Active sonar man-machine cooperation multi-target tracking method

Abstract

The invention belongs to the technical field of underwater acoustic signal processing, and particularly relates to an active sonar man-machine cooperation multi-target tracking method. Comprising the following steps: carrying out balanced denoising processing on background noise, and automatically detecting a threshold value to obtain azimuth and distance information of a target; meanwhile, the detection threshold value is manually adjusted through manual intervention; establishing a track for the detection point through an automatic algorithm and distributing a track ID; meanwhile, through manual intervention, suspected target points are manually recorded; calculating the proximity of each detection point and each track through an automatic algorithm, and carrying out track association according to a proximity matrix; meanwhile, through manual intervention, track ID affiliation is manually corrected; according to the invention, a hybrid tracking mode of'automatic-based and manual-assisted 'is formed, a full-automatic tracking algorithm is used for processing a conventional process, an abnormal decision is manually intervened, a high-reliability man-machine cooperative active sonar target tracking system is established, and the tracking robustness in a complex environment is improved by dynamically switching a manual / automatic mode.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic signal processing technology, and specifically relates to an active sonar human-machine collaborative multi-target tracking method. Background Technology

[0002] Active sonar obtains target location and range information by emitting sound waves and performing a series of signal processing steps on the received echoes, including beamforming, matched filtering, background estimation, and thresholding, thus achieving single-cycle target detection. Due to the complexity of the underwater environment, reverberation, clutter, and multipath propagation inevitably lead to errors, false alarms, and missed detections in target detection results. Multi-target tracking, building upon single-cycle target detection, establishes appropriate motion and observation models and utilizes adaptive filtering and data association algorithms to provide continuous target trajectories. This technique can accurately estimate target positions, eliminate false alarms, and compensate for missed targets. Furthermore, multi-cycle association is also a crucial foundation for target feature extraction and target recognition.

[0003] Traditional active sonar target tracking systems rely on manual guidance. Operators combine real-time target information with their personal experience to determine the target's location and establish a track. Due to limitations in human response speed and capability, they typically only support tracking up to six targets simultaneously. With the development of target tracking technology, multi-target fully automatic tracking technology has been widely applied in radar tracking, video tracking, and other fields. However, underwater target tracking started relatively late. Due to the complexity of the underwater environment, relying entirely on automatic algorithms may lead to tracking interruptions or mistracking due to sudden environmental changes or model mismatch.

[0004] In summary, due to the complexity of the underwater environment, the dynamic changes of the target, and the limitations of sensor performance, active sonar target tracking systems face the following problems if they rely solely on manual or fully automatic tracking:

[0005] 1) Bottlenecks of purely automatic algorithms: Active sonar detects targets by emitting sound waves, but in complex seabed topography and between water layers, sound waves are reflected multiple times, creating continuous reverberation noise that may submerge target echoes, leading to missed detections. Simultaneously, marine biomes, bubble swarms, seabed sediments, seamounts, and reefs generate non-uniform clutter with statistical characteristics similar to real targets, easily generating false alarms and causing data association errors. Furthermore, in situations of target intersection or obstruction, acoustic echoes superimpose at the receiver, and purely automatic algorithms, lacking long-term trajectory memory, are prone to interrupting tracking or mistracking.

[0006] 2) The drawbacks of purely manual tracking: While purely manual tracking has advantages in situations with high interference and concealed targets, it is highly dependent on operator experience. The acoustic characteristics of underwater targets need to be manually identified, resulting in high training costs and strong subjectivity. At the same time, long-term monitoring can easily lead to operator fatigue, and it is even more difficult to simultaneously monitor multiple targets. Summary of the Invention

[0007] The purpose of this invention is to provide an active sonar multi-target tracking method that combines manual intervention and automatic algorithms. By dynamically switching between manual and automatic modes, the tracking robustness in complex environments is improved, and it is especially suitable for scenarios with strong reverberation, multiple clutter, and target cross-occlusion.

[0008] To address the aforementioned technical problems, this invention provides an active sonar human-machine collaborative multi-target tracking method, comprising:

[0009] Target detection; the target's location and distance information are obtained by equalizing and denoising the background noise and automatically detecting a threshold; the detection threshold is manually adjusted through manual intervention.

