A collision avoidance system multi-target tracking method and apparatus
Patent Information
- Application Number
- CN202610744476.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-27
- Publication Date
- 2026-10-27
- Estimated Expiration
- 2046-05-27
Smart Images
Figure CN122568515B_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of underwater sonar detection and multi-target tracking technology, and in particular to a multi-target tracking method and apparatus for a collision avoidance system. Background Technology
[0002] With the increasing use of underwater vehicles, safe navigation has become a crucial issue. Underwater vehicles, especially unmanned submersibles and automated underwater vehicles (AUVs), often face unknown and complex underwater environments and the risk of collisions with other underwater obstacles when performing missions such as ocean exploration, scientific research, and environmental monitoring. To ensure the safety of underwater vehicles during missions, the application of obstacle avoidance systems (OAS) is particularly important. The target tracking capability of collision avoidance sonar plays a vital role in enhancing the environmental situational awareness and achieving accurate early warning within collision avoidance systems.
[0003] However, collision avoidance sonar multi-target tracking currently faces the following challenges: First, the number of targets in the underwater environment is usually unknown and changes over time (targets appear or disappear), and the target state also changes dynamically over time, making traditional tracking methods based on a fixed number of targets difficult to apply. Second, while receiving target echoes, sonar inevitably receives interference measurements from reverberation, environmental noise, and carrier self-noise. These false alarm measurements and real target measurements may exhibit similar spatial distribution characteristics in the range-azimuth plane, leading to difficulties in data correlation. Third, underwater targets often have diverse shapes, incomplete acoustic observations, and complex trajectories, further exacerbating the difficulty of track maintenance under low signal-to-mixing conditions. These challenges collectively constitute the core difficulties of collision avoidance sonar multi-target tracking.
[0004] Therefore, in order to adapt to the complex underwater environment and complete the obstacle target tracking task in the collision avoidance system of unmanned vehicles, there is an urgent need for a high-performance and accurate collision avoidance sonar multi-target tracking algorithm to enhance the robustness of the collision avoidance algorithm in complex environments and ensure the early warning capability and recognition reliability of the collision avoidance system.
[0005] Application content
[0006] This application describes a multi-target tracking method and apparatus for a collision avoidance system, which can solve the above-mentioned technical problems.
[0007] According to a first aspect, a multi-target tracking method for a collision avoidance system is provided, comprising the following steps: receiving beam domain data from a collision avoidance sonar, wherein the beam domain data includes echo intensity information in multiple beam directions; obtaining a joint measurement set at the current moment based on the beam domain data; and performing [further steps] based on the joint measurement set. - A generalized labeled Bernoulli filter is used to predict the obstacle target's predicted state set at the current moment. Based on the obstacle target's predicted state set, the target signal-to-mixing ratio, and the joint measurement set, a joint measurement likelihood model is constructed. The joint measurement likelihood model includes a target intensity likelihood term and a false alarm intensity likelihood term. The target intensity likelihood term is a closed-form expression derived based on the negative exponential distribution characteristics of the collision avoidance sonar echo intensity, using no-information priors and marginalizing the target signal-to-mixing ratio. - The generalized labeled Bernoulli filter is updated to obtain the multi-target posterior density of the obstacle target at the current time; - The generalized labeled Bernoulli filter includes embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio to automatically penalize false alarm measurements and perform multi-hypothesis weight updates to enhance obstacle targets.
[0008] In one possible implementation, the joint measurement set includes at least one candidate obstacle target measurement point, which includes distance information, beam direction information, and echo intensity information of the candidate obstacle target.
[0009] In one possible implementation, embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio includes: calculating the joint likelihood ratio of a subset of measurement hypotheses relative to the predicted state of the obstacle target based on the target likelihood value and the false alarm likelihood value; determining the associated hypothesis weights based on the joint likelihood ratio; and updating the multi-target posterior density of the δ-generalized label multi-Bernoulli filter based on the associated hypothesis weights.
