Multi-radar-based indoor multi-target detection method and device, terminal and medium

Abstract

The application discloses a multi-radar-based indoor multi-target detection method and device, a terminal and a medium. The method generates each candidate global correlation hypothesis across radars according to each actual measurement data at the current moment; predicts each prior state estimation and a prediction covariance matrix of each motion trajectory at the current moment based on a posterior state estimation of the motion trajectory at the last moment and a target covariance matrix; in the candidate global correlation hypothesis, calculates a comprehensive cost based on a comprehensive cost function, the actual measurement data, each prior state estimation and the prediction covariance matrix, including a position consistency error, a Doppler consistency error and a physical constraint penalty term; calculates a cumulative comprehensive cost of each candidate global correlation hypothesis in a sliding time window, and selects a motion trajectory as the motion trajectory at the current moment according to the cumulative comprehensive cost. Thus, the problem that the prior art ignores trajectory time consistency and motion direction constraints, resulting in target tracking loss and system false positives, is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of indoor detection and radar sensing technology, and in particular to indoor multi-target detection methods, devices, terminals and media based on multiple radars. Background Technology

[0002] With the rapid development of the Internet of Things, 6G communication and sensing integration, and smart healthcare and smart homes, the demand for high-precision positioning and sensing services in indoor scenarios is becoming increasingly urgent. Traditional satellite navigation technology cannot achieve stable and reliable positioning indoors due to problems such as building obstruction and multipath effects. Alternative technologies such as Wi-Fi, Bluetooth, ultra-wideband (UWB), vision, and LiDAR have limitations such as insufficient accuracy, the need for active tagging, privacy leaks, and high costs.

[0003] Millimeter-wave radar, with its advantages of being unobtrusive, protecting privacy, having strong penetration, and operating in all weather conditions, has become the preferred technology for indoor human body monitoring and positioning. However, single radars suffer from problems such as limited field of view, blind spots, obstruction, and large positioning errors at long distances. Therefore, distributed multi-radar collaborative sensing has become an inevitable trend for solving positioning problems in complex indoor scenarios.

[0004] However, the cross-radar target association, a core component of multi-radar systems, suffers from critical flaws. Existing nearest-neighbor-based association algorithms rely solely on single-frame Euclidean distance matching, neglecting trajectory temporal consistency and motion direction constraints. This leads to symmetrical cross-matching errors, resulting in false "ghost points" and trajectory jumps, causing target tracking loss and false alarms. Consequently, the reliability of multi-radar systems in scenarios such as smart healthcare and security monitoring is severely compromised.

[0005] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an indoor multi-target detection method, device, terminal and medium based on multiple radars to address the above-mentioned defects of the prior art. The aim is to solve the problems of the prior art ignoring trajectory time consistency and motion direction constraints, which easily leads to symmetrical cross-matching errors, false "ghost points" and trajectory jumps, resulting in target tracking loss and system false alarms.

[0007] The technical solution adopted by this invention to solve the problem is as follows: In a first aspect, embodiments of the present invention provide an indoor multi-target detection method based on multiple radars, wherein the method includes: Obtain the actual measurement data of each radar at the current moment, and generate each candidate global correlation hypothesis across radars based on the actual measurement data; If the current time is not the initial time, obtain the posterior state estimate and target covariance matrix of the motion trajectory at the previous time, and determine the prior state estimate and prediction covariance matrix of each motion trajectory at the current time based on the posterior state estimate and target covariance matrix of the previous time. In each of the candidate global association hypotheses, a comprehensive cost is calculated based on the comprehensive cost function, the actual measurement data, each of the prior state estimates and the prediction covariance matrix. The comprehensive cost function is achieved by weighted summation of the positional consistency error, the Doppler consistency error and the physical constraint penalty term. Calculate the first cumulative comprehensive cost of each of the candidate global association hypotheses within the sliding time window, and determine the optimal global association hypothesis among the candidate global association hypotheses based on the first cumulative comprehensive cost; The trajectory of the motion under the optimal global association hypothesis is taken as the trajectory of the motion at the current moment.

[0008] In one implementation, if the current time is the initial time, the method includes: Acquire the actual measurement data at each moment within the initialization time window, and generate the corresponding initial global correlation hypothesis across radars based on the actual measurement data at each moment; Calculate the second cumulative integrated cost of each initialization global association hypothesis at the end of the initialization time window; Based on each of the second cumulative comprehensive costs, an initial global correlation hypothesis is selected, and the motion trajectory at the initial time is determined based on the selected initial global correlation hypothesis.

[0009] In one implementation method, in each of the candidate global association hypotheses, a comprehensive cost is calculated based on a comprehensive cost function, according to the actual measurement data, each of the prior state estimates, and the prediction covariance matrix, including: The optimal trace matching scheme for each radar is determined based on the actual measurement data of each radar at the current moment, the prior state estimates, and the predicted covariance matrix. In each of the candidate global association hypotheses, each motion trajectory is updated according to the optimal trajectory matching scheme to determine the posterior state estimate and target covariance matrix of the motion trajectory at the current time. The comprehensive cost of the candidate global association hypothesis is calculated using a comprehensive cost function based on the prior state estimate, the predicted covariance matrix, the posterior state estimate, and the target covariance matrix.

[0010] In one implementation method, determining the optimal trajectory matching scheme for each radar based on the actual measurement data of each radar at the current time, the prior state estimates, and the predicted covariance matrix includes: Each of the prior state estimates is mapped to the measurement space to determine the corresponding predicted measurement value; Based on the predicted covariance matrix, the matching cost between each actual measurement data and each predicted measurement value of each radar is calculated one by one. The optimal trace matching scheme for each radar is determined based on the matching cost described for each radar.

[0011] In one implementation method, determining the optimal track matching scheme for each radar based on the respective matching costs of each radar includes: The matching costs are arranged and combined to determine a two-dimensional cost matrix; The two-dimensional cost matrix is ​​transformed into a bipartite graph matching problem, and the Hungarian algorithm is called to perform global optimization to solve the bipartite graph matching problem, thereby determining the optimal trace matching scheme for each radar.