[0010] Track establishment; tracks are automatically established for detection points and track IDs are assigned; at the same time, suspected target points are manually recorded through manual intervention;

[0011] Track association; the proximity between each detection point and each track is calculated automatically using an algorithm, and track association is performed based on the proximity matrix; at the same time, manual intervention is used to manually correct the track ID attribution;

[0012] Track maintenance: The position of all tracks is predicted for the next moment using an automatic algorithm. When a track is not associated with a real detection point, the predicted point is extrapolated. At the same time, manual intervention is used to manually correct or delete abnormal extrapolated or lost tracks.

[0013] Target identification; targets are identified using automatic algorithms, and the target attributes of each track are provided; at the same time, the target attributes of tracks with known target attributes are manually modified through human intervention.

[0014] Preferably, in the target detection, by adding manual intervention points, it is possible to manually adjust the detection threshold by manually marking suspicious areas in the full background output screen.

[0015] Preferably, in the track establishment process, an automatic algorithm establishes tracks for all detection points in the first cycle and assigns a track ID to each track. In non-first cycles, new tracks are established for detection points that are not associated with existing tracks and track IDs are assigned. At the same time, by adding manual intervention points, manual recording is allowed at locations where no target point is automatically detected but a target is suspected, and new tracks are established and track IDs are assigned.

[0016] Preferably, in the track association, the automatic algorithm adopts the nearest neighbor data association algorithm.

[0017] Preferably, the proximity between each detection point and each track is calculated using the nearest neighbor data association algorithm, and the tracks are associated based on the proximity matrix. If the proximity between a detection point and a track is less than a threshold, the track is associated with that track. At the same time, by adding a manual intervention point, the track ID attribution can be manually corrected when tracks intersect or split.

[0018] Preferably, in the track maintenance, the automatic algorithm employs the Kalman filter algorithm.

[0019] Preferably, the Kalman filter algorithm is used to predict the position of all tracks at the next moment. When a track is not associated with a real detection point, the predicted point is extrapolated. When the extrapolation of a track exceeds a set threshold, the track is judged to be lost. At the same time, by adding a manual intervention point, manual correction or deletion is allowed for abnormal extrapolation or lost tracks.

[0020] Preferably, in the target recognition, the automatic algorithm employs a multi-period correlation target multi-dimensional feature fusion recognition algorithm.

[0021] This invention also provides an active sonar human-machine collaborative multi-target tracking system, which executes the active sonar human-machine collaborative multi-target tracking method as described above, including:

[0022] The target detection module performs equalization and noise reduction on the background noise and automatically detects a threshold to obtain the target's location and distance information; it also has a manual intervention point to manually adjust the detection threshold.

[0023] The track establishment module automatically creates tracks for detection points and assigns track IDs; it also has a manual intervention point to manually record suspected target points.

[0024] The track association module uses an automatic algorithm to calculate the proximity between each detection point and each track, and associates tracks based on the proximity matrix; it also has a manual intervention point to manually correct the track ID attribution.

[0025] The track maintenance module uses an automatic algorithm to predict the position of all tracks at the next moment. When a track is not associated with a real detection point, the predicted point is extrapolated. It also has a manual intervention point to manually correct or delete abnormal extrapolated or lost tracks.

[0026] The target recognition module uses automatic algorithms to identify targets and provides target attributes for each track; it also has a manual intervention point to manually modify the target attributes of tracks with known target attributes.

