An obstacle target detection method for an unmanned underwater vehicle collision avoidance system

Abstract

The present application belongs to the technical field of underwater target detection, and particularly relates to a kind of obstacle target detection method for unmanned underwater vehicle collision avoidance system, comprising: step 1) receiving the beam data collected by the collision avoidance sonar of unmanned underwater vehicle;Step 2) adjacent N frame beam data is processed by background cancellation using interframe difference method;Step 3) the scale information and position information of the target to be detected are obtained by clustering method after background cancellation processing data;Step 4) data association is carried out, and the target detection result is output after tracking filtering processing according to the result of data association.The method of the present application further enhances the robustness of detection algorithm in complex environment, improves the efficiency of obstacle target detection of collision avoidance system, and guarantees the early warning ability and target existence confidence of collision avoidance system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of underwater target detection, and particularly relates to an obstacle target detection method for an unmanned underwater vehicle collision avoidance system. BACKGROUND

[0002] Under the background of increasing use of underwater vehicles, safe navigation has become an important issue. Underwater vehicles, especially unmanned underwater vehicles and automated underwater vehicles (AUVs), often face unknown and complex underwater environments when performing tasks such as ocean exploration, scientific research, and environmental monitoring, and there is a risk of collision with other underwater obstacles. To ensure the safety of underwater vehicles when performing tasks, the application of an obstacle avoidance system (OAS) is particularly important.

[0003] Currently, researchers mainly start from the perspective of sonar images and detect obstacle targets in sonar images. However, mapping sonar data into sonar images can cause loss of target information, and when target signals are masked by reverberation, image processing-based methods will generally have difficulty detecting obstacle targets; using a fixed detection threshold to detect obstacle targets is easily affected by the background environment, and a too high detection threshold can cause obstacle target omission, and a too low detection threshold can cause excessive false alarms.

[0004] In the actual application process of unmanned underