set of normalized complex modulation constellations;

[0015] Step 1.2: Construct a time-domain received signal model containing noise, analyze the time delay and Doppler shift introduced by the target reflection, and model the signal reflection matrix after the first signal contacts the target as follows:

[0016]

[0017] Among them, F Tx is the emission matrix, F Rx is the receiving matrix, is the Doppler effect caused by the target velocity, f D,0 is the Doppler frequency shift, T0 is the total symbol duration, is the propagation delay caused by the target distance, τ0 is the propagation delay caused by the target distance, Δf is the subcarrier spacing, is the phase term, and W is the matrix representation of additive white Gaussian noise.

[0018] Furthermore, step 2 specifically includes the following steps:

[0019] Step 2.1: Eliminate the influence of known modulation data through normalization, separate target parameters from noise, and generate the normalized matrix F:

[0020]

[0021] in is the phase term, is a fixed phase offset, is the normalized noise term;

[0022] Step 2.2: Use two-dimensional FFT to decompose the received signal in the frequency domain, and independently estimate the discrete grid values of delay and Doppler shift. The signal estimation likelihood function is:

[0023]

[0024] Among them, the inner summation of the function is: For each subcarrier k, a discrete Fourier transform is performed on the symbol dimension l to estimate the Doppler frequency shift f D , Outer summation: For each symbol period l, perform inverse discrete Fourier transform on subcarrier dimension k to estimate the delay τ, is the phase term, The phase shift is eliminated by taking the real part Re, and only the phase shift related to τ and f is retained. D relevant information;

[0025] Step 2.3: Based on the quantified results of time delay and Doppler shift, combined with the speed of light and subcarrier spacing parameters, calculate the target dynamic range.

[0026] Furthermore, step 3 specifically includes the following steps:

[0027] Step 3.1: Adaptively adjust the segmentation threshold according to the radar angular resolution to segment the point cloud into independent object clusters. The distance threshold is calculated using the following formula:

[0028] r AB ≤C0+C1min{r OA ,r OB}

[0029] Among them, r AB is the distance between two radar target points A and B, r OA ,r OB is the distance between the detection end and the target points A and B, The constant C0 enables the algorithm to handle sensor noise and overlapping of close-range pulses, while C1 causes the distance between points to increase as their distance from the OFDM radar increases;

[0030] Step 3.2: Convert the clustered point cloud into a vectorized line feature map and use the iterative endpoint fitting (IEPF) algorithm to record the start, end, and variance attributes of the line segments to support efficient analysis of dynamic scenes.

[0031] Furthermore, step 4 specifically includes the following steps:

[0032] Step 4.1: Complete the construction of the world model knowledge base (WMKS) by storing the vectorized line feature map and dynamic target bounding box information and associating confidence attributes;

[0033] Step 4.2: Update the dynamic target confidence by using the free space conflict method to distinguish between static and dynamic targets;

[0034] Step 4.3: Map update and global consistency maintenance based on geometric matching and pose optimization. Use geometric matching algorithm to align historical and current scan data based on least squares method, optimize drone pose estimation, and maintain global map consistency. Figure 1 Consistency.

[0035] The present invention further discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0036] The present invention further discloses a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the method of the present invention when the computer program / instruction is executed by a processor.

[0037] The present invention further discloses a computer program product, comprising a computer program / instruction, which implements the steps of the method of the present invention when executed by a processor.

[0038] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0039] (1) This paper proposes a high-precision map construction method based on OFDM radar communication and perception integration technology. Through algorithms such as two-dimensional frequency domain decomposition, adaptive threshold clustering, and confidence competition, it achieves high-speed target detection and high-precision environment modeling, significantly reducing hardware load and optimizing computational efficiency. The system adopts dynamic anti-interference design and prior knowledge fusion mechanism to maintain stable performance in complex electromagnetic environments and extreme weather. It is particularly suitable for payload-constrained platforms such as drones, balancing battery life improvement with real-time decision-making capabilities.