[0010] In one possible implementation, after obtaining the multi-target posterior density of the obstacle target at the current time, the method further includes: predicting the obstacle target based on the posterior density of the δ-generalized label Bernoulli filter at the previous time to obtain the obstacle target prediction state set at the current time; constructing multiple measurement hypothesis subsets based on the joint measurement set, wherein the measurement hypothesis subsets are subsets of the joint measurement set and are used to represent the set of candidate obstacle target measurement information that may be generated by the same obstacle target under the corresponding association hypothesis.
[0011] In one possible implementation, the process of... - The generalized label Bernoulli filter predicts the obstacle target prediction state set at the current time, including: based on the posterior density of the δ-generalized label Bernoulli filter at the previous time, the obstacle target is predicted using the δ-generalized label Bernoulli filter to obtain the obstacle target prediction state set at the current time.
[0012] In one possible implementation, a joint measurement likelihood model is constructed based on the set of predicted obstacle target states and the joint measurement set, including: constructing multiple subsets of measurement hypotheses based on the joint measurement set; determining a target likelihood term for each subset of measurement hypotheses relative to the predicted obstacle target state for each predicted obstacle target state and each subset of measurement hypotheses; the target likelihood term includes the target intensity likelihood function term; determining a false alarm likelihood term corresponding to each subset of measurement hypotheses; the false alarm likelihood term includes the false alarm intensity likelihood function term; and constructing the joint measurement likelihood model based on the target likelihood term and the false alarm likelihood term.
[0013] In one possible implementation, the method further includes: outputting the tracking trajectory result of the obstacle target at the current moment based on the posterior density.
[0014] According to a second aspect, a multi-target tracking device for a collision avoidance system is provided, comprising: a receiving module for receiving beam domain data of a collision avoidance sonar, the beam domain data including echo intensity information in multiple beam directions; a prediction module for obtaining a joint measurement set at the current moment based on the beam domain data; and performing prediction based on the joint measurement set. - A generalized labeled Bernoulli filter predicts the current set of predicted obstacle target states; a construction module is used to construct a joint measurement likelihood model based on the obstacle target predicted state set, the target signal-to-mixing ratio, and the joint measurement set. The joint measurement likelihood model includes a target intensity likelihood term and a false alarm intensity likelihood term. The target intensity likelihood term is a closed-form expression derived from the negative exponential distribution characteristics of the collision avoidance sonar echo intensity, using no prior information and marginalizing the target signal-to-mixing ratio; an update module is used to perform... - The generalized labeled Bernoulli filter is updated to obtain the multi-target posterior density of the obstacle target at the current time; - The generalized labeled Bernoulli filter includes embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio to automatically penalize false alarm measurements and perform multi-hypothesis weight updates to enhance obstacle targets.
[0015] According to a fourth aspect, an electronic device is provided, including a memory and one or more processors, wherein a computer program is stored on the memory, and the computer program, when executed by the one or more processors, implements the multi-target tracking method of the collision avoidance system as described in any of the above embodiments. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 A schematic diagram of the algorithm flow of a multi-target tracking method for a collision avoidance system provided in an embodiment of this application;
[0018] Figure 2 A flowchart illustrating a multi-target tracking method for a collision avoidance system provided in an embodiment of this application;
[0019] Figure 3 A schematic acoustic image of a target scanned by sonar, provided in an embodiment of this application.
[0020] Figure 4 The strength measurement-based method provided in the embodiments of this application - Tracking trajectory results processed by the GLMB algorithm and - A schematic diagram of the tracking trajectory results after processing by the GLMB algorithm;
[0021] Figure 5 This application provides a schematic diagram of a multi-target tracking module for a collision avoidance system. Detailed Implementation
[0022] The solution provided in this specification will now be described with reference to the accompanying drawings.
[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be described below with reference to the accompanying drawings.
