Target tracking method and related equipment

Abstract

The embodiment of the invention discloses a target body tracking method and related equipment, which are used for tracking a target body under the condition of improving the accuracy, efficiency and real-time performance of target body tracking in a complex scene. The method comprises the following steps: acquiring multi-frame UWB radar information, detecting the multi-frame UWB radar information by using a double-threshold detection algorithm, determining whether a true target body exists or not, determining a target track of the true target body based on a target track confirmation threshold, and determining the target track of the true target body according to the target track confirmation threshold. Determining a plurality of initial association hypotheses of the true target body based on the target track of the true target body by using an MHT multi-hypothesis tracking algorithm, determining a target association hypothesis in the plurality of initial association hypotheses, and generating a target track corresponding to the target association hypothesis, and dynamically adjusting the weight of each sub-model in the IMM model based on the maneuvering characteristics of the true target body in the target track corresponding to the target association hypothesis by using the IMM model, and tracking the true target body based on the target track corresponding to the target association hypothesis by using the adjusted IMM model.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of target tracking, and more particularly, to a target tracking method, a target tracking device, an electronic device, and a computer readable storage medium. BACKGROUND

[0002] With the development of intelligent transportation, autonomous driving, and unmanned systems, the demand for tracking technology that can accurately locate the target position and speed in real time is rapidly increasing.

[0003] The existing target tracking method is to use only radar technology to calculate the position and speed of the target by measuring the time, frequency, and amplitude of the reflected signal, and to achieve tracking function.

[0004] However, the existing technology has obvious deficiencies: on the one hand, the tracking process is often accompanied by a large delay, which cannot reflect the real-time state of the target in time; on the other hand, the utilization efficiency of the Doppler effect is low, resulting in inaccurate speed measurement. In summary, the existing technology has low accuracy, efficiency, and real-time performance in complex scenarios. SUMMARY

[0005] Embodiments of the present application provide a target tracking method, a target tracking device, an electronic device, and a computer readable storage medium, which are used to track the target in a complex scene to improve the accuracy, efficiency, and real-time performance of target tracking.

[0006] In a first aspect, the embodiments of the present application provide a target tracking method, comprising:

[0007] Obtaining a plurality of UWB radar information;

[0008] Detecting the plurality of UWB radar information using a double-threshold detection algorithm to determine whether there is a true target, and if it is determined that there is a true target, determining a target track of the true target based on a target track confirmation threshold;

[0009] Determining a plurality of initial association hypotheses of the true target based on the target track of the true target using an MHT multi-hypothesis tracking algorithm, determining a target association hypothesis in the plurality of initial association hypotheses, and generating a target track corresponding to the target association hypothesis;

[0010] Adjusting the weight of each sub-model in the IMM model based on the maneuvering characteristics of the true target in the target track corresponding to the target association hypothesis using an IMM model, to obtain an adjusted IMM model;

[0011] The tracking unit predicts the motion state of the true target body based on the target track corresponding to the target association hypothesis by using the adjusted IMM model, so as to perform tracking of the true target body.

[0012] In a second aspect, an embodiment of the present application provides a target body tracking device, comprising:

[0013] A obtaining unit is configured to obtain multiple frames of UWB radar information.

[0014] A determining unit is configured to detect whether there is a true target body by using a double-threshold detection algorithm, and determine a target track of the true target body based on a target track confirmation threshold if it is determined that there is a true target body.

[0015] The determining unit is further configured to determine multiple initial association hypotheses of the true target body based on the target track of the true target body by using an MHT multi-hypothesis tracking algorithm, determine a target association hypothesis from the multiple initial association hypotheses, and generate a target track corresponding to the target association hypothesis.

[0016] An adjusting unit is configured to dynamically adjust weights of each sub-model in an IMM model based on a maneuvering characteristic of the true target body in the target track corresponding to the target association hypothesis by using the IMM model, so as to obtain an adjusted IMM model.

[0017] A tracking unit is configured to predict the motion state of the true target body based on the target track corresponding to the target association hypothesis by using the adjusted IMM model, so as to perform tracking of the true target body.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, comprising:

[0019] A central processing unit, a memory, an input and output interface, a wired or wireless network interface, and a power supply.

[0020] The memory is a transitory storage memory or a persistent storage memory.

[0021] The central processing unit is configured to communicate with the memory, and perform instruction operation in the memory to execute the foregoing target body tracking method.

[0022] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which comprises instructions, and when the instructions are run on a computer, the computer executes the foregoing target body tracking method.

[0023] In a fifth aspect, an embodiment of the present application provides a computer program product comprising instructions, and when the computer program product is run on a computer, the computer executes the foregoing target body tracking method.

[0024] From the above technical solutions, the embodiments of the present application have the following advantages: by combining the high-precision measurement of the UWB radar and the efficient screening of the double-threshold detection algorithm, the true target body can be accurately identified, false information can be effectively eliminated, and the detection accuracy can be improved; by using the MHT multi-hypothesis tracking algorithm, the multiple possible association paths of the target are skillfully managed, the pruning strategy is used to remove low-probability hypotheses, and the high-quality path is focused, so that the calculation complexity can be reduced and the tracking efficiency can be improved; by means of the adaptive dynamic adjustment capability of the IMM model, the complex motion state of the target can be accurately matched, the prediction result can be optimized in real time, and the effect of tracking the maneuvering target can be enhanced. In summary, the target body tracking method combining the MHT multi-hypothesis tracking algorithm and the IMM model can realize real-time tracking of the UWB radar in a complex environment, reduce tracking delay, improve Doppler utilization, and significantly improve the accuracy, efficiency and real-time performance of target body tracking in a complex scene.