[0040] (2) The present invention solves the problems of the existing ISAC system in dynamic scenarios, such as high hardware redundancy, insufficient radar point cloud parsing efficiency, low high-speed target perception accuracy, and insufficient real-time and robust mapping due to the lack of environmental prior constraints, through the delay-Doppler joint estimation and prior knowledge fusion mechanism, thus meeting the autonomous navigation of flying intelligent bodies in high-voltage line inspection scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a schematic diagram of the high-voltage line inspection of the flying intelligent body;

[0042] Figure 2 A flow chart of the method for environmental perception and high-precision map construction of flying agents in complex dynamic scenarios;

[0043] Figure 3 To solve the distance r AB Schematic diagram;

[0044] Figure 4 This is a comparison chart of the distance error between the OFDM radar ranging method and the traditional ranging method. DETAILED DESCRIPTION

[0045] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0046] like Figure 1 Figure 2 shows a flying agent inspecting high-voltage power lines. Equipped with an OFDM radar and a map-optimized SLAM system, the flying agent performs inspections in a complex and dynamic environment. The OFDM radar module integrated into the agent's front end transmits high-frequency signals using orthogonal frequency division multiplexing (OFDM) technology. Its multi-subcarrier modulation and cyclic prefix design significantly enhance its ability to mitigate multipath interference.

[0047] like Figure 2 As shown, an embodiment of the present invention provides a method for complex environment perception and map construction of a UAV based on OFDM radar and graph optimization SLAM, the method comprising the following steps:

[0048] 1. Construction of OFDM signal transmission and reception model

[0049] (1) Transmitted signal modeling

[0050] A complete signal frame transmitted in an OFDM radar system is represented by an M×N matrix:

[0051]

[0052] Where N rows represent the frequency components of N orthogonal subcarriers, M columns represent M consecutive OFDM symbol periods, and element a k,lSelected from the set of normalized complex modulation constellations, it means that the modulation symbol of the k-th subcarrier in the l-th symbol period meets the power constraint:

[0053] E{|a k,l | 2}=1(2)

[0054] F TX Perform inverse fast Fourier transform (IFFT) on each column to convert the frequency domain symbols into time domain waveforms. And add a length of T before the result. G The cyclic prefix of . Then concatenate the results of each column to get the time domain signal s(t):

[0055]

[0056] Where Δf is the subcarrier spacing, T G is the length of the guard interval, T0 is the total duration of the symbol, T0 = 1Δf + T G .

[0057] (2) Received signal modeling

[0058] There are multiple reflecting targets in the environment. The received signal r(t) is composed of the transmitted signal reflected by the target, time delay, Doppler frequency shift and additive white Gaussian noise. Its expression is:

[0059]

[0060] Among them, H is the number of reflective targets in the environment, τ h is the delay, f D,h is the Doppler shift, is the complex attenuation factor of the hth target, where |b h | represents the amplitude attenuation of the signal after being reflected by the hth target, represents the phase shift introduced by the reflection path, The variance is σ 2 complex Gaussian white noise.

[0061] The backscattered transmission matrix for the first target will be converted into the receiving matrix F Rx :

[0062]

[0063] in, is the Doppler effect caused by the target velocity, f D,0 is the Doppler shift, is the propagation delay caused by the target distance, τ0 is the propagation delay caused by the target distance, is the phase term, W is the matrix representation of additive white Gaussian noise, whose elements are independent and identically distributed random values with variance σ 2 All phase shifts that remain constant throughout the frame are uniformly represented as the phase term

[0064] 2. Joint Delay-Doppler Estimation and Range Calculation Signal Preprocessing and Noise Suppression

[0065] (1) Signal normalization and noise separation

[0066] For the receiving matrix F Rx With the emission matrix F Tx Perform element-by-element division to eliminate the phase influence of known modulation symbols and generate a normalized matrix F:

[0067]

[0068] in, is the phase term, is a fixed phase offset, is the normalized noise term.