[0012] In one implementation method, in each of the candidate global association hypotheses, each motion trajectory is updated according to each optimal trajectory matching scheme to determine the posterior state estimate and target covariance matrix of the motion trajectory at the current time, including: Determine whether each motion trajectory matches the actual measurement data based on the optimal trajectory matching scheme described above; If so, calculate the Kalman gain corresponding to each actual measurement data in each candidate global association hypothesis, correct the prior state estimate of the corresponding motion trajectory based on each Kalman gain and each actual measurement data, and determine the posterior state estimate of the motion trajectory and the target covariance matrix at the current moment. If not, the prior state and predicted covariance matrix of the motion trajectory are used as the posterior state estimate and target covariance matrix of the motion trajectory at the current moment.

[0013] In one implementation, the physical constraint penalty term is composed of a weighted sum of an existence penalty, a spatial boundary penalty, and a human dynamics penalty.

[0014] Secondly, embodiments of the present invention also provide an indoor multi-target detection device based on multiple radars, wherein the indoor multi-target detection device based on multiple radars includes: The hypothesis generation module is used to obtain the actual measurement data of each radar at the current time, and generate candidate global correlation hypotheses across radars based on the actual measurement data. The state estimation module is used to obtain the posterior state estimate and target covariance matrix of the motion trajectory at the previous time if the current time is not the initial time, and to determine the prior state estimate and prediction covariance matrix of each motion trajectory at the current time based on the posterior state estimate and target covariance matrix of the previous time. The cost calculation module is used to calculate the comprehensive cost in each of the candidate global association hypotheses based on the comprehensive cost function, according to the actual measurement data, each of the prior state estimates and the prediction covariance matrix. The comprehensive cost function is implemented by weighted summation of the position consistency error, Doppler consistency error and physical constraint penalty term. The hypothesis filtering module is used to calculate the first cumulative comprehensive cost of each of the candidate global association hypotheses within the sliding time window, and to determine the optimal global association hypothesis among the candidate global association hypotheses based on the first cumulative comprehensive cost. The trajectory determination module is used to take the motion trajectory of the optimal global association hypothesis as the motion trajectory at the current moment.

[0015] Thirdly, embodiments of the present invention also provide a terminal, the terminal including a memory and one or more processors; the memory stores one or more programs; the programs include instructions for executing the indoor multi-target detection method based on multiple radars as described above; the processor is used to execute the programs.

[0016] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a plurality of instructions, wherein the instructions are adapted to be loaded and executed by a processor to implement any of the above-described multi-radar-based indoor multi-target detection methods.

[0017] The beneficial effects of this invention are as follows: In this embodiment, candidate global correlation hypotheses are generated across radars based on actual measurement data at the current moment; prior state estimates and predicted covariance matrices for each trajectory at the current moment are predicted based on the posterior state estimate and target covariance matrix of the previous moment's trajectory; among the candidate global correlation hypotheses, a comprehensive cost function is used to calculate the comprehensive cost based on actual measurement data, prior state estimates, and the predicted covariance matrix. The comprehensive cost function is achieved by weighted summation of position consistency error, Doppler consistency error, and physical constraint penalty terms; the cumulative comprehensive cost of each candidate global correlation hypothesis is calculated within a sliding time window, and the trajectory of the candidate global correlation hypothesis is selected as the trajectory at the current moment based on the cumulative comprehensive cost. This effectively solves the problem of existing technologies neglecting trajectory time consistency and motion direction constraints, leading to target tracking loss and system false alarms. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the indoor multi-target detection method based on multiple radars provided in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of a scenario provided in an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of ghost point generation provided in an embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram illustrating the specific process of the indoor multi-target detection method based on multiple radars provided in this embodiment of the invention.

[0023] Figure 5 This is a schematic diagram of ghost point resolution based on Doppler consistency provided in an embodiment of the present invention.

[0024] Figure 6 This is a schematic diagram illustrating the relationship between algorithm performance and target quantity provided in an embodiment of the present invention.

[0025] Figure 7 This is a schematic diagram of the internal modules of the indoor multi-target detection device based on multiple radars provided in an embodiment of the present invention.

[0026] Figure 8 This is a schematic diagram of the terminal provided in the embodiment of the present invention. Detailed Implementation

[0027] This invention discloses a method, device, terminal, and medium for indoor multi-target detection based on multiple radars. To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0028] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0029] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0030] In the core data fusion stage of multi-radar systems, cross-radar target association has a critical flaw. Existing nearest neighbor-based association algorithms rely solely on single-frame target Euclidean distance matching, ignoring trajectory temporal consistency and motion direction constraints. This easily leads to symmetrical cross-matching errors, resulting in false "ghost points" and trajectory jumps, causing target tracking loss and system false alarms. This seriously affects the reliability of multi-radar systems in scenarios such as smart healthcare and security monitoring.

[0031] For example, within a specific area, radar R1 and radar R2 simultaneously observe two closely spaced or moving towards each other and intersecting targets P1 and P2 from different perspectives. Because the local distance between the two targets is extremely close, the Euclidean distance or local cost function of a single frame will tend to be similar at this time, losing its distinguishability. If the algorithm incorrectly associates the trajectory of P1 observed by radar R1 with the trajectory of P2 observed by radar R2 (i.e., symmetrical cross-matching occurs), then in the subsequent geometric solution and state fusion stages, the system will calculate a false target location that does not exist in the real physical world at the spatial intersection of the lines of sight of radar R1 and radar R2. Figure 3 As shown, these false targets derived from erroneous associations are called "ghost points." As the target continues to move, these ghost points will form false trajectories in space independent of the real target. In smart healthcare and security monitoring systems, the appearance of ghost points is disastrous. It not only causes the tracking identity of the real target to change or be lost, but also triggers frequent false alarms in the system, seriously undermining the high reliability that multi-radar systems are originally intended to provide.