[0025] Correspondingly, the target body tracking device, the electronic device, the computer readable storage medium and the computer program product containing instructions provided by the present application also have the above technical effects. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 An architecture schematic diagram of a target body tracking system disclosed by the embodiments of the present application;

[0027] Figure 2 A flowchart of a target body tracking method disclosed by the embodiments of the present application;

[0028] Figure 2-1 A schematic diagram of a trajectory management mechanism disclosed by the embodiments of the present application;

[0029] Figure 2-2 A method flowchart of a pruning strategy disclosed by the embodiments of the present application;

[0030] Figure 2-3 A first and second threshold detection result schematic diagram disclosed by the embodiments of the present application;

[0031] Figure 2-4 An IMM multi-model Kalman detection result schematic diagram disclosed by the embodiments of the present application;

[0032] Figure 2-5 An overall position deviation result schematic diagram disclosed by the embodiments of the present application;

[0033] Figure 3 A structure schematic diagram of a target body tracking device disclosed by the embodiments of the present application;

[0034] Figure 4 A structure schematic diagram of an electronic device disclosed by the embodiments of the present application. Detailed Implementation

[0035] This application provides a target tracking method, a target tracking device, an electronic device, and a computer-readable storage medium for tracking targets in complex scenarios while improving the accuracy, efficiency, and real-time performance of target tracking.

[0036] Please see Figure 1 The architecture of the target tracking system in this application embodiment includes:

[0037] UWB module 101 and MCU module 102. When tracking a target, UWB module 101 can connect to MCU module 102. UWB module 101 can perform RF radio frequency and signal processing obstacle detection, obtain multiple frames of UWB radar information, and report this information to MCU module 102 via SPI or CAN bus. MCU module 102 can use a dual-threshold detection algorithm to detect the acquired multiple frames of UWB radar information, determine the presence of a real target, and determine the target trajectory of the real target based on a target trajectory confirmation threshold. Then, using the MHT (Multi-Hypothesis Tracking) algorithm, it determines multiple initial association hypotheses of the real target based on the target trajectory. From these initial hypotheses, it identifies the target association hypothesis and generates the target trajectory corresponding to the hypothesis. Next, using the IMM (Integrated Model) model, it dynamically adjusts the weights of each sub-model in the IMM model based on the maneuverability of the real target in the target trajectory corresponding to the association hypothesis. Finally, using the adjusted IMM model, it predicts the motion state of the real target based on the target trajectory corresponding to the association hypothesis for tracking.

[0038] In other words, in this embodiment of the application, the MCU terminal 102 can perform track initiation and track management (including initialization, confirmation of target trajectory, maintenance of trajectory life cycle, etc.), MHT-based data interconnection (solving the trajectory association problem under multiple targets and false alarm interference), and IMM-based maneuvering target tracking (performing maneuver detection and model switching prediction for specific tracks).

[0039] based on Figure 1 Please refer to the target tracking system shown. Figure 2 , Figure 2 This is a flowchart illustrating a target tracking method disclosed in an embodiment of this application. The method includes:

[0040] 201. Acquire multiple frames of UWB radar information.

[0041] In an optional embodiment, the present application is applicable to the fields of intelligent transportation, autonomous driving, unmanned systems, etc., and can be applied in simple or complex scenarios (such as high-density targets and high-maneuverability scenarios, dense crowds, and trajectory intersection situations). The UWB-specific parameters are as follows: the working frequency band is 3.1 GHz-10.6 GHz, the present application uses CH9 7.9 GHz; the bandwidth ratio is greater than 20%, typically 500 MHz; the speed resolution formula is wherein λ is the wavelength (37.97 mm), Tcoh is the coherent processing time, 128-point FFT is used, 128 CIRs are required, the coherent processing time is 128*1 ms, and a speed resolution of 14.8 cm / s can be obtained. The distance resolution formula is wherein c is the speed of light ≈3*10^8 m / s, B is the bandwidth 500 MHz, the distance resolution is 30 cm, and the update step of the ranging is half of the distance resolution, i.e., 15 cm. The pulse compression system uses pulse coding combined with pulse compression processing to avoid the typical leakage problem of FMCW, which can greatly alleviate the problems of track breakage and false association of traditional radars in complex scenarios (multiple targets / obstructions / maneuverability).

[0042] 202. Detecting multiple frames of UWB radar information by using a double-threshold detection algorithm to determine whether there is a true target body, and if it is determined that there is a true target body, determining the target track of the true target body based on a target track confirmation threshold.

[0043] In an optional embodiment, the true target body includes target bodies such as vehicles and pedestrians, which are not limited here. The double-threshold detection algorithm is used to detect whether the speed and / or distance change degree in the multiple frames of UWB radar information reaches a preset threshold, and whether the speed fluctuation degree is within a preset range, to determine whether there is a true target body. The target track confirmation threshold includes the number of target continuous detection frames T1 required for track confirmation and the number of target continuous non-detection frames T2 required for track discard, which are used to determine whether to create or maintain a target track. The track of the true target body includes the motion track of the true target body in the past period of time, including the record of the change of the position, speed, etc. with time.

[0044] 203. Determining multiple initial association hypotheses of the true target body based on the target track of the true target body by using an MHT multi-hypothesis tracking algorithm, determining a target association hypothesis in the multiple initial association hypotheses, and generating a target track corresponding to the target association hypothesis.