[0069] This operation converts the received signal into a new matrix form that contains the target parameters and noise. The final matrix F is transmitted to the target delay τ and Doppler frequency shift f D Estimator of . W is the additive white Gaussian noise matrix, and the noise variance is:

[0070]

[0071] (2) Two-dimensional frequency domain decomposition and delay-Doppler joint estimation

[0072] First, for the case where the number of targets H is exactly 1 and there is no prior information about the targets, the present invention defines an estimator. The first step is to define a parameter vector:

[0073]

[0074] Where τ represents the propagation delay, f D represents the Doppler shift, is the phase offset.

[0075] The log-likelihood function is:

[0076]

[0077] Among them, the inner summation of the function is: For each subcarrier k, a discrete Fourier transform is performed on the symbol dimension l to estimate the Doppler frequency shift f DOuter summation: For each symbol period l, an inverse discrete Fourier transform is performed on the subcarrier dimension k to estimate the delay τ. is the phase term, The phase shift is eliminated by taking the real part Re, and only the phase shift related to τ and f is retained. D Related information.

[0078] When searching for the maximum value of θ on a discrete grid, a method based on Fast Fourier Transform (FFT) can be used. D The quantification is as follows:

[0079]

[0080] Among them, τ Q,n is the quantized delay value, N FFT is the number of FFT points, f D,Q,m is the quantized Doppler frequency shift value. Since the Doppler frequency shift value can be negative, f D,Q,m Symmetrically distributed around zero.

[0081] In order to find the discrete value closest to the true value and, by substituting equations (10) and (11) into equation (9), we can obtain:

[0082]

[0083] in

[0084]

[0085] Among them, M FFT =M,N FFT =N.

[0086] For time delay τ and Doppler frequency shift f D Perform maximum likelihood estimation to maximize the value of |A(m,n)|:

[0087]

[0088] Among them, IFFT k,MFFT (m) represents the fast Fourier transform of each column of the F matrix, Indicates that each row of the result is inversely fast Fourier transformed, |A(m,n)| is the modulus of A(m,n), eliminating the phase offset impact.

[0089] Define the maximum estimate and for:

[0090]

[0091] Among them, the search range of Doppler frequency shift is symmetrical interval -m max ...m max , the search range of the delay is defined as the non-negative interval 0...n max The ratio parameter of the cyclic prefix to the OFDM symbol length is set to G = 1 / 4, and the maximum ratio parameter of the Doppler frequency shift to the subcarrier spacing is set to D = 1 / 10.

[0092] (2) Calculating the target dynamic distance

[0093] Based on the quantified results of time delay and Doppler shift, combined with the speed of light and subcarrier spacing parameters, the target dynamic range r is calculated:

[0094]

[0095] Where Δf is the subcarrier spacing, c is the speed of light, is the estimated value of the delay.

[0096] 3. Point cloud processing and feature map generation

[0097] (1) Perform dynamic threshold clustering on the radar scan point cloud, adaptively adjust the segmentation threshold according to the radar angular resolution, and segment the point cloud into independent object clusters

[0098] The present invention determines cluster membership by considering the distance between two consecutive points. If the distance between the two points is within a threshold range, the points are considered to belong to the same cluster, otherwise the points belong to two different clusters.

[0099] like Figure 3 As shown, for points A and B obtained from OFDM radar scanning, the distance r between them is AB It can be calculated by the following formula:

[0100]

[0101] Where α is r OA With r OB The angle between them.

[0102] The distance threshold is calculated using the following formula:

[0103] r AB ≤C0+C1min{r OA ,r OB}(19)

[0104] in, The constant C0 enables the algorithm to handle sensor noise and the overlap of closely spaced pulses, while C1 causes the distance between points to increase as their distance from the OFDM radar increases.