[0024] In the description of the embodiments of this application, the words "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a specific manner.
[0025] In the description of the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, B existing alone, and A and B existing simultaneously. Furthermore, unless otherwise stated, the term "multiple" means two or more.
[0026] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.
[0027] In the practical application of unmanned underwater vehicles (UUVs), due to the complex and variable underwater environment, traditional collision avoidance tracking algorithms use the target's distance and orientation as measurements for data correlation. However, in reverberant environments, interference from reverberation, ambient noise, and carrier self-noise can easily generate false alarms. Measurements generated by false alarms can lead to incorrect association of the target tracking trajectory with the false alarm measurements, resulting in false tracks and affecting the obstacle avoidance system's judgment. The generation of false tracks affects the performance of the collision avoidance system, causing the UUV to frequently perform unnecessary avoidance maneuvers, wasting time and energy, reducing operational efficiency, and ultimately affecting the overall performance of the equipment. With the increasing use of UUVs, the application of collision avoidance systems is crucial to ensuring the navigation safety of these vehicles.
[0028] Target tracking and target state estimation provide crucial information for early warning and maneuvering in collision avoidance systems. To suppress false tracks, this invention employs a collision avoidance sonar multi-target tracking method based on intensity measurement during target tracking. This method balances multi-target tracking performance with strong anti-interference capabilities, significantly reducing the occurrence of false tracks in real-world multi-target tracking scenarios. To address the issue of false tracks arising during multi-target tracking in reverberant environments, target echo intensity is incorporated into the measurement model, constructing a multi-target tracking framework that combines range, azimuth, and intensity measurements. A likelihood function incorporating intensity measurement is then applied. - Updating the target using a generalized labeled Bernoulli filter improves the reliability of data association under low signal-to-mixing conditions and solves the problem of false alarm measurements in target tracking trajectory association under reverberation conditions. Introducing intensity information into target measurements for data association and updating of target state significantly improves the algorithm's anti-interference capability, suppresses false tracks, and enhances the algorithm's tracking performance.
[0029] Figure 1This is a schematic diagram of the algorithm flow for a multi-target tracking method in a collision avoidance system provided in an embodiment of this application. Figure 1 As shown,
[0030] First, the beam domain data of the collision avoidance sonar is received. The beam domain data includes echo intensity values (beam intensity) corresponding to multiple beam directions and multiple range cells.
[0031] Then, in the constant false alarm rate (CFAR) detection zone, for any unit to be detected in the beam domain data, reference units around the unit to be detected are selected, and the background reverberation intensity is estimated based on the echo intensity of the reference units. The detection threshold corresponding to the unit to be detected is determined according to the background reverberation intensity and the preset CFAR coefficient. The echo intensity of the unit to be detected is compared with the detection threshold, and the detection decision result is output: if the echo intensity of the unit to be detected is greater than the detection threshold, the unit to be detected is determined as a detection point. The set of detection points exceeding the detection threshold is obtained by traversing multiple beam directions and multiple distance units in the beam domain data.
[0032] Next, feature extraction is performed on the set of detection points exceeding the detection threshold. For example, the detection points are clustered to obtain at least one candidate obstacle target point cluster, and candidate obstacle target measurement information is extracted based on the candidate obstacle target point cluster to form the joint measurement set at the current time. The detection points are clustered to obtain at least one candidate obstacle target point cluster, and candidate obstacle target measurement information is extracted based on the candidate obstacle target point cluster to form the joint measurement set at the current time. .
[0033] Then, a joint measurement likelihood model incorporating the intensity likelihood function is constructed. Based on the negative exponential distribution of collision avoidance sonar echo intensity, a closed-form expression for the intensity likelihood function is derived. When the signal-to-mixing ratio (SMR) is unknown, an uninformative Jeffreys prior is used to marginalize the SMR parameters, avoiding additional errors introduced by parameter estimation. The target intensity likelihood function is: ,
[0034] False alarm likelihood function .