[0045] In an optional embodiment, the multiple initial association hypotheses are multiple possible target association hypotheses generated by the MHT algorithm, each hypothesis representing a possible motion path of the target or an association manner with the observation data. The target track corresponding to the target association hypothesis is a motion track of the target body generated according to the selected target association hypothesis.

[0046] 204. Using the IMM model, the weights of each sub-model in the target trajectory corresponding to the target association assumption are dynamically adjusted based on the maneuvering characteristics of the real target body, to obtain the adjusted IMM model.

[0047] In one optional implementation, the IMM model comprises a constant velocity sub-model, a constant acceleration sub-model, and a uniform rotor model. The weights of each sub-model are dynamically adjusted to adapt to the target's maneuvering characteristics. The true target's maneuvering characteristics refer to motion state information such as changes in acceleration and / or turning.

[0048] 205. Using the adjusted IMM model, the motion state of the real target body is predicted based on the target trajectory corresponding to the target association assumption, so as to track the real target body.

[0049] In one alternative implementation, for example, in traffic monitoring, UWB radar is used to track vehicles. After acquiring multiple frames of radar information, a dual-threshold detection algorithm is used to determine whether a vehicle exists. If it exists, its trajectory is determined. Then, the MHT algorithm is used to generate multiple initial association hypotheses, and the most likely hypothesis is selected to generate the trajectory. Finally, the IMM model is used to adjust the weights and predict the vehicle's motion state.

[0050] Thus, by combining the high-precision measurement of UWB radar with the efficient screening of the dual-threshold detection algorithm, true targets can be accurately identified, false information can be effectively eliminated, and detection accuracy can be improved. The MHT (Multi-Hypothesis Tracking) algorithm cleverly manages multiple possible associated paths of the target, uses pruning strategies to remove low-probability hypotheses, and focuses on high-quality paths, reducing computational complexity and improving tracking efficiency. Leveraging the adaptive dynamic adjustment capability of the IMM (Integrated Model) model, it accurately matches the complex motion state of the target, optimizes prediction results in real time, and enhances the tracking effect of maneuvering targets. In summary, this target tracking method combining the MHT and IMM models can achieve real-time tracking with a single UWB radar at a density of >10 targets / km², with a tracking delay of ≤3 frames for maneuvering targets (acceleration ≥10m / s²) and a Doppler utilization rate of ≥80% (compared to ≤65% for traditional methods). In other words, this application can achieve real-time tracking of UWB radar in complex environments, reduce tracking delay, improve Doppler utilization, and significantly improve the accuracy, efficiency, and real-time performance of target tracking in complex scenarios.

[0051] In an optional embodiment, the double-threshold detection algorithm includes a first threshold detection algorithm and a second threshold detection algorithm, and the double-threshold detection algorithm is used to detect the multiple frames of UWB radar information to determine whether there is a true target object, including: using the first threshold detection algorithm to detect whether the speed and / or distance change degree in the multiple frames of UWB radar information reaches a preset speed and / or distance change degree threshold, to determine whether there is a target object, and using the second threshold detection algorithm to determine whether the speed fluctuation degree in the multiple frames of UWB radar information is within a preset speed fluctuation degree range, to determine whether the target object is a true target object, to determine whether there is a true target object.

[0052] Specifically, the target object tracking method of the present application involves track initiation and track management, and the trigger logic is that when a target with a speed / distance change that is not noise is detected in a plurality of continuous frames (target object), a global track template attempts to match the detection point. Specifically, a double-threshold detection algorithm (including a first threshold detection algorithm and a second threshold detection algorithm) is used. The first threshold detection algorithm is based on the Neyman-Pearson criterion for detection, which is a core method of statistical detection theory. Assuming “no target”, “target”, the criterion aims to maximize the detection rate (P_D) under a limited false alarm rate (P_FA). A likelihood ratio test is established Set threshold η. If Λ(z)>η, it is judged as a target, and the threshold η is based on the background power estimate value (actual signal-to-noise ratio). The second threshold detection algorithm is based on the consistency test of the Doppler shift in the sliding window. The estimated target Doppler frequency (i.e. speed) of each frame is analyzed in the sliding window to determine whether it is continuous and stable (such as RMS fluctuation < threshold), a window width W (such as 5 frames) is set, and the calculation is as formula one:

[0053] Formula one

[0054] If σv<δv, it is considered that the Doppler is stable, and it is a true target.

[0055] In this way, the first threshold detection algorithm is used to preliminarily judge based on the speed and / or distance change degree. If the speed or distance change exceeds the preset threshold, it is considered that there may be a target object. This step can quickly screen out potential targets and reduce the amount of data for subsequent processing. The second threshold detection algorithm is used for further confirmation based on the speed fluctuation degree. If the speed fluctuation is within a reasonable range, it is considered to be a true target object, otherwise it may be noise or interference, which can effectively distinguish true targets from false targets and improve the accuracy of detection. Through double judgment, the false positive rate can be effectively reduced, and the reliability of target detection can be improved.

[0056] In an optional embodiment, the target track confirmation threshold value comprises a target continuous detection frame number T1 required for track confirmation and a target continuous non-detection frame number T2 required for track discard, and determining the target track of the true target object based on the target track confirmation threshold value comprises: if the continuous T1 frames of UWB radar information in the multiple frames of UWB radar information do not appear the situation of disconnection of the true target object, creating or maintaining the target track of the true target object based on the multiple frames of UWB radar information, and if the continuous T2 frames of UWB radar information in the multiple frames of UWB radar information appear the situation of disconnection of the true target object, releasing the multiple frames of UWB radar information, and not creating or deleting the target track of the true target object.