[0105] (2) Vectorized Line Feature Map

[0106] Combining the above steps, the iterative endpoint fitting (IEPF) algorithm is used to convert the clustered point cloud into a vectorized line feature map, recording the starting point, end point and variance attributes of the line segment.

[0107] Algorithm 1: Iterative endpoint fitting (IEPF) algorithm:

[0108] Input: point set Distance threshold δ

[0109] Output: polyline segment set

[0110] 1. Initialization:

[0111] 1.1 Create an empty stack The original point set Push onto the stack

[0112] 1.2 Initialize the polyline segment set

[0113] 2. Iterative segmentation:

[0114]

[0115] a. Pop the subset from the top of the stack

[0116] b. Fitting line segment L fit :

[0117] calculate Starting point p start and the endpoint p end

[0118] Construct the parametric equation of a straight line: L fit ={p start +t(p end -p start )|t∈[0,1]}

[0119] c. Maximum distance calculation:

[0120] Traversal All points p i , calculate the vertical distance

[0121] Record the maximum distance d max =max{d i} and the corresponding point p split

[0122] d. Segmentation determination:

[0123] ifd max >δ:

[0124] i. In p split Department of Generals Split into two subsets and

[0125] ii. and Push onto the stack

[0126] e.Else: Line segment L fit join in

[0127] 3. Output result: Returns a set of polyline segments

[0128] 4. Combined with the graph optimization SLAM framework of the World Model Knowledge Base (WMKS), pose correction and map update are achieved through geometric matching and confidence competition mechanism.

[0129] (1) Construction of World Model Knowledge Base (WMKS)

[0130] Vectorized line features generated by the Iterative Endpoint Fitting (IEPF) algorithm, along with line segment start and end points and variance attributes, are stored in WMKS as a base map for static environments. A confidence attribute is added to static line features, with the initial confidence set to 300. This confidence is gradually increased to a threshold of 600 based on the number of successful matches before being stored in WMKS. Communication with WMKS is achieved through a messaging mechanism, with local caching ensuring real-time performance. Detected dynamic objects (such as moving vehicles and pedestrians) are marked as independent object clusters, and their bounding box information (position, size) and confidence level are stored. Dynamic objects are stored only in the local cache and not written to WMKS.

[0131] (2) Distinguishing between moving objects and static objects

[0132] The present invention uses a free-space collision method to detect both moving and static objects. In this method, objects detected in the current scan are compared with free-space polygons generated from the previous OFDM radar scan. When a free-space collision is detected (overlap with the historical polygon ≥ 50%), the dynamic confidence score is updated in steps of +10. If it exceeds a threshold of 100, the object is marked as moving.

[0133] (3) Map update and global consistency maintenance based on geometric matching algorithm and pose optimization

[0134] The present invention uses a geometric matching algorithm to match newly detected line segments with line segments in the static map to update the map and maintain global consistency.

[0135] Algorithm 2: Geometric matching algorithm:

[0136] Input: New detection line segment set L new ={l1,l2,…,l N}

[0137] Static Line Map M static ={s1,s2,...,s M}

[0138] Overlap area threshold θ overlap =50%

[0139] Maximum length difference threshold ΔL max =0.2m

[0140] Maximum direction deviation threshold Δθ max =5°

[0141] Output: Set P of successfully matched line segments matched ={(l i ,s j )}

[0142] 1. Initialization:

[0143] 1.1 Create an empty set P matched and candidate list C candidate

[0144] 1.2 Traversing M static Each static line segment s in j :

[0145] a. Buffer radius r j =0.4m

[0146] b. Generate buffer polygon B using the GEOS library j

[0147] 2. Overlap determination and candidate matching screening:

[0148] 2.1 Traversing L new Each new line segment l in i :

[0149] a. To M static Each s in j , calculate l i With B j The overlapping area A ij