[0032] To address the aforementioned shortcomings of existing technologies, this invention provides an indoor multi-target detection method based on multiple radars. The method generates candidate global association hypotheses across radars based on actual measurement data at the current moment; predicts prior state estimates and predicted covariance matrices for each trajectory at the current moment based on the posterior state estimate of the previous moment's trajectory and the target covariance matrix; in each candidate global association hypothesis, a comprehensive cost function is calculated based on the actual measurement data, prior state estimates, and the predicted covariance matrices. The comprehensive cost function is achieved by weighted summation of position consistency error, Doppler consistency error, and physical constraint penalty terms; the cumulative comprehensive cost of each candidate global association hypothesis is calculated within a sliding time window, and the trajectory of the candidate global association hypothesis is selected as the trajectory at the current moment based on the cumulative comprehensive cost. This effectively solves the problem of existing technologies neglecting trajectory time consistency and motion direction constraints, leading to target tracking loss and false alarms.

[0033] Exemplary method: like Figure 1 As shown, the method includes: Step S100: Obtain the actual measurement data of each radar at the current time, and generate each candidate global correlation hypothesis across radars based on the actual measurement data.

[0034] For radar networks consisting of multiple single-instance single-receive (SISO) or multiple-instance multiple-receive (MIMO) radars, multiple sets of measurements, such as range, azimuth, velocity, coordinate point cloud, and signal-to-noise ratio, are acquired simultaneously. Data verification and timestamp alignment are performed to ultimately obtain a unified and valid actual measurement dataset from multiple radars at the current moment.

[0035] like Figure 2 As shown in (b), in the multi-radar system of the present invention, multiple radar nodes are deployed at different physical locations in the room (such as different corners or ceilings) to conduct independent, multi-view joint observation of the same spatial scene. Figure 2 (a) is a traditional multiple-transmitter, multiple-receiver radar.

[0036] This invention provides a unified modeling of all possible one-to-one correspondences between cross-radar measurements. Assuming the presence of N observable targets within a scene, the cross-radar measurement correlation is modeled as a permutation mapping. The system maintains a parallel scale in the background. The global association hypothesis space, where each global association hypothesis is denoted as . For the actual measurement data of each radar acquired at the current moment, the actual measurement data is synchronized in time and transformed into a unified world coordinate system. Then, the system performs cross-radar measurement time modeling to generate candidate global correlation hypotheses corresponding to the current moment.

[0037] For example, suppose the actual measurement data of radar 1 are: A1, A2, A3; the actual measurement data of radar 2 are: B1, B2, B3; and the motion trajectories are: trajectory 1, trajectory 2, trajectory 3. Then, based on the actual measurement data of each radar, there are 6 candidate global association hypotheses generated, including: H1: Trajectory 1: [A1, B1]; Trajectory 2: [A2, B2]; Trajectory 3: [A3, B3]; H2: Trajectory 1: [A1, B2]; Trajectory 2: [A2, B1]; Trajectory 3: [A3, B3]; H3: Trajectory 1: [A1, B1]; Trajectory 2: [A2, B3]; Trajectory 3: [A3, B2]; H4: Trajectory 1: [A1, B3]; Trajectory 2: [A2, B2]; Trajectory 3: [A3, B1]; H5: Track 1: [A2, B3]; Track 2: [A1, B2]; Track 3: [A3, B1]; H6: Trajectory 1: [A3, B2]; Trajectory 2: [A2, B1]; Trajectory 3: [A1, B3].

[0038] In one implementation, step M200, if the current time is the initial time, the method includes: Step M201: Obtain the actual measurement data at each time within the initialization time window, and generate the corresponding initial global correlation hypothesis across radars based on the actual measurement data at each time. Step M202: Calculate the second cumulative integrated cost of each initial global association hypothesis at the end of the initialization time window; Step M203: Select an initial global association hypothesis based on each of the second cumulative comprehensive costs, and determine the motion trajectory at the initial time based on the selected initial global association hypothesis.

[0039] To prevent premature locking of erroneous trajectories due to measurement noise and clutter during the cold start phase, this invention sets the length to be... An initialization time window (typically 20 frames) is defined. Actual measurement data for each moment within the initialization time window is acquired, and global correlation hypothesis modeling is performed based on the actual measurement data for each moment to obtain the corresponding initial global correlation hypothesis.

[0040] After modeling the correlation assumptions of the measurement data, the actual measurement data within the initialization time window can be filtered, correlated, and fitted to obtain the motion trajectory corresponding to each observed target. Within the initialization time window, all initial global correlation assumptions and motion trajectories are maintained in parallel, and the cumulative cost is continuously calculated. At the end of the window, backtracking and optimization are performed to select the assumption with the minimum cumulative cost as the reliable starting state, i.e., the posterior state estimate at the initial time. Here, the initial time corresponds to the end frame of the initialization time window. For example, when the initialization time window is 20 frames, the initial time is the 20th frame.

[0041] In one implementation, the second cumulative integrated cost of each initialization global association hypothesis at the end of the initialization time window is calculated, including: The comprehensive cost function is used to calculate the comprehensive cost of each initial global association hypothesis within the initialization time window. The comprehensive cost includes one or more of the following: position consistency error, Doppler velocity error, and physical constraint penalty term. Based on the comprehensive cost at each time point, the cumulative comprehensive cost at the end of the initialization time window is calculated as the second cumulative comprehensive cost.

[0042] To evaluate the reasonableness of each global correlation hypothesis over time, this invention addresses each global correlation hypothesis... A comprehensive cost function was constructed, which integrates multiple constraints such as position, velocity, and physical rules: , in, For position consistency error, For Doppler velocity error, This is a physical constraint penalty term. and These represent the weight parameters, and Fixed weights based on statistical quantiles or adaptive weights can be used. and When using adaptive weights, the position consistency weights can be dynamically adjusted based on the current environmental clutter density and signal-to-noise ratio. Weight of physical penalty term The weighting ratio.