[0035] Substituting the target intensity likelihood function and the false alarm likelihood function... - A generalized labeled Bernoulli filter is used for prediction and updating to obtain the multi-objective posterior density.
[0036] Finally, the motion state, number of targets, and continuous track labels of each obstacle are extracted from the posterior density, and stable and reliable obstacle tracking results are output for collision avoidance decision-making.
[0037] The construction of the joint measurement likelihood model includes:
[0038] Let the obstacle target expansion state be: Target measurement modeling is ,in ,in and Let be the position of the target in Cartesian coordinates. and Corresponding to the target's speed, For the target turning speed, Indicates the target signal-to-mix ratio. , It is the azimuth angle. For distance, This indicates the target intensity value.
[0039] Assuming that kinematic measurements and strength measurements are independent, then the target likelihood is... Similar to a false alarm for
[0040] (1)
[0041] (2)
[0042] in, Let be the objective likelihood function. Let be the target intensity likelihood function. Let be the likelihood function for false alarm intensity. This represents the spatial distribution density of false alarms.
[0043] The construction of the intensity likelihood function based on the negative exponential distribution includes:
[0044] Against a Gaussian reverberation background, after matched filtering and square detection, the collision avoidance sonar beam intensity approximately follows a negative exponential distribution. The detection threshold is... Normalized background noise power .exist Under the given conditions, the conditional probability density of false alarm intensity is:
[0045] (3)
[0046] When the information is mixed, At that time, the target intensity was Under the condition that the conditional probability density is
[0047] (4)
[0048] Therefore, the signal-to-mix ratio is known. hour:
[0049]
[0050] (6)
[0051] Marginalization processing for unknown information scrambling ratios includes:
[0052] In actual collision avoidance scenarios for aircraft, the signal-to-object confusion ratio is... Typically unknown. To avoid the complexity and uncertainty of online estimation, this invention introduces Jeffreys' uninformed prior, taking...
[0053] ,
[0054] in For the range of values of the smear ratio, After marginalization, the target intensity likelihood without parameters is obtained:
[0055] .
[0056] This closed-form can be directly used in filters without estimating the target signal-to-mixing ratio. At high signal-to-mixing ratios, the obstacle echo intensity is significantly higher than the background. After intensity likelihood weighting, the filter's trajectory starts more rapidly, allowing the collision avoidance system to identify new obstacles earlier and providing the vehicle with more avoidance time. This application's embodiment uses Jeffreys prior marginalization to handle unknown signal-to-mixing ratios, avoiding online parameter estimation. The algorithm complexity is controllable, and the method is based on mature Gaussian mixture theory. - The GLMB framework is easy to deploy on existing collision avoidance sonar signal processing platforms.
[0057] The enhanced δ-GLMB filter update includes:
[0058] Substituting the above joint likelihood into The -GLMB update equation has the following posterior density:
[0059]
[0060] in,
[0061] ,
[0062] ,
[0063] ,
[0064] ,
[0065] The probability of target detection. Let be the likelihood function of the target space. Let be the target intensity likelihood function. The spatial intensity density function for false alarms. Let be the likelihood function for false alarm intensity.
[0066] because It reflects the contrast between the measurement intensity and the false alarm background. When the intensity of the detection point is consistent with the distribution of false alarms, the ratio is small, which will automatically reduce the association weight, thereby effectively punishing the association of false alarms and enhancing the association confidence of real obstacle targets.
[0067] Figure 2 This is a flowchart illustrating a multi-target tracking method for a collision avoidance system provided in an embodiment of this application. Figure 2 As shown, the method includes:
[0068] Step S210: Receive beam domain data from the collision avoidance sonar, wherein the beam domain data includes echo intensity distribution information in multiple beam directions.
[0069] By introducing echo intensity information, real obstacles and reverberation false alarms differ in intensity dimension.