[0150] b. If A ij ≥θ overlap ×Area(B j), (l i ,s j ) Add C candidate

[0151] 3. Attribute verification and matching confirmation:

[0152] 3.1 vs. C candidate Each candidate pair (l i ,s j ):a. Length check:

[0153] Extracts j Length L old and l i Length L new

[0154] Calculate the length difference ΔL=|L new -L old |

[0155] If ΔL>ΔL max , marked as length mismatch, skipping subsequent verification b. Length verification:

[0156] Extracts j The direction angle θ old and l i The direction angle θ new

[0157] Calculate the direction difference Δθ=|θ new -θ old |

[0158] If Δθ>Δθ max , marked as direction mismatch, skip matching

[0159] c. Match successful:

[0160] If the verification is passed, (l i ,s j ) Join P matched

[0161] 4. Rejection scenario:

[0162] If the overlapping area of a candidate pair meets the requirements but the length or direction deviation exceeds the limit, it will be directly excluded and not included in the matching results.

[0163] 5. Output result: set P matched

[0164] Confidence competition mechanism:

[0165] The static existence confidence is expressed as follows:

[0166]

[0167] Among them, the threshold: C exist ≥600 deposited into WMKS, C exist ≤-600 Delete objects

[0168] The dynamic confidence is expressed as follows:

[0169]

[0170] The dynamic confidence is initially 0, increases to +10 in case of free space collision, and decreases to -10 otherwise. Objects with a dynamic confidence ≥ 100 are marked as moving objects, and those with a dynamic confidence below -100 are restored to static candidates.

[0171] If the match is successful, merge the line segments, update the WMKS object's geometric properties, perform pose estimation, and trigger pose correction:

[0172] The proposed estimation method attempts to match the points obtained from the current scan with the static objects modeled in the previous scan. It is important to note that only the static objects matched in the object matching step will be used to generate the pose estimate. ref and move to position P new The scan data S obtained at ref and S new , for S new Every point P in i , and S ref The corresponding point P i '. Define the error function of the pose transformation as the sum of the squares of the Euclidean distances between corresponding points and calculate the least squares solution:

[0173]

[0174] The optimal rotation ω and translation T are solved by this formula, where n is the number of corresponding point pairs.

[0175] The translation T is defined as:

[0176] T=(T x ,T y ) (25)

[0177] T represents the translation in the two-dimensional plane, T x and T y are the translation amounts in the x and y directions respectively.

[0178] Among them, R ω The rotation matrix of the rotation angle ω is expressed as:

[0179]

[0180] The optimal solution of ω is obtained by the following formula:

[0181]

[0182] Optimal translation and It is obtained by the following formula:

[0183]

[0184] Where X, Y, and Z are:

[0185]

[0186] Among them, S x , S y The current scanning point coordinates are accumulated, S x′ , S y' is the accumulation of matched historical scanning point coordinates. , respectively expressed as:

[0187]

[0188]

[0189] The new rotation and translation are then applied to the current pose estimate to reduce the position error between the two scans, and the process is repeated until the solution converges, for 20 iterations.

[0190] The optimized pose is then applied to the world coordinates of the WMKS object to complete the map construction.

Claims

1. A method for environmental perception and high-precision map construction of a flying agent in complex dynamic scenes, characterized by: The steps include: Step 1: Construct an OFDM signal transmission and reception model and generate a time domain signal matrix through orthogonal frequency division multiplexing modulation; Step 2: Based on the time domain signal matrix, two-dimensional frequency domain decomposition is used to realize the delay-Doppler joint estimation, and the speed of light parameter is combined to solve the dynamic target distance; Step 3: Process the radar point cloud using an adaptive threshold clustering algorithm and generate a vectorized line feature map using iterative endpoint fitting. Step 4: Integrate the vectorized line feature map and the world model knowledge base to build a confidence competition mechanism, dynamically distinguish static / moving targets through geometric matching and free space conflict method, and use least squares optimization to achieve the posture correction of the flying agent and maintain global map consistency.