[0043] Position consistency error is used to evaluate the geometric error at the intersection of multiple radar measurements under the same global correlation assumption. This invention calculates the position consistency error using Euclidean distance based on measured and predicted position coordinates. , in, Indicates the measured position coordinates. This indicates the predicted location coordinates.

[0044] Doppler consistency error: At any time k, the velocity vector estimate of the target in the world coordinate system is obtained using the extended Kalman filter (EKF). The estimated vector velocity is calculated as a unit vector along the line-of-sight direction from both radars (Radar A and Radar B) to the estimated target location. and By performing spatial projection, the radial velocity predicted by the model is obtained: , Suppose that the measured radial velocities provided by the two radars at the same moment are respectively and For a single measurement correlation pair, its Doppler consistency residual is defined as the sum of the squares of the Euclidean distances between the measured radial velocity and the predicted radial velocity: .

[0045] Ultimately, under a certain global correlation hypothesis Within this framework, by summing the Doppler residuals of all associated pairs identified as the same target, the Doppler consistency error under this assumption can be obtained. Alternatively, depending on the degree of nonlinearity, unscented Kalman filtering (UKF) or particle filtering (PF) can be used to predict the observed target to obtain higher state estimation accuracy; if computing power is limited and the motion model is relatively simple, standard Kalman filtering (KF) can also be used in a reduced order.

[0046] Since the core of Doppler consistency lies in the fact that the real target follows the kinematic constraints of a single physical entity, while ghost points are mathematically spliced ​​solutions that violate the laws of kinematics, such as... Figure 5 As shown in (b), this illustrates the process of predicting false ghost points under erroneous association conditions. Figure 5 In Figure (a), the prediction process for the true target under correct correlation conditions is shown. Therefore, this method can be used as a "ghost filter." For a real physical target, it has a unique and objectively existing real velocity vector in space. When radar A and radar B observe it simultaneously, the radial velocity they measure... and Just the same real velocity vector From different perspectives and The physical projection of the object. Therefore, the velocity estimated by the extended Kalman filter. The projection equations in these two directions can be perfectly matched, resulting in the calculated residuals. Minimal (containing only the system's inherent measurement noise). For false targets (ghost points), a ghost point is essentially an incorrect cross-combination of radar A's observation of "target 1" and radar B's observation of "real target 2". In pure spatial geometry within a single frame, the ranging circles of these two measurements may indeed intersect at a point (i.e., the location of the ghost point). However, in the velocity dimension, radar A's measured radial velocity comes from the projection of "target 1's" true velocity, while radar B's measured radial velocity comes from the projection of "target 2's" true velocity. Since the true motion directions and speeds of target 1 and target 2 are usually quite different, the system attempts to estimate a unified virtual velocity vector for this incorrectly intersecting ghost point. However, under most non-degenerate geometric conditions, there is no single vector. This allows the projection of the radial velocity of target 1 at viewpoint A to simultaneously satisfy the condition that the radial velocity of target 2 is equal to the projection of target 1 at viewpoint B. Therefore, for the erroneous correlation assumption containing ghost points, a significant and irreconcilable contradiction inevitably arises between the predicted and measured radial velocities, leading to Doppler residuals. A punitive numerical surge occurred.

[0047] This invention introduces an extended Kalman filter to predict target velocities and projects them onto the line-of-sight directions of each radar, utilizing high-precision Doppler measurements from the radar as an independent verification standard. The continuous motion trajectories of real physical targets are highly consistent in the time dimension, and their projected Doppler closely matches the measured Doppler; however, incorrect combinations of ghost points inevitably produce huge Doppler residuals. This feature greatly expands the feature dimension of cross-radar measurement correlation, enabling the system to maintain high target discriminability even in extreme cases where purely geometric features fail (such as when multiple targets have highly overlapping spatial positions).

[0048] To enhance the algorithm's robustness in complex environments, a physical constraint penalty term is introduced. The physical constraint penalty term consists of an existence penalty. Spatial boundary penalty and human dynamics punishment The weighted summation consists of: , in , and These are the internal weighting coefficients for each sub-penalty term. The specific definitions of each sub-term are as follows: Existential penalty is used to suppress association hypotheses that cannot generate physical understanding; it indicates how many "false targets" exist in the global association hypothesis. If in the global association hypothesis... If, under this assumption, a feasible fusion location cannot be formed by combining cross-radar measurements of any target (e.g., geometrically unsolvable, numerically divergent, or clearly violating the basic measurement consistency constraint), then the target is considered to lack physical existence under this assumption. The existence penalty is defined as follows: , in This is an indicator function; it returns 1 when the condition is true and 0 when the condition is false. Used to represent the global association hypothesis Next, the Is the objective infeasible (not satisfying physical or geometric constraints)? This is a function or decision maker defined in this invention. An existence penalty term is used to impose a discrete penalty on infeasible associations, significantly reducing their likelihood of being selected in the global hypothesis decision.

[0049] Spatial boundary penalty: Consider the known physical boundaries of the indoor scene: , For any target under the global correlation assumption The estimated position (i.e., the prior state estimate, obtained based on Kalman filtering) is then determined. Define its in and The distance that crosses the boundary in the direction is: , , in, Spatial boundary penalty Defined as: , in, This is a normalized scale for position consistency error, used to ensure that the penalty term and the basic cost have similar dimensions.

[0050] To suppress the global correlation assumptions that violate the characteristics of reasonable human motion, a human dynamics penalty based on velocity, acceleration, and inter-frame displacement is introduced. Consider a human motion model involving a jogging scene, where the maximum permissible velocity, acceleration, and inter-frame displacement are respectively... , , During the sampling period Below, definition: , ， , in, and These are the normalized scales for velocity and acceleration, respectively. The dynamic penalty term is defined as follows: .

[0051] This invention, by constructing physical constraint penalty terms, can directly and effectively suppress "pseudo-solutions" that exist mathematically but are illegal in physical reality. Through the refined design of three comprehensive penalty terms—existence penalty, spatial boundary penalty, and human dynamics penalty—a systematic physical constraint mechanism is established. Combined with a penalty weight allocation mechanism, this feature ensures that any single significant physical violation will lead to a punitive increase in the overall cost of the assumption, thereby greatly enhancing the system's tracking robustness in unstructured environments and under high clutter interference.

[0052] Step S200: If the current time is not the initial time, obtain the posterior state estimate and target covariance matrix of the motion trajectory at the previous time, and determine the prior state estimate and prediction covariance matrix of each motion trajectory at the current time based on the posterior state estimate and target covariance matrix of each motion trajectory at the previous time.

[0053] After the parallel initialization phase, the system officially enters the online tracking mode. First, it reads the posterior state estimates of each motion trajectory obtained after the target tracking was completed at the previous moment (including state information such as position, velocity, and acceleration), as well as the target covariance matrix characterizing the error distribution and uncertainty of the posterior state estimates. Using an extended Kalman filter state transition model, it calculates the prior state vector and prediction covariance matrix of each target object at the current moment. This achieves the recursive prediction of the target state from the optimal estimate at the previous moment to the predicted state at the current moment, solving the problem of continuous temporal prediction of target motion states. It features low computational cost and high real-time performance, making it suitable for online real-time tracking of multiple trajectories and multiple targets.

[0054] Step S300: In each of the candidate global association hypotheses, based on the comprehensive cost function, the comprehensive cost is calculated according to the actual measurement data, each of the prior state estimates and the prediction covariance matrix. The comprehensive cost function is achieved by weighted summation of position consistency error, Doppler consistency error and physical constraint penalty term.

[0055] The actual measurement data corresponding to the same motion trajectory may differ under different candidate global correlation hypotheses. For each candidate global correlation hypothesis, the comprehensive cost function described above is used to calculate the comprehensive cost corresponding to that candidate global correlation hypothesis based on the actual measurement data, the prior state estimates of each motion trajectory, and the predicted covariance matrix.

[0056] For example, candidate global association hypothesis H1: Trajectory 1: [A1,B1]; Trajectory 2: [A2,B2]; Trajectory 3: [A3,B3]; candidate global association hypothesis H2: Trajectory 1: [A1,B2]; Trajectory 2: [A2,B1]; Trajectory 3: [A3,B3]; then, for the comprehensive cost of candidate global association hypothesis H1, the comprehensive cost of trajectory 1 can be calculated based on the prior state estimates and predicted covariance matrices of the actual measurement data A1, B1 and trajectory 1, and the comprehensive cost of trajectory 2 can be calculated based on the prior state estimates and predicted covariance matrices of the actual measurement data A2, B2 and trajectory 2; similarly, the comprehensive cost of trajectory 3 can be calculated; combining the comprehensive costs of each estimate yields the comprehensive cost of candidate global association hypothesis H1. For candidate global association hypothesis H2, the actual measurement data corresponding to trajectory 1 are A1 and B2. Then, the comprehensive cost of trajectory 1 is calculated based on the actual measurement data A1 and B2 and the prior state estimation and prediction covariance matrix of trajectory 1. Similarly, the comprehensive cost of trajectory 2 and trajectory 3 in candidate global association hypothesis H2 is calculated, and thus the comprehensive cost of candidate global association hypothesis H2 is obtained.

[0057] In one implementation, for each of the candidate global association hypotheses, a comprehensive cost is calculated based on the comprehensive cost function, according to the actual measurement data, each of the prior state estimates, and the prediction covariance matrix, including: Step S301: Determine the optimal trace matching scheme for each radar based on the actual measurement data of each radar at the current time, the prior state estimates, and the prediction covariance matrix. Step S302: In each of the candidate global association hypotheses, update each motion trajectory according to each optimal trajectory matching scheme, and determine the posterior state estimate and target covariance matrix of the motion trajectory at the current time. Step S303: Calculate the comprehensive cost of the candidate global association hypothesis using the comprehensive cost function based on the prior state estimate, the predicted covariance matrix, the posterior state estimate, and the target covariance matrix.

[0058] First, for each radar, the actual measurement data is matched with each motion trajectory. By combining the actual measurement data of each radar at the current moment with the prior state estimate and prediction covariance matrix of each motion trajectory at the current moment, the matching cost between the actual measurement data and the motion trajectory is calculated. Based on each matching cost, the optimal matching scheme for that radar is obtained.

[0059] For example, given hypothesis H1: Radar A's measurement point data: A1, A2, A3; Radar B's measurement point data: B1, B2, B3. Within a single radar, the trajectory corresponding to each point is found. That is, for each existing historical motion trajectory 1, 2, 3, a prediction is first made using the posterior state of the previous moment to obtain the predicted state at the current moment. Then, the matching cost between each point and each trajectory is calculated to obtain the set of points for each trajectory. For example, for Radar A, if the allocation of points A1, A2, A3 to trajectories 1, 2, and 3 results in the lowest matching cost, then this matching scheme is the optimal matching.

[0060] After obtaining the optimal trajectory matching scheme for each radar, the optimal trajectory matching scheme provides a source for the matching status between each actual measurement data and each trajectory in the candidate global association, as well as the basic matching cost. Thus, based on the matching status, the actual measurement data in the candidate global association hypothesis can be used to update the corresponding motion trajectory, and the posterior state estimate of the motion trajectory at the current time and the target covariance matrix can be obtained.

[0061] Based on each candidate global association hypothesis, the state residuals and uncertainty change costs before and after trajectory update are calculated based on the state difference between the prior and posterior state estimates of the motion trajectory, the change in filtering error reflected by the prediction covariance matrix and the target covariance matrix. At the same time, the comprehensive cost is obtained by weighted fusion of the matching residuals between measurement and trajectory, cross-radar consistency constraints and target motion physical constraints, which is used to quantitatively evaluate the rationality and reliability of each global association hypothesis.