[0070] In this application embodiment, the collision avoidance sonar echo intensity is used as an additional measurement to construct a random finite set multi-target tracking framework with joint range-azimuth-intensity measurement.
[0071] Step S220: Based on the beam domain data, obtain the joint measurement set at the current moment.
[0072] In one possible implementation, the joint measurement set includes at least one candidate obstacle target measurement point, which includes distance information, beam direction information, and echo intensity information of the candidate obstacle target.
[0073] Step S230, proceed - Generalized labeled Bernoulli filter prediction yields the set of predicted obstacle target states at the current moment.
[0074] In one possible implementation, the obstacle target is predicted based on the posterior density of the δ-generalized label Bernoulli filter at the previous time step, and the predicted state set of the obstacle target at the current time step is obtained.
[0075] In one possible implementation, multiple measurement hypothesis subsets are constructed based on the joint measurement set. These measurement hypothesis subsets are subsets of the joint measurement set and are used to represent a set of candidate obstacle target measurement information that may be generated by the same obstacle target under the corresponding association hypothesis.
[0076] Step S240: Based on the obstacle target prediction state set, the target signal-to-mixing ratio, and the joint measurement set, construct a joint measurement likelihood model. The joint measurement likelihood model includes a target intensity likelihood term and a false alarm intensity likelihood term. The target intensity likelihood term is a closed expression derived based on the negative exponential distribution characteristics of the collision avoidance sonar echo intensity. It is obtained by marginalizing the target signal-to-mixing ratio using no informational priors. The target signal-to-mixing ratio is used to characterize the significance of the candidate obstacle target echo relative to the background reverberation or clutter.
[0077] In one possible implementation, the target intensity likelihood term is given in formula (8) above. The false alarm intensity likelihood term is given in formula (2) above.
[0078] In one possible implementation, embodiments of this application derive a closed-form expression for the intensity likelihood function based on the assumption of a negative exponential distribution under reverberation background, and utilize Jeffreys prior marginalization to solve the problem of unknown signal-to-mixing ratio.
[0079] In one possible implementation, a joint measurement likelihood model is constructed based on the set of predicted obstacle target states and the joint measurement set, including: constructing multiple subsets of measurement hypotheses based on the joint measurement set; determining a target likelihood term for each subset of measurement hypotheses relative to the predicted obstacle target state for each predicted obstacle target state and each subset of measurement hypotheses; the target likelihood term includes the target intensity likelihood function term; determining a false alarm likelihood term corresponding to each subset of measurement hypotheses; the false alarm likelihood term includes the false alarm intensity likelihood function term; and constructing the joint measurement likelihood model based on the target likelihood term and the false alarm likelihood term.
[0080] In one possible implementation, the target signal-to-mixing ratio (SMR) is used to characterize the prominence of a candidate obstacle target echo relative to background reverberation or clutter. Specifically, the SMR can be determined based on echo intensity information from the candidate obstacle target's measurement data, as well as the background reverberation or clutter intensity in the area surrounding the candidate obstacle target. A high SMR indicates that the candidate obstacle target echo is more prominent relative to the background; a low SMR indicates that the candidate obstacle target is more likely to be affected by background reverberation, clutter, or false alarms.
[0081] The embodiments of this application utilize a modified joint likelihood ratio to automatically penalize false alarm measurements and a multi-hypothesis weight update mechanism to enhance real obstacle targets.
[0082] Step S250: Perform strength measurement-based... - The generalized labeled Bernoulli filter is updated to obtain the multi-target posterior density of the obstacle target at the current time; the intensity measurement-based... - The generalized labeled Bernoulli filter includes embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio to automatically penalize false alarm measurements and perform multi-hypothesis weight updates to enhance obstacle targets.
[0083] In one possible implementation, the joint likelihood ratio is the ratio between the target intensity likelihood value of the measurement hypothesis subset relative to the predicted state of the obstacle target and the false alarm intensity likelihood value corresponding to the measurement hypothesis subset, as detailed in formula (13) above. The intensity likelihood function automatically penalizes low-power false alarms, significantly reducing the generation of false tracks and preventing the collision avoidance system from triggering unnecessary maneuvers due to false target tracks.