[0057] Specifically, for example, when the radar continuously detects the presence of a certain vehicle for multiple frames and does not lose its signal, it is considered that the track of the vehicle is reliable, so as to create or continue to maintain the track of the vehicle. When the radar does not detect the signal of a certain vehicle for multiple frames, it is considered that the vehicle may have left the monitoring area or the signal is blocked, at which time the radar resource related to the vehicle is released, and the track of the vehicle is no longer tracked.

[0058] In this way, by dynamically adjusting the creation or maintenance of the target track through the target track confirmation threshold value, the flexibility and adaptability of track management can be improved, and resource waste can be avoided.

[0059] In an optional embodiment, the method further comprises: dynamically adjusting an initial continuous detection frame number T3 required for track confirmation and an initial continuous non-detection frame number T4 required for track discard based on a maximum target speed of a scene, a minimum acceleration resolution, and a radar frame period, to obtain the target continuous detection frame number T1 and the target continuous non-detection frame number T2.

[0060] Specifically, the target track confirmation threshold value can be determined in an adaptive adjustment manner. The specific formula is wherein Vmax is the maximum target speed of the scene (such as 30 km / h), a is the minimum acceleration resolution, and T is the radar frame period. The adaptive setting can be For example, Confirming the track means confirming for the continuous T1 frames without disconnection, and discarding the track means releasing the associated resources after the continuous T2 frames without corresponding. Secondly, the application can also store the track. Specifically, the state can be maintained according to a unique track ID, including the position, the speed, and the matching history. Each track should maintain the content shown in Table 1.

[0061]

[0062] Table 1

[0063] It needs to be understood that in a high-speed or more maneuverable scene, the target speed is faster, and more continuous detection frame numbers can be required to confirm the trajectory, and increasing the continuous detection frame numbers can reduce the misjudgment probability. In a low-speed or stable scene, fewer continuous detection frame numbers can be required, and reducing the continuous detection frame numbers can improve the response speed. In this way, by dynamically adjusting the frame numbers of the trajectory confirmation and discard, the adaptability of the system to different scenes can be improved, and misjudgment and missed judgment can be reduced.

[0064] In an optional embodiment, the target track of the true target body is created or maintained based on the multi-frame UWB radar information, including: creating or maintaining an initial track of the true target body based on the multi-frame UWB radar information, the initial track including multiple trajectories, if it is determined that the distance difference and / or the speed difference between the multiple trajectories are less than or equal to a preset distance difference threshold value and / or a preset speed difference threshold value, merging the multiple trajectories to obtain the target track of the true target body, if it is determined that the distance difference and / or the speed difference between the multiple trajectories are greater than the preset distance difference threshold value and / or the preset speed difference threshold value, and the angle difference between the multiple trajectories is greater than a preset angle difference threshold value, splitting the initial track based on the multiple trajectories to obtain multiple target tracks of the true target body.

[0065] Specifically, merging the track means that when the distance difference (Ad) and the speed difference (Av) between the multiple trajectories are less than or equal to a preset threshold, it is considered that the two trajectories belong to different observations of the same target body, and then they are merged into one track to avoid misjudgment as two target bodies. Splitting the track means that if the distance difference and the speed difference between the multiple trajectories are greater than the preset threshold, and the angle difference (D) is also greater than a preset angle difference threshold value, it is considered that the trajectory is broken, and the initial track needs to be split to obtain multiple target tracks of the true target body. For example, in a complex scene, a target is blocked and then reappears, and its track can change. At this time, by splitting the judgment, the original track can be split into multiple new tracks to accurately track the motion state of the target.

[0066] In this way, by merging or splitting the track, the relationship between the multiple trajectories can be effectively processed, and the accuracy and reliability of the track can be improved.

[0067] For the convenience of understanding the trajectory management mechanism of the present application, please refer to Figure 2-1 , Figure 2-1 is a schematic diagram of a trajectory management mechanism disclosed in an embodiment of the present application, which is Figure 2-1It can be known that the track management mechanism is as follows: firstly, the target is continuously detected, then a first threshold (NP detection) is performed, if not passed, it is judged as a noise point; if passed, a second threshold (Doppler consistency detection) is performed, if not passed, it is judged as a drift interference, if passed, a track is created / associated. Then, the track is continuously confirmed for T1 frames, so that it becomes a stable target and track maintenance is performed, including disconnection T2 frame release, multi-target matching split judgment and data structure update and the like. In this way, the track management mechanism of the present application effectively reduces the misjudgment rate and improves the target detection accuracy through continuous target detection and double threshold detection algorithm; the track is created or associated after double threshold detection, so that the tracking stability is enhanced; the track is continuously confirmed for T1 frames and maintained, so that the system efficiency is improved; the multi-target matching split judgment adapts to complex scenes and avoids track confusion; the data structure is dynamically updated, so that the information timeliness and system adaptability are ensured.

[0068] In an optional embodiment, a plurality of initial association hypotheses of a true target body are determined based on a target track of the true target body by using an MHT multi-hypothesis tracking algorithm, a target association hypothesis is determined from the plurality of initial association hypotheses, and a target track corresponding to the target association hypothesis is generated, including: determining a plurality of initial association hypotheses of a target body based on a target track of the true target body by using an MHT multi-hypothesis tracking algorithm, retaining a plurality of association hypotheses with greater possibility from the plurality of initial association hypotheses by pruning, determining a target association hypothesis from the plurality of association hypotheses with greater possibility based on a minimum total cost, a maximum target association probability and / or a longest continuous matching track index, and determining a target track corresponding to the target association hypothesis, wherein the minimum total cost index includes a distance error and / or a speed error index.