[0062] In one implementation, determining the optimal trajectory matching scheme for each radar based on the actual measurement data of each radar at the current time, the prior state estimates, and the predicted covariance matrix includes: Step S3011: Map each of the prior state estimates to the measurement space and determine the corresponding predicted measurement values; Step S3012: Calculate the matching cost between each actual measurement data and each predicted measurement value for each radar based on the predicted covariance matrix. Step S3013: Determine the optimal trace matching scheme for each radar based on the matching costs of each radar.

[0063] Specifically, the prior state is mapped to the measurement space through the radar observation model, and the corresponding predicted measurement value is calculated. Then, based on the comprehensive cost function constructed above, the matching cost between each actual measurement data and the predicted measurement value (or predicted motion trajectory) in the current moment is calculated one by one, and each actual measurement data is matched with each motion trajectory according to the matching cost; finally, the measurement matching results of each radar are integrated and verified to ensure that a single measurement belongs to only a single target trajectory, thereby forming the optimal trajectory matching scheme between the actual measurement data corresponding to each radar and each motion trajectory.

[0064] In one implementation, determining the optimal track matching scheme for each radar based on the respective matching costs of each radar includes: Step S30131: Arrange and combine the matching costs to determine the two-dimensional cost matrix; Step S30132: Convert the two-dimensional cost matrix into a bipartite graph matching problem, and call the Hungarian algorithm to perform global optimization to solve the bipartite graph matching problem, and determine the optimal trace matching scheme for each radar.

[0065] like Figure 4 As shown, the calculated matching costs are arranged and combined to construct a two-dimensional cost matrix reflecting the suitability of global allocation. For "measurement-trajectory" pairs with extremely high physical constraint penalties, the system directly assigns an infinite value to the two-dimensional cost matrix to prevent such illegal matching from occurring.

[0066] After obtaining the two-dimensional cost matrix, it is transformed into a typical bipartite graph matching problem, and the Hungarian algorithm is used for global optimization to find the solution. This algorithm is able to... With a time complexity of (where N is the number of targets or measurements), this invention quickly finds the globally optimal allocation scheme that minimizes the sum of the cost matrix, serving as the optimal trajectory matching scheme between the actual measurements and the motion trajectories. In the solution process, this invention calculates the matching cost between the actual measurement data of each radar and each predicted measurement value in parallel, and then determines the matching scheme between the actual measurement data of each radar and each motion trajectory based on each matching cost. Compared to the traditional Nearest Neighbor Association (GNN) algorithm, the Hungarian algorithm completely avoids the "local optima leading to global degradation" phenomenon caused by multiple targets being too close together, ensuring the allocation accuracy in dense multi-target scenarios.

[0067] Alternatively, in scenarios with a very large number of targets and higher real-time requirements, the Hungarian algorithm can be replaced with the Jonker-Volgenant (JV) algorithm, the Munkres algorithm, or a suboptimal greedy nearest neighbor association (GNN) algorithm to reduce the time complexity of solving the bipartite graph matching problem.

[0068] In one implementation, in each of the candidate global association hypotheses, each motion trajectory is updated according to each optimal trajectory matching scheme to determine the posterior state estimate and target covariance matrix of the motion trajectory at the current time, including: Step S3021: Determine whether each motion trajectory matches the actual measurement data according to each optimal trajectory matching scheme; Step S3022: If yes, calculate the Kalman gain corresponding to each actual measurement data in each of the candidate global association hypotheses, correct the prior state estimate of the corresponding motion trajectory based on each Kalman gain and each actual measurement data, and determine the posterior state estimate of the motion trajectory and the target covariance matrix at the current moment. Step S3023: If not, then the prior state and predicted covariance matrix of the motion trajectory are used as the posterior state estimate and target covariance matrix of the motion trajectory at the current moment.

[0069] Specifically, according to the optimal trajectory matching scheme, this invention maintains independent trajectory state branches for different candidate global association hypotheses, and the updated states are not shared between branches. In each candidate global association hypothesis, cross-radar correlation measurements belonging to the same target are jointly updated by the same filter. For each motion trajectory successfully matched to the actual measurement data, the actual measurement data is substituted into the update equation of the extended Kalman filter to calculate the Kalman gain. Based on the actual measurement data and the Kalman gain, the prior state estimate and predicted covariance matrix of the motion trajectory at the current time are corrected to obtain the posterior state estimate and the updated target covariance matrix at the current time.

[0070] For motion trajectories that fail to match actual measurement data, they are marked as "missed" and only their prior state estimate is retained as the posterior state estimate at the current moment (i.e., pure prediction without update), and their "continuous lost frame count" counter is incremented; if the counter exceeds a preset threshold, it is determined that the target has left the observation area and the motion trajectory is deregistered from the system.

[0071] For isolated measurements that fail to be assigned to any trajectory, the system treats them as potential emerging targets or sporadic clutter and sends them to the starting candidate queue for subsequent trajectory initiation logic judgment.

[0072] Step S400: Calculate the first cumulative comprehensive cost of each candidate global association hypothesis within the sliding time window, and determine the optimal global association hypothesis among the candidate global association hypotheses based on the first cumulative comprehensive cost.

[0073] To suppress the impact of single-frame observation noise and sporadic correlation errors on the output results, this invention designs a sliding time window decision mechanism of length L to evaluate candidate global correlation hypotheses for each radar and select the optimal global correlation hypothesis. This mechanism does not make decisions based on the instantaneous cost at the current moment, but rather filters the optimal hypothesis by evaluating the first cumulative cost of candidate global correlation hypotheses within the sliding time window, selecting the candidate global correlation hypothesis with the smallest first cumulative comprehensive cost. Here, the first cumulative comprehensive cost corresponding to a candidate global correlation hypothesis is the comprehensive cost accumulated from each moment within the sliding time window to the current moment.