[0084] In this application embodiment, the intensity likelihood ratio is embedded into the update step of the δ-generalized label Bernoulli filter to achieve joint likelihood-driven adaptive false alarm penalty and real target enhancement.
[0085] In one possible implementation, the tracking trajectory of the obstacle target at the current moment is output based on the posterior density.
[0086] This application's embodiments introduce echo intensity information, creating a difference in intensity between real obstacles and reverberant false alarms. The intensity likelihood function automatically penalizes low-power false alarms, significantly reducing the generation of false tracks and preventing the collision avoidance system from triggering unnecessary maneuvers due to false target tracks. The data association between joint range and azimuth intensity is more reliable. In cases of strong reverberation or single-frame missed detection, tracks are less prone to drifting or interruption due to false alarm association, resulting in smaller target number estimation bias and providing continuous and reliable obstacle motion information for the collision avoidance system. At high signal-to-mixing ratios, obstacle echo intensity is significantly higher than background, leading to a faster track initiation after intensity likelihood weighting, enabling the collision avoidance system to identify new obstacle targets earlier and providing the vehicle with more avoidance time. Furthermore, Jeffreys prior marginalization is used to handle unknown signal-to-mixing ratios, avoiding online parameter estimation. The algorithm complexity is controllable, and the method is based on the mature Gaussian Mixture Model (GLMB) framework, making it easy to deploy on existing collision avoidance sonar signal processing platforms.
[0087] The following describes the implementation details and pseudocode of the algorithm in the embodiments of this application.
[0088] Prediction steps: Input the hypothetical components from the previous time step and the Bernoulli parameters of the newborn target, obtain the optimal subset of hypotheses for the surviving and newborn targets through the k-shortest path algorithm, merge them to form the prediction hypothesis components, and normalize the weights.
[0089] For example, the posterior hypothesis components of the δ-generalized label multi-Bernoulli filter from the previous time step and the Bernoulli parameters of the newborn target are input. A subset of survival target hypotheses is generated based on the target survival model, and a subset of newborn target hypotheses is generated based on the Bernoulli parameters of the newborn target. A subset of high-weight hypotheses is selected from the subset of survival target hypotheses and the subset of newborn target hypotheses using the K-shortest path algorithm. The selected subset of survival target hypotheses and the subset of newborn target hypotheses are merged to form the prediction hypothesis components at the current time step, and the weights of the prediction hypothesis components are normalized.
[0090] Update steps: For each prediction hypothesis, generate association mappings using ordinal assignment, combined with improved... Calculate the assumed weights, update the single-objective state density in Gaussian mixture form using unscented Kalman filtering or extended Kalman filtering, and retain the one with the largest weight. An assumption component is used to control the computational load.
[0091] For example, in the prediction phase, multiple prediction hypothesis components have been obtained, each representing a possible combination of obstacle targets. The current joint measurement set contains measurement information for multiple candidate obstacle targets. For each prediction hypothesis component, a subset of measurement hypotheses is constructed based on the current joint measurement set. Multiple association mappings are generated using a sorting and allocation algorithm based on the matching relationship between the subset of measurement hypotheses and the predicted states of the obstacle targets. For each association mapping, a joint likelihood ratio is calculated based on the corresponding target intensity likelihood value and false alarm intensity likelihood value, and the weight of the posterior hypothesis component is determined based on the joint likelihood ratio. Simultaneously, for each obstacle target under the association mapping, the Gaussian mixture-form single-target predicted state density is updated using unscented Kalman filtering or extended Kalman filtering to obtain the corresponding single-target posterior state density. The updated posterior hypothesis components are pruned, retaining those whose weights meet preset conditions to control the computational cost of the δ-GLMB filter.