[0069] Specifically, the minimum total cost is used to fuse the distance error and the speed error to evaluate the rationality of the hypothesis. The maximum target association probability is used to reflect the possibility of association between the hypothesis and the target. The longest continuous matching track is used to reflect the continuity and stability of the hypothesis.

[0070] Specifically, after the track initiation and track management, the data association based on multi-hypothesis tracking (MHT) can be performed in the target tracking process. More specifically, the multi-hypothesis structure refers to maintaining multiple possible association paths for each track per frame, generating new hypotheses and calculating likelihoods. The first layer refers to the direct wave corresponding to the target, modeling the direct wave (Line-of-Sight, LOS). It is assumed that the detection points belong to the direct path reflection of the real target. Each target generates a "candidate trajectory" branch. The second layer refers to considering the possibility of multipath echoes, modeling the multipath (Non-Line-of-Sight, NLOS), assuming that some reflections are caused by indirect paths such as the ground, walls, etc. The multipath model is established according to the historical trajectory environment context (such as angle, delay, etc. Bias template). There should be signal feature differences between multipath hypotheses and direct wave hypotheses (such as SNR drop, Doppler ambiguity, etc.). Pruning strategies are used for top-down decreasing likelihood ordering, retaining the top K good hypotheses, and discarding the rest. The specific formula is K=K0+Nclutter, which limits the maximum number of branches of the hypothesis tree expansion and avoids the exponential explosion of the calculation amount, especially in clutter dense areas. Wherein, K is the upper limit of the total number of hypotheses to be retained, K0 is the reserved branch of the basic target number (2-3 hypotheses per target), and Nclutter is the estimated value of the number of clutter points in the detection frame. It can be dynamically adjusted according to the quality of each frame observation point, for example: K=min(Kmax,K0+λ Nclutter), where λ∈[0.5,2.0] controls the influence degree of clutter on the number of hypotheses. The confirmation and rejection operation refers to confirming or removing the management of low-probability hypotheses after exceeding the maximum number of hypotheses. The pruning order is sorted according to the "minimum total cost ->maximum target association probability->longest continuous matching trajectory" combined index, and the top K hypotheses are selected.

[0071] The cost evaluation refers to introducing the following Doppler-based cost term in the hypothesis matching evaluation, such as formula two:

[0072] Formula two

[0073] Wherein, Vobs is the Doppler velocity observed in the current frame, Vpred is the predicted velocity according to the historical trajectory (such as Kalman prediction), σv is the standard deviation of the velocity measurement noise, and Wv is the degree of consistency weight factor. The chi-square likelihood of the distance deviation and the velocity difference is fused to form a continuous updating and path reservation mechanism, realizing complete data association. The essence is the standardized residual square, which is used to evaluate the deviation between the current target velocity and the historical trajectory velocity. The weight W𝑣 is used together with the position residual term weight W𝑝 to form a composite cost function as formula three:

[0074] Formula three

[0075] In this way, the MHT multi-hypothesis tracking algorithm can effectively manage multiple initial association hypotheses, reduce computational complexity, and improve tracking efficiency through pruning and sorting.

[0076] For the convenience of understanding the pruning strategy of the present application, please refer to Figure 2-2 , Figure 2-2 The method flowchart of a pruning strategy disclosed in an embodiment of the present application is as follows: Figure 2-2 As can be seen, the pruning strategy method flowchart is as follows: starting from the root history track, it is divided into LOS (direct wave) hypothesis and NLOS (multipath wave) hypothesis. For the LOS hypothesis, verification is performed according to the Doppler effect and distance matching principle, and if it is consistent, a new track candidate is generated. For the NLOS hypothesis, delay, low signal-to-noise ratio (SNR), and multipath mode factors are considered, and matched with a multipath template to identify and process multipath signals. This strategy can effectively manage track hypotheses, remove impossible track branches, and retain the most likely path. The advantage of this pruning strategy is that it can effectively reduce the amount of calculation, improve tracking efficiency, enhance the adaptability of the system to complex environments, and improve the accuracy and reliability of target tracking.

[0077] In an optional implementation, the IMM model includes a constant velocity sub-model, a constant acceleration sub-model, and a uniform rotation sub-model, the maneuvering characteristics of the target body represent the acceleration change and / or turning situation motion state of the target body, and the weights of the sub-models in the IMM model are dynamically adjusted based on the maneuvering characteristics of the true target body in the target track corresponding to the target association hypothesis, to obtain an adjusted IMM model, including: the weights of the constant velocity sub-model, the constant acceleration sub-model, and the uniform rotation sub-model in the IMM model are dynamically adjusted based on the acceleration change and / or turning situation motion state of the true target body in the target track corresponding to the target association hypothesis, to obtain an adjusted IMM model.

[0078] Specifically, in the target body tracking process, after track initiation and track management and data interconnection based on multi-hypothesis tracking (MHT), maneuvering target tracking based on IMM multi-model Kalman can be performed. More specifically, for each current track, a 3-model IMM: CV (constant velocity model) CA (constant acceleration model) CT (uniform rotation model) can be used, and each model is represented by a state vector [x, y, Vx, Vy] or [x, y, Vx, Vy, ax, ay]. The filtering process is as follows:

[0079] Step 1 State Mixing (Mixing): The purpose is to give each model an "initial state of the previous frame", considering the possibility of switching from other models. Use the model transition probability matrix Π to weight the average of each model's state and covariance: as formula four:

[0080] Formula Four

[0081] Where Pij: transition probability from model i to j.