[0027] The present invention also provides an underwater acoustic signal processing device, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, it implements an active sonar human-machine collaborative multi-target tracking method as described above.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] This invention combines the advantages of both approaches, forming a hybrid tracking mode of "automatic primary and manual secondary." The fully automated tracking algorithm handles routine processes, while manual intervention facilitates abnormal decision-making, establishing a highly reliable human-machine collaborative active sonar target tracking system. The entire process begins with target detection and ends with target recognition, with the output of each module serving as the input for the next. The manual intervention points are designed as parallel paths, allowing real-time intervention during the automatic algorithm's execution to enhance the system's robustness and adaptability. Attached Figure Description

[0030] Figure 1 This is a flowchart of an active sonar human-machine collaborative multi-target tracking method according to the present invention. Detailed Implementation

[0031] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The advantages and features of the present invention will become clearer from the following description. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0032] like Figure 1 As shown in the figure, this embodiment of the invention specifically provides an active sonar human-machine collaborative multi-target tracking method, which includes the following steps:

[0033] Step 1: Human-machine collaborative design of the target detection module;

[0034] In the target detection module, the background is first equalized and denoised based on the background noise, and then the threshold detection is automatically performed to obtain the target's location and distance information. At the same time, a manual intervention point is added, which can mark suspicious areas in the full background output screen and manually adjust the detection threshold.

[0035] Step 2: Human-machine collaborative design of the trajectory establishment module;

[0036] In the track establishment module, the automatic algorithm establishes tracks for all detection points in the first cycle and assigns a track ID to each track. In subsequent cycles, it establishes new tracks for detection points that are not associated with existing tracks and assigns IDs. At the same time, a manual intervention point is added. In locations where no target point is automatically detected but a target is suspected, manual entry can be performed to establish a new track and assign a track ID.

[0037] Step 3: Human-machine collaborative design of the trajectory association module;

[0038] In the track association module, the automatic algorithm uses the nearest neighbor data association algorithm to calculate the proximity between each detection point and each track, and performs track association based on the proximity matrix. If the proximity between a detection point and a certain track is less than a threshold, it is associated with that track. At the same time, a manual intervention point is added so that the track ID can be manually corrected when tracks intersect or split.

[0039] Step 4: Human-machine collaborative design of the trajectory maintenance module;

[0040] In the track maintenance module, the automatic algorithm uses the Kalman filter algorithm to predict the position of all tracks at the next moment. When a track is not associated with a real detection point, the predicted point is extrapolated. When the extrapolation of a track exceeds the set threshold, the track is judged to be lost. At the same time, a manual intervention point is added to manually correct or delete tracks that are abnormally extrapolated or lost.

[0041] Step 5: Human-machine collaborative design of the target recognition module;

[0042] In the target recognition module, the automatic algorithm uses multi-period correlation of multi-dimensional target features to perform target recognition and gives the target attributes of each track. At the same time, a manual intervention point is added to manually modify the target attributes of tracks with known target attributes (such as water targets that can be excluded by radar), which can save the computing resources occupied by the track and reduce the false alarm rate.

[0043] This invention also provides an active sonar human-machine collaborative multi-target tracking system, which executes the active sonar human-machine collaborative multi-target tracking method as described above, including:

[0044] The target detection module performs equalization and noise reduction on the background noise and automatically detects a threshold to obtain the target's location and distance information; it also has a manual intervention point to manually adjust the detection threshold.

[0045] The track establishment module automatically creates tracks for detection points and assigns track IDs; it also has a manual intervention point to manually record suspected target points.

[0046] The track association module uses an automatic algorithm to calculate the proximity between each detection point and each track, and associates tracks based on the proximity matrix; it also has a manual intervention point to manually correct the track ID attribution.

[0047] The track maintenance module uses an automatic algorithm to predict the position of all tracks at the next moment. When a track is not associated with a real detection point, the predicted point is extrapolated. It also has a manual intervention point to manually correct or delete abnormal extrapolated or lost tracks.

[0048] The target recognition module uses automatic algorithms to identify targets and provides target attributes for each track; it also has a manual intervention point to manually modify the target attributes of tracks with known target attributes.

[0049] This invention also provides an underwater acoustic signal processing device, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, it implements an active sonar human-machine collaborative multi-target tracking method as described above.

[0050] The above description is merely a description of preferred embodiments of the present invention and is not intended to limit the scope of the present invention in any way. Any changes or modifications made by those skilled in the art based on the above disclosure shall fall within the protection scope of the claims.