[0092] State extraction: Extract the number of targets and the state of each track from the posterior hypothesis with the largest weight, and output the result.
[0093] The tagged continuous tracking trajectory is directly used for obstacle situation assessment of the collision avoidance system.
[0094] The algorithm pseudocode is shown in Tables 1 and 2:
[0095]
[0096] The following is based on Figure 3 and Figure 4This paper describes the tracking track results of embodiments of this application. Lake test data verifies that the proposed algorithm has a significant ability to suppress false tracks. During the observation period, a target was located to the right front of the sonar, gradually approaching the sonar. The 25th frame acoustic image was generated after forward-looking sonar scanning, as shown below. Figure 3 As shown in the figure, the location marked by the red box is the target's location. The tracking trajectory result obtained after processing this set of data using this algorithm is as follows: Figure 4 (a) The tracking track results obtained after processing by the GLMB algorithm are as follows: Figure 4 (b), where gray dots represent measured values and colored dots represent target tracks. (Through) Figure 4 A comparison of (a) and (b) shows that... The GLMB algorithm produced a large number of false tracks. However, the intensity measurement-based approach provided in this application's embodiments... - The GLMB algorithm did not produce any false tracks, demonstrating a good ability to distinguish between false alarms and targets.
[0097] Figure 5 This is a schematic diagram of a multi-target tracking module in a collision avoidance system provided in an embodiment of this application. Figure 5 As shown, a multi-target tracking device for a collision avoidance system includes: a receiving module for receiving beam domain data from a collision avoidance sonar, the beam domain data including echo intensity information in multiple beam directions; a prediction module for obtaining a joint measurement set at the current time based on the beam domain data; and performing prediction based on the joint measurement set. - A generalized labeled Bernoulli filter predicts the current set of predicted obstacle target states; a construction module is used to construct a joint measurement likelihood model based on the obstacle target predicted state set, the target signal-to-mixing ratio, and the joint measurement set. The joint measurement likelihood model includes a target intensity likelihood term and a false alarm intensity likelihood term. The target intensity likelihood term is a closed-form expression derived from the negative exponential distribution characteristics of the collision avoidance sonar echo intensity, using no prior information and marginalizing the target signal-to-mixing ratio; an update module is used to perform... - The generalized labeled Bernoulli filter is updated to obtain the multi-target posterior density of the obstacle target at the current time; - The generalized labeled Bernoulli filter includes embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio to automatically penalize false alarm measurements and perform multi-hypothesis weight updates to enhance obstacle targets.
[0098] According to another embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, it implements the multi-target tracking method of the collision avoidance system as described above.
[0099] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a multi-target tracking method for a collision avoidance system.
[0100] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.
[0101] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.
Claims
1. A multi-target tracking method for a collision avoidance system, characterized in that, Includes the following steps: Receive beam domain data from collision avoidance sonar, wherein the beam domain data includes echo intensity information in multiple beam directions; Based on the beam domain data, the joint measurement set at the current moment is obtained; Based on joint measurement set - Generalized labeled Bernoulli filter prediction yields the current set of predicted obstacle target states; Based on the obstacle target prediction state set, the target signal-to-mix ratio, and the joint measurement set, a joint measurement likelihood model is constructed. The joint measurement likelihood model includes a target intensity likelihood term and a false alarm intensity likelihood term. The target intensity likelihood term is a closed expression derived based on the negative exponential distribution characteristics of the collision avoidance sonar echo intensity. It is obtained by marginalizing the target signal-to-mix ratio using no information prior. conduct - The generalized labeled Bernoulli filter is updated to obtain the multi-target posterior density of the obstacle target at the current time; - The generalized labeled Bernoulli filter includes embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio to automatically penalize false alarm measurements and perform multi-hypothesis weight updates to enhance obstacle targets.