[0082] Step 2 Parallel Kalman Filter Update: For each model j, perform a set of Kalman filtering steps (or extended / universal Kalman filtering), as formula five:

[0083] Formula Five

[0084] Output the state estimate and covariance under each model.

[0085] Step 3 Model Probability Update:

[0086] Update the posterior probability of each model using the Bayesian principle, as formula six:

[0087] Formula Six

[0088] Model Likelihood , which can be obtained by observation error residuals and covariance (Gaussian likelihood function).

[0089] Step 4 Fusion Output:

[0090] Calculate the final state estimate as the weighted average of all models, as formula seven:

[0091] Formula Seven

[0092] The fused state output provides the target final position, velocity estimate, and is sent to the trajectory management module.

[0093] It needs to be understood that the model switching strategy of the present application is to dynamically adjust the transition probability according to certain conditions, such as changes in heading, distance mutation rate, or mutation error jump; for transition not obvious track, reduce the model switching frequency, prevent false switching. Design the transition probability matrix Π:

[0094]

[0095] Where this matrix expresses the model persistence tendency (diagonal dominant), but allows model jump, the indicators include the indicators shown in Table 2 as follows:

[0096]

[0097] Table 2

[0098] In this way, the IMM model can adaptively match the target maneuvering characteristics by dynamically adjusting the weight of each sub-model, thereby improving the accuracy and flexibility of tracking.

[0099] The technical effects achieved by the present application are illustrated below by test results. For details, please refer to Figure 2-3 、 Figure 2-4 and Figure 2-5 , Figure 2-3 is a first and second threshold detection result schematic diagram disclosed by an embodiment of the present application, Figure 2-4 is an IMM multi-model Kalman detection result schematic diagram disclosed by an embodiment of the present application, Figure 2-5 is an overall position deviation result schematic diagram disclosed by an embodiment of the present application. Figure 2-3 The first and second threshold detection results are shown: the Neyman-Pearson detection result of the left figure shows that the detection result is 1 (target exists) within about 60 frames, and 0 (no target) after 60 frames; the Doppler continuity detection result of the right figure shows that the detection result is 0 (no target) within about 5 frames, and 1 (target exists) after 5 frames. This shows that the double threshold detection algorithm can effectively distinguish between true targets and noise, reduce the misjudgment rate, and improve the target detection accuracy. Figure 2-4 The IMM multi-model Kalman detection result is shown: the UWB radar target tracking-trajectory comparison of the left figure shows the comparison between the true trajectory and the IMM estimated trajectory, the CV model, the CA model, and the CT model. The IMM estimated trajectory (red solid line) is relatively close to the true trajectory (black dashed line), indicating that the IMM model effectively fuses the advantages of each sub-model and improves the tracking accuracy. The right figure shows the IMM model probability change over time. The CV, CA, and CT model probabilities change dynamically with the frame number. The CV probability (green) is relatively high at the beginning, the CA probability (blue) rises at about 30 frames, and the CT probability (pink) significantly increases at about 70 frames, which shows that the IMM model can dynamically adjust the weight according to the target motion state, adapt to the target maneuvering characteristics, and enhance the tracking effect. Figure 2-5The overall position deviation results are shown: the left overall X direction comparison and the middle overall Y direction comparison respectively show the comparison between the true X, Y coordinates and the IMM estimated X, Y coordinates. In the X direction, the true X (black dotted line) and the IMM estimated X (red solid line) have the same overall trend, but the true X starts to rise rapidly at about 60 frames, and the IMM estimated X is slightly delayed; in the Y direction, the true Y (black dotted line) and the IMM estimated Y (red solid line) have the same overall trend, but the true Y suddenly rises at about 40 frames, and the IMM estimated Y is slightly delayed. This shows that when the target maneuverability changes, the IMM estimation has a certain delay. The overall position error of the right figure shows the change of the position error with the frame number. The error is small at about 40 frames, the error increases after 40 frames, and the error obviously increases at about 80 frames, which shows that the error will increase when the target motion state changes.

[0100] According to the data, the average position error in the CT stage is 3.347 m, and the average deviation in the Y direction is 3.169 m, which shows that when the target performs uniform rotation, although there is error, the overall is in an acceptable range. The overall average position error is 1.212 m, which shows that the algorithm of the present application can track the target with small error on the whole. The maximum Y direction deviation is 5.693 m, and when the target motion state changes dramatically, a larger deviation may occur, but this situation is relatively rare. The error in the CV stage (1-30 frames) is 0.042 m, the error in the CA stage (31-70 frames) is 0.488 m, and the error in the CT stage (71-100 frames) is 3.347 m. As the complexity of the target motion state increases, the error will increase, but the amount of error increase is not large, and the algorithm of the present application can effectively track the target.

[0101] In summary, Figure 2-3 The test results of the double threshold detection algorithm show that the double threshold detection algorithm can effectively filter out true targets and reduce false positives. Figure 2-4 The test results of the IMM multi-model Kalman filter show that the IMM multi-model Kalman filter can dynamically adjust the model weight according to the target motion state, and improve the tracking accuracy. Figure 2-5 The test results of the overall position error show that the overall position error is within a reasonable range, but the error will increase when the target maneuverability changes, and the algorithm of the present application can still achieve accurate tracking.

[0102] Further, please refer to Figure 3 An embodiment of the target body tracking device in the present application embodiment comprises:

[0103] An acquisition unit is configured to acquire multiple frames of UWB radar information.

[0104] The determination unit is configured to detect the multiple frames of UWB radar information by using a double-threshold detection algorithm, determine whether a true target body exists, and if it is determined that the true target body exists, determine a target track of the true target body based on a target track confirmation threshold.