[0074] Define candidate global association hypotheses At any moment k First cumulative cost for: , in, for The single-frame synthesis cost of the candidate global association hypothesis at time step [time]. Then, the optimal global association hypothesis output at the current time step [time]. satisfy: .

[0075] This invention employs the cumulative cost of a sliding time window to screen the optimal global association hypothesis, upgrading the decision-making basis from "single-frame instantaneous cost" to "multi-frame historical cumulative cost." This mechanism effectively resists observation jitter and occasional clutter interference in single frames. Through cumulative consistency optimization based on historical windows, the system can keenly capture and automatically remove occasional erroneous association paths that are "geometrically feasible in the short term but lack long-term temporal continuity," ultimately ensuring that the output true target trajectory has extremely high coherence and smoothness.

[0076] Optionally, for the sliding time window, the window length L can be dynamically adjusted based on real-time indicators such as the motion maneuverability of the observed target (e.g., the degree of acceleration abrupt change or the magnitude of the predicted residual), the spatial density of environmental clutter, or the measurement signal-to-noise ratio. Through this adaptive adjustment, the system can achieve a dynamic optimal balance between anti-interference robustness and tracking real-time performance.

[0077] Step S500: Use the motion trajectory of the optimal global association hypothesis as the motion trajectory at the current moment.

[0078] Since the motion trajectories in the candidate global correlation hypotheses are updated motion estimates for the current moment, the motion trajectory of the optimal global correlation hypothesis selected from all candidate global correlation hypotheses will be used as the motion trajectory of each target object at the current moment.

[0079] This invention constructs a complete framework for ghost point discrimination and stable tracking from the perspective of global spatiotemporal consistency. It not only fundamentally solves the industry pain point of multi-base station radar systems being prone to erroneous measurement combinations when multiple targets are densely intersecting and spatially close, but also effectively identifies and suppresses false physical targets induced by erroneous associations without increasing additional hardware costs. This provides a stable measurement foundation for advanced multimodal data fusion and scene semantic analysis in the backend.

[0080] This invention constructs a complete space containing N! global correlation hypotheses. However, for dense multi-target or large-area wide-area monitoring scenarios, the size of the hypothesis space explodes exponentially with the increase of the number of targets N, severely limiting hardware computing power. To overcome the computing power bottleneck and significantly increase the number of concurrent multi-targets applicable to the system, the benchmark global parallel solution method can be optimized and transformed into a multi-hypothesis tracking algorithm incorporating pruning strategies. For example... Figure 6 As shown, Figure 6 (a) shows the relationship between tracking accuracy and the number of targets. Figure 6 Figure (b) shows the relationship between computational complexity and the number of targets. In practical implementation, instead of retaining all combinations, the size of the hypothesis tree is controlled by introducing dimensionality reduction and pruning strategies (for example, pruning strategies are used to remove low-confidence hypotheses with probabilities below a set threshold, retain high-probability hypothesis branches, merge redundant hypotheses with similar probabilities, and then update the target trajectory state using measurement data based on the remaining valid hypotheses). This alternative solution method based on a pruned version of MHT (Multi-Hypothesis Tracking) can accurately retain high-probability real physical association paths while effectively controlling memory consumption and computational time complexity, thereby achieving real-time stable tracking in large-scale dense target scenarios.

[0081] Based on the above embodiments, the present invention also provides an indoor multi-target detection device based on multiple radars, such as... Figure 7 As shown, the device includes: Assumption generation module 01 is used to obtain the actual measurement data of each radar at the current time, and generate each candidate global correlation hypothesis across radars based on the actual measurement data. The state estimation module 02 is used to obtain the posterior state estimate and target covariance matrix of the motion trajectory at the previous time if the current time is not the initial time, and to determine the prior state estimate and prediction covariance matrix of each motion trajectory at the current time based on the posterior state estimate and target covariance matrix of the previous time. Cost calculation module 03 is used to calculate the comprehensive cost in each of the candidate global association hypotheses based on the comprehensive cost function, according to the actual measurement data, each of the prior state estimates and the prediction covariance matrix. The comprehensive cost function is implemented by weighted summation of position consistency error, Doppler consistency error and physical constraint penalty term. Assuming the filtering module 04 is used to calculate the first cumulative comprehensive cost of each of the candidate global association hypotheses within the sliding time window, and to determine the optimal global association hypothesis among the candidate global association hypotheses based on the first cumulative comprehensive cost; The trajectory determination module 05 is used to take the motion trajectory of the optimal global association hypothesis as the motion trajectory at the current moment.

[0082] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 8 As shown, the terminal includes a processor, memory, network interface, and display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-radar-based indoor multi-target detection method. The display screen can be an LCD screen or an e-ink screen.

[0083] Those skilled in the art will understand that Figure 8 The schematic diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0084] In one implementation, the terminal's memory stores one or more programs, and these programs are configured to be executed by one or more processors, and the programs contain instructions for performing a multi-radar-based indoor multi-target detection method.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0086] In summary, this invention discloses an indoor multi-target detection method, device, terminal, and medium based on multiple radars. The method generates candidate global association hypotheses across radars based on actual measurement data at the current moment; predicts prior state estimates and predicted covariance matrices for each trajectory at the current moment based on the posterior state estimate and target covariance matrix of the previous moment's trajectory; in the candidate global association hypotheses, a comprehensive cost function is calculated based on actual measurement data, prior state estimates, and the predicted covariance matrix, with the comprehensive cost function achieved by weighted summation of position consistency error, Doppler consistency error, and physical constraint penalty terms; the cumulative comprehensive cost of each candidate global association hypothesis is calculated within a sliding time window, and the trajectory of the candidate global association hypothesis is selected as the trajectory at the current moment based on the cumulative comprehensive cost. This effectively solves the problem of existing technologies neglecting trajectory time consistency and motion direction constraints, leading to target tracking loss and false alarms.