2. The method according to claim 1, characterized in that, The joint measurement set includes at least one candidate obstacle target measurement point, and the candidate obstacle target measurement point includes distance information, beam direction information and echo intensity information of the candidate obstacle target measurement point.
3. The method according to any one of claims 1-2, characterized in that, Embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio includes: Calculate the joint likelihood ratio of the measurement hypothesis subset relative to the predicted state of the obstacle target based on the target likelihood value and the false alarm likelihood value; The weights of the association hypothesis are determined based on the joint likelihood ratio; The multi-objective posterior density of the δ-generalized label Bernoulli filter is updated based on the associated hypothesis weights.
4. The method according to any one of claims 1-2, characterized in that, After obtaining the multi-target posterior density of the obstacle target at the current moment, the method further includes: Based on the posterior density of the δ-generalized label Bernoulli filter at the previous time step, the obstacle target is predicted to obtain the set of predicted obstacle target states at the current time step. Based on the joint measurement set, multiple measurement hypothesis subsets are constructed. The measurement hypothesis subsets are subsets of the joint measurement set and are used to represent the set of candidate obstacle target measurement information that may be generated by the same obstacle target under the corresponding association hypothesis.
5. The method according to any one of claims 1-2, characterized in that, The process - The generalized label Bernoulli filter predicts the obstacle target prediction state set at the current time, including: based on the posterior density of the δ-generalized label Bernoulli filter at the previous time, the obstacle target is predicted using the δ-generalized label Bernoulli filter to obtain the obstacle target prediction state set at the current time.
6. The method according to any one of claims 1-2, characterized in that, Based on the set of predicted states of obstacle targets and the joint measurement set, a joint measurement likelihood model is constructed, including: Multiple subsets of measurement hypotheses are constructed based on the joint measurement set; For each obstacle target predicted state and each measurement hypothesis subset, a target intensity likelihood term is determined relative to the obstacle target predicted state for each measurement hypothesis subset; the target intensity likelihood term includes a target intensity likelihood function term. Determine the false alarm intensity likelihood term corresponding to the subset of measurement hypotheses; the false alarm intensity likelihood term includes the false alarm intensity likelihood function term; The joint measurement likelihood model is constructed based on the target intensity likelihood term and the false alarm intensity likelihood term.
7. The method according to any one of claims 1-2, characterized in that, The method further includes: outputting the tracking trajectory result of the obstacle target at the current moment based on the posterior density.
8. A multi-target tracking device for a collision avoidance system, characterized in that, include: The receiving module is used to receive beam domain data of the collision avoidance sonar, wherein the beam domain data includes echo intensity information in multiple beam directions; The prediction module is used to obtain the joint measurement set at the current time based on the beam domain data; and to perform prediction based on the joint measurement set. - Generalized labeled Bernoulli filter prediction yields the current set of predicted obstacle target states; The construction module is used to construct a joint measurement likelihood model based on the obstacle target prediction state set, the target signal-to-mixing ratio, and the joint measurement set. The joint measurement likelihood model includes a target intensity likelihood term and a false alarm intensity likelihood term. The target intensity likelihood term is a closed expression derived based on the negative exponential distribution characteristics of the collision avoidance sonar echo intensity. It is obtained by marginalizing the target signal-to-mixing ratio using no information prior. The update module is used to perform... - The generalized labeled Bernoulli filter is updated to obtain the multi-target posterior density of the obstacle target at the current time; - The generalized labeled Bernoulli filter includes embedding the ratio of the target intensity likelihood term and the false alarm intensity likelihood term into the joint likelihood ratio to automatically penalize false alarm measurements and perform multi-hypothesis weight updates to enhance obstacle targets.
9. An electronic device, characterized in that, The system includes a memory and one or more processors, wherein the memory stores a computer program that, when executed by the one or more processors, implements a multi-target tracking method for a collision avoidance system as described in any one of claims 1 to 7.
10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by one or more processors, implements a multi-target tracking method for a collision avoidance system as described in any one of claims 1 to 7.