[0105] The determination unit is further configured to determine multiple initial association hypotheses of the true target body based on the target track of the true target body by using an MHT multi-hypothesis tracking algorithm, determine a target association hypothesis from the multiple initial association hypotheses, and generate a target track corresponding to the target association hypothesis.

[0106] The adjustment unit is configured to dynamically adjust weights of each sub-model in an IMM model based on a maneuvering characteristic of the true target body in the target track corresponding to the target association hypothesis by using the IMM model, and obtain an adjusted IMM model.

[0107] The tracking unit is configured to predict a motion state of the true target body based on the target track corresponding to the target association hypothesis by using the adjusted IMM model, and track the true target body.

[0108] In an optional implementation, the determination unit is specifically configured to:

[0109] determine whether a target body exists by detecting whether a degree of change in speed and / or distance in the multiple frames of UWB radar information reaches a preset degree of change in speed and / or distance by using the first threshold detection algorithm, and determine whether the target body is a true target body by determining whether a degree of fluctuation in speed in the multiple frames of UWB radar information is within a preset range of the degree of fluctuation in speed by using the second threshold detection algorithm, to determine whether a true target body exists.

[0110] In an optional implementation, the determination unit is specifically configured to:

[0111] If a true target body is not lost in continuous T1 frames of UWB radar information in the multiple frames of UWB radar information, a target track of the true target body is created or maintained based on the multiple frames of UWB radar information, if a true target body is lost in continuous T2 frames of UWB radar information in the multiple frames of UWB radar information, the multiple frames of UWB radar information are released, and the target track of the true target body is not created or deleted, and the target track confirmation threshold includes a target continuous detection frame number T1 required for track confirmation and a target continuous non-detection frame number T2 required for track abandonment.

[0112] In an optional implementation, the determination unit is specifically configured to:

[0113] The initial continuous detection frame number T1 and the initial continuous non-detection frame number T2 are obtained by dynamically adjusting an initial continuous detection frame number T3 required for trajectory confirmation and an initial continuous non-detection frame number T4 required for trajectory abandonment based on a maximum target speed of a scene, a minimum acceleration resolution, and a radar frame period.

[0114] In an optional implementation, the determining unit is specifically configured to:

[0115] The initial track of the true target body is created or maintained based on the multiple frames of UWB radar information, the initial track includes multiple tracks, if it is determined that a distance difference and / or a speed difference between the multiple tracks is less than or equal to a preset distance difference threshold value and / or a preset speed difference threshold value, the multiple tracks are merged to obtain a target track of the true target body, if it is determined that the distance difference and / or the speed difference between the multiple tracks is greater than the preset distance difference threshold value and / or the preset speed difference threshold value, and an angle difference between the multiple tracks is greater than a preset angle difference threshold value, the initial track is split based on the multiple tracks to obtain multiple target tracks of the true target body.

[0116] In an optional implementation, the determining unit is specifically configured to:

[0117] The multiple initial association hypotheses of the target body are determined based on the target track of the true target body by using an MHT multi-hypothesis tracking algorithm, a plurality of more likely association hypotheses in the multiple initial association hypotheses are reserved by pruning, a target association hypothesis is determined in the plurality of more likely association hypotheses based on a minimum total cost, a maximum target association probability, and / or a longest continuous matching track index, and a target track corresponding to the target association hypothesis is determined, the minimum total cost index includes a distance error and / or a speed error index.

[0118] In an optional implementation, the adjusting unit is specifically configured to:

[0119] The weights corresponding to a constant speed sub-model, a constant acceleration sub-model, and a uniform turning sub-model in the IMM model are dynamically adjusted based on an acceleration change and / or a turning condition motion state of the true target body in the target track corresponding to the target association hypothesis, to obtain the adjusted IMM model, the IMM model includes the constant speed sub-model, the constant acceleration sub-model, and the uniform turning sub-model, and the maneuvering characteristic of the target body represents the acceleration change and / or the turning condition motion state of the target body.

[0120] Further, please refer to Figure 4 An embodiment of the electronic device in the embodiment of the present application includes:

[0121] The central processor 401, the memory 405, the input and output interface 404, the wired or wireless network interface 403 and the power supply 402;

[0122] The memory 405 is a volatile storage memory or a persistent storage memory.

[0123] The central processor 401 is configured to communicate with the memory 405 and execute the instruction operation in the memory 405 to perform the method in the foregoing Figure 2 embodiments.

[0124] Further, the embodiments of the present application further provide a computer readable storage medium including instructions, when the instructions run on a computer, causing the computer to execute the method in the foregoing Figure 2 embodiments.

[0125] Further, the embodiments of the present application further provide a computer program product including instructions, when the computer program product runs on a computer, causing the computer to execute the method in the foregoing Figure 2 embodiments.

[0126] It should be understood that, although each step in the flowchart involved in each of the above-described embodiments is shown in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each of the above-described embodiments can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0127] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0128] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0129] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments of the present application.

[0130] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of a software functional unit.

[0131] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such an understanding, the technical solutions of the present application essentially, or the part that contributes to the prior art, or all or a part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), magnetic disk or optical disk, and various other media that can store program codes.