[0087] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for indoor multi-target detection based on multiple radars, characterized in that, The method includes: Obtain the actual measurement data of each radar at the current moment, and generate each candidate global correlation hypothesis across radars based on the actual measurement data; If the current time is not the initial time, obtain the posterior state estimate and target covariance matrix of the motion trajectory at the previous time, and determine the prior state estimate and prediction covariance matrix of each motion trajectory at the current time based on the posterior state estimate and target covariance matrix of the previous time. In each of the candidate global association hypotheses, a comprehensive cost is calculated based on the comprehensive cost function, the actual measurement data, each of the prior state estimates and the prediction covariance matrix. The comprehensive cost function is achieved by weighted summation of the positional consistency error, the Doppler consistency error and the physical constraint penalty term. Calculate the first cumulative comprehensive cost of each of the candidate global association hypotheses within the sliding time window, and determine the optimal global association hypothesis among the candidate global association hypotheses based on the first cumulative comprehensive cost; The trajectory of the motion under the optimal global association hypothesis is taken as the trajectory of the motion at the current moment.

2. The indoor multi-target detection method based on multiple radars according to claim 1, characterized in that, If the current time is the initial time, the method includes: Acquire the actual measurement data at each moment within the initialization time window, and generate the corresponding initial global correlation hypothesis across radars based on the actual measurement data at each moment; Calculate the second cumulative integrated cost of each initialization global association hypothesis at the end of the initialization time window; Based on each of the second cumulative comprehensive costs, an initial global correlation hypothesis is selected, and the motion trajectory at the initial time is determined based on the selected initial global correlation hypothesis.

3. The indoor multi-target detection method based on multiple radars according to claim 1, characterized in that, In each of the candidate global association hypotheses, based on the comprehensive cost function, the comprehensive cost is calculated according to the actual measurement data, each of the prior state estimates, and the prediction covariance matrix, including: The optimal trace matching scheme for each radar is determined based on the actual measurement data of each radar at the current time, the prior state estimates, and the predicted covariance matrix. In each of the candidate global association hypotheses, each motion trajectory is updated according to the optimal trajectory matching scheme to determine the posterior state estimate and target covariance matrix of the motion trajectory at the current time. The comprehensive cost of the candidate global association hypothesis is calculated using a comprehensive cost function based on the prior state estimate, the predicted covariance matrix, the posterior state estimate, and the target covariance matrix.

4. The indoor multi-target detection method based on multiple radars according to claim 3, characterized in that, The optimal trajectory matching scheme for each radar is determined based on the actual measurement data of each radar at the current time, the prior state estimates, and the predicted covariance matrix, including: Each of the prior state estimates is mapped to the measurement space to determine the corresponding predicted measurement value; Based on the predicted covariance matrix, the matching cost between each actual measurement data and each predicted measurement value of each radar is calculated one by one. The optimal trace matching scheme for each radar is determined based on the matching cost described for each radar.

5. The indoor multi-target detection method based on multiple radars according to claim 4, characterized in that, Determining the optimal track matching scheme for each radar based on its respective matching costs includes: The matching costs are arranged and combined to determine a two-dimensional cost matrix; The two-dimensional cost matrix is ​​transformed into a bipartite graph matching problem, and the Hungarian algorithm is called to perform global optimization to solve the bipartite graph matching problem, thereby determining the optimal trace matching scheme for each radar.

6. The indoor multi-target detection method based on multiple radars according to claim 3, characterized in that, In each of the candidate global association hypotheses, each trajectory is updated according to the optimal trajectory matching scheme to determine the posterior state estimate and target covariance matrix of the trajectory at the current time, including: Determine whether each motion trajectory matches the actual measurement data based on the optimal trajectory matching scheme described above; If so, calculate the Kalman gain corresponding to each actual measurement data in each candidate global association hypothesis, correct the prior state estimate of the corresponding motion trajectory based on each Kalman gain and each actual measurement data, and determine the posterior state estimate of the motion trajectory and the target covariance matrix at the current moment. If not, the prior state and predicted covariance matrix of the motion trajectory are used as the posterior state estimate and target covariance matrix of the motion trajectory at the current moment.

7. The indoor multi-target detection method based on multiple radars according to claim 1, characterized in that, The physical constraint penalty term consists of a weighted sum of existence penalty, spatial boundary penalty, and human dynamics penalty.

8. An indoor multi-target detection device based on multiple radars, characterized in that, The device includes: The hypothesis generation module is used to obtain the actual measurement data of each radar at the current time, and generate candidate global correlation hypotheses across radars based on the actual measurement data. The state estimation module is used to obtain the posterior state estimate and target covariance matrix of the motion trajectory at the previous time if the current time is not the initial time, and to determine the prior state estimate and prediction covariance matrix of each motion trajectory at the current time based on the posterior state estimate and target covariance matrix of each motion trajectory at the previous time. The cost calculation module is used to calculate the comprehensive cost in each of the candidate global association hypotheses based on the comprehensive cost function, according to the actual measurement data, each of the prior state estimates and the prediction covariance matrix. The comprehensive cost function is implemented by weighted summation of the position consistency error, Doppler consistency error and physical constraint penalty term. The hypothesis filtering module is used to calculate the first cumulative comprehensive cost of each of the candidate global association hypotheses within the sliding time window, and to determine the optimal global association hypothesis among the candidate global association hypotheses based on the first cumulative comprehensive cost. The trajectory determination module is used to take the motion trajectory of the optimal global association hypothesis as the motion trajectory at the current moment.

9. A terminal, characterized in that, The terminal includes a memory and one or more processors; the memory stores one or more programs; the programs contain instructions for executing the multi-radar-based indoor multi-target detection method as described in any one of claims 1-7; the processors are used to execute the programs.

10. A computer-readable storage medium storing a plurality of instructions thereon, characterized in that, The instructions are loaded and executed by the processor to implement the steps of the multi-radar-based indoor multi-target detection method according to any one of claims 1-7.

Images