Claims

1. An active sonar human-in-the-loop multi-target tracking method, characterized in that, The method comprises: Target detection; By equalizing the background noise and automatically detecting the threshold value to obtain the orientation and distance information of the target; meanwhile, by manual intervention, the detection threshold value is manually adjusted; Track establishment; by an automatic algorithm, a track is established for a detected point and a track ID is assigned; meanwhile, by manual intervention, a suspected target point is manually recorded; Track association; by an automatic algorithm, the proximity of each detected point to each track is calculated, and track association is performed according to the proximity matrix; meanwhile, by manual intervention, the track ID attribution is manually corrected; Track maintenance; by an automatic algorithm, the position of each track at the next moment is predicted, and when a track is not associated with a real detected point, the predicted point is extrapolated; meanwhile, by manual intervention, an abnormally extrapolated or lost track is manually corrected or deleted; Target identification; by an automatic algorithm, the target is identified, and the target attribute of each track is given; meanwhile, by manual intervention, the target attribute of a track with a known target attribute is manually modified.

2. The active sonar human-in-the-loop multi-target tracking method of claim 1, wherein, In the target detection, by adding a manual intervention point, a suspected area is manually marked on a full background output picture, and the detection threshold value is manually adjusted.

3. The active sonar human-in-the-loop multi-target tracking method of claim 1, wherein, In the track establishment, by an automatic algorithm, a track is established for all detected points in a first period, and a track ID is assigned to each track; in a non-first period, a new track is established for a detected point that is not associated with an existing track, and a track ID is assigned; meanwhile, by adding a manual intervention point, a new track and a track ID are established and assigned in a position where no target point is automatically detected but a target is suspected.

4. The active sonar human-in-the-loop multi-target tracking method of claim 1, wherein, In the track association, the automatic algorithm adopts a nearest neighbor data association algorithm.

5. The active sonar human-in-the-loop multi-target tracking method of claim 4, wherein, By the nearest neighbor data association algorithm, the proximity of each detected point to each track is calculated, and track association is performed according to the proximity matrix; if the proximity of a detected point to a track is less than a threshold value, the detected point is associated with the track; meanwhile, by adding a manual intervention point, the track ID attribution is manually corrected when tracks intersect or split.

6. The active sonar human-in-the-loop multi-target tracking method of claim 1, wherein, In the track maintenance, the automatic algorithm adopts a Kalman filter algorithm.

7. The active sonar human-in-the-loop multi-target tracking method of claim 6, wherein, By the Kalman filter algorithm, the position of each track at the next moment is predicted, and when a track is not associated with a real detected point, the predicted point is extrapolated; when a track is continuously extrapolated for more than a set threshold value, the track is judged to be lost; meanwhile, by adding a manual intervention point, an abnormally extrapolated or lost track is manually corrected or deleted.

8. The active sonar human-in-the-loop multi-target tracking method of claim 1, wherein, In the target identification, the automatic algorithm adopts a multi-cycle associated target multi-dimensional feature fusion identification algorithm.

9. An active sonar human-in-the-loop multi-target tracking system, performing an active sonar human-in-the-loop multi-target tracking method according to any one of claims 1 to 8, characterized in that, The method comprises: A target detection module, which equalizes the background noise and automatically detects the threshold value to obtain the orientation and distance information of the target; meanwhile, the detection threshold value is manually adjusted by a manual intervention point; A track establishment module, which establishes a track for a detected point by an automatic algorithm and assigns a track ID; meanwhile, a suspected target point is manually recorded by a manual intervention point; A track association module, which calculates the proximity of each detected point to each track by an automatic algorithm, and performs track association according to the proximity matrix; meanwhile, the track ID attribution is manually corrected by a manual intervention point; The track maintenance module predicts the position of all tracks at the next time through an automatic algorithm, and extrapolates the predicted point when a track is not associated with a real detection point; meanwhile, the track maintenance module has a manual intervention point to manually correct or delete an abnormal extrapolated or lost track. The target identification module identifies a target through an automatic algorithm, and gives the target attribute of each track; meanwhile, the target identification module has a manual intervention point to manually modify the target attribute of a track with a known target attribute.

10. An underwater acoustic signal processing device comprising a memory, a processor and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the active sonar human-machine cooperative multi-target tracking method according to any one of claims 1-8.

Images