2. The method for environmental perception and high-precision map construction of a flying agent in a complex dynamic scene according to claim 1 is characterized in that: Step 1 specifically includes the following steps: Step 1.1: Design the transmit signal frame based on the OFDM radar system, define the subcarrier spacing, guard interval length, and symbol duration, and generate a time-domain signal matrix containing orthogonal frequency division multiplexing modulation. The transmitted OFDM signal matrix is represented by an M×N matrix as follows: Among them, N rows represent the frequency components of N orthogonal subcarriers, M columns represent M consecutive OFDM symbol periods, and element a k,l is a set of normalized complex modulation constellations; Step 1.2: Construct a time-domain received signal model containing noise, analyze the time delay and Doppler shift introduced by the target reflection, and model the signal reflection matrix after the first signal contacts the target as follows: Among them, F Tx is the emission matrix, F Rx is the receiving matrix, is the Doppler effect caused by the target velocity, f D,0 is the Doppler frequency shift, T0 is the total symbol duration, is the propagation delay caused by the target distance, τ0 is the propagation delay caused by the target distance, Δf is the subcarrier spacing, is the phase term, and W is the matrix representation of additive white Gaussian noise.

3. The method for environmental perception and high-precision map construction of a flying agent in a complex dynamic scene according to claim 2 is characterized in that: Step 2 specifically includes the following steps: Step 2.1: Eliminate the influence of known modulation data through normalization, separate target parameters from noise, and generate the normalized matrix F: in is the phase term, is a fixed phase offset, is the normalized noise term; Step 2.2: Use two-dimensional FFT to decompose the received signal in the frequency domain, and independently estimate the discrete grid values of delay and Doppler shift. The signal estimation likelihood function is: Among them, the inner summation of the function is: For each subcarrier k, a discrete Fourier transform is performed on the symbol dimension l to estimate the Doppler frequency shift f D , Outer summation: For each symbol period l, perform inverse discrete Fourier transform on subcarrier dimension k to estimate the delay τ, is the phase term, The phase shift is eliminated by taking the real part Re, and only the phase shift related to τ and f is retained. D relevant information; Step 2.3: Based on the quantified results of time delay and Doppler shift, combined with the speed of light and subcarrier spacing parameters, calculate the target dynamic range.

4. The method for environmental perception and high-precision map construction of a flying agent in a complex dynamic scene according to claim 1 is characterized in that: Step 3 specifically includes the following steps: Step 3.1: Adaptively adjust the segmentation threshold according to the radar angular resolution to segment the point cloud into independent object clusters. The distance threshold is calculated using the following formula: r AB ≤C0+C1min{r OA ,r OB } Among them, r AB is the distance between two radar target points A and B, r OA ,r OB is the distance between the detection end and the target points A and B, The constant C0 enables the algorithm to handle sensor noise and overlapping of close-range pulses, while C1 causes the distance between points to increase as their distance from the OFDM radar increases; Step 3.2: Convert the clustered point cloud into a vectorized line feature map and use the iterative endpoint fitting (IEPF) algorithm to record the start, end, and variance attributes of the line segments to support efficient analysis of dynamic scenes.

5. The method for environmental perception and high-precision map construction of a flying agent in a complex dynamic scene according to claim 1 is characterized in that: Step 4 specifically includes the following steps: Step 4.1: Complete the construction of the world model knowledge base (WMKS) by storing the vectorized line feature map and dynamic target bounding box information and associating confidence attributes; Step 4.2: Update the dynamic target confidence by using the free space conflict method to distinguish between static and dynamic targets; Step 4.3: Map update and global consistency maintenance based on geometric matching and pose optimization. A geometric matching algorithm is used to align historical and current scan data based on the least squares method to optimize the drone pose estimation and maintain global map consistency.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to claim 1.

7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images