Claims

1. A target object tracking method characterized by, The method comprises: acquiring multiple frames of UWB radar information; detecting the multiple frames of UWB radar information by using a double-threshold detection algorithm to determine whether a true target body exists, and if it is determined that a true target body exists, determining a target track of the true target body based on a target track confirmation threshold; determining multiple initial association hypotheses of the true target body based on the target track of the true target body by using an MHT multi-hypothesis tracking algorithm, determining a target association hypothesis in the multiple initial association hypotheses, and generating a target track corresponding to the target association hypothesis; dynamically adjusting weights of each sub-model in an IMM model based on a maneuvering characteristic of the true target body in the target track corresponding to the target association hypothesis by using the IMM model to obtain an adjusted IMM model; predicting a motion state of the true target body based on the target track corresponding to the target association hypothesis by using the adjusted IMM model to track the true target body.

2. The method of claim 1, wherein, The double-threshold detection algorithm comprises a first threshold detection algorithm and a second threshold detection algorithm, and the step of detecting the multiple frames of UWB radar information by using a double-threshold detection algorithm to determine whether a true target body exists comprises: detecting whether a speed and / or a distance change degree in the multiple frames of UWB radar information reaches a preset speed and / or distance change degree threshold by using the first threshold detection algorithm to determine whether a target body exists; determining whether a speed fluctuation degree in the multiple frames of UWB radar information is within a preset speed fluctuation degree range by using the second threshold detection algorithm to determine whether the target body is a true target body and to determine whether a true target body exists.

3. The method of claim 1, wherein, The target track confirmation threshold comprises a target continuous detection frame number T1 required for track confirmation and a target continuous non-detection frame number T2 required for track discard; The step of determining a target track of the true target body based on a target track confirmation threshold comprises: if a true target body disconnection situation does not occur in continuous T1 frames of UWB radar information in the multiple frames of UWB radar information, creating or maintaining a target track of the true target body based on the multiple frames of UWB radar information; if a true target body disconnection situation occurs in continuous T2 frames of UWB radar information in the multiple frames of UWB radar information, releasing the multiple frames of UWB radar information, and not creating or deleting a target track of the true target body.

4. The method of claim 3, wherein, The method further comprises: dynamically adjusting an initial continuous detection frame number T3 required for track confirmation and an initial continuous non-detection frame number T4 required for track discard based on a maximum target speed of a scene, a minimum acceleration resolution, and a radar frame period to obtain the target continuous detection frame number T1 and the target continuous non-detection frame number T2.

5. The method of claim 3, wherein, The step of creating or maintaining a target track of the true target body based on the multiple frames of UWB radar information comprises: creating or maintaining an initial track of the true target body based on the multiple frames of UWB radar information, the initial track comprising multiple tracks; if it is determined that a distance difference and / or a speed difference between the multiple tracks is less than or equal to a preset distance difference threshold and / or a preset speed difference threshold, merging the multiple tracks to obtain a target track of the true target body; If it is determined that the distance difference and / or the speed difference between the multiple trajectories is greater than the preset distance difference threshold value and / or the preset speed difference threshold value, and the angle difference between the multiple trajectories is greater than the preset angle difference threshold value, the initial track is split based on the multiple trajectories to obtain target tracks of the multiple true target bodies.

6. The method of claim 1, wherein, The MHT multi-hypothesis tracking algorithm determines multiple initial association hypotheses of the target body based on the target track of the true target body, determines a target association hypothesis in the multiple initial association hypotheses, and generates a target track corresponding to the target association hypothesis, including: The MHT multi-hypothesis tracking algorithm determines multiple initial association hypotheses of the target body based on the target track of the true target body; The multiple initial association hypotheses are pruned to retain multiple association hypotheses with greater likelihood; Based on a minimum total cost, a maximum target association probability, and / or a longest continuous matching track indicator, a target association hypothesis is determined in the multiple association hypotheses with greater likelihood, and a target track corresponding to the target association hypothesis is determined, wherein the minimum total cost indicator includes a distance error and / or a speed error indicator.

7. The method of claim 1, wherein, The IMM model includes a constant speed sub-model, a constant acceleration sub-model, and a uniform rotation sub-model, and the maneuvering characteristics of the true target body represent the acceleration change and / or turning situation motion state of the target body; The IMM model is dynamically adjusted based on the maneuvering characteristics of the true target body in the target track corresponding to the target association hypothesis, to obtain an adjusted IMM model, including: The IMM model is dynamically adjusted based on the acceleration change and / or turning situation motion state of the true target body in the target track corresponding to the target association hypothesis, to obtain the adjusted IMM model.

8. A target body tracking apparatus characterized by comprising: It includes: An acquisition unit is configured to acquire multiple frames of UWB radar information; A determination unit is configured to detect whether there is a true target body by using a double-threshold detection algorithm to detect the multiple frames of UWB radar information, and determine a target track of the true target body based on a target track confirmation threshold value if it is determined that there is a true target body; The determination unit is further configured to determine multiple initial association hypotheses of the true target body based on the target track of the true target body by using a MHT multi-hypothesis tracking algorithm, determine a target association hypothesis in the multiple initial association hypotheses, and generate a target track corresponding to the target association hypothesis; An adjustment unit is configured to dynamically adjust weights of each sub-model in an IMM model based on maneuvering characteristics of the true target body in a target track corresponding to a target association hypothesis by using the IMM model, to obtain an adjusted IMM model; A tracking unit is configured to predict a motion state of the true target body based on the target track corresponding to the target association hypothesis by using the adjusted IMM model, to track the true target body.

9. An electronic device, comprising: It includes: A central processing unit and a memory; The memory is a volatile memory or a persistent memory; The central processing unit is configured to communicate with the memory and execute instruction operations in the memory to perform the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises instructions which, when run on a computer, cause the computer to perform the method of any one of claims 1-7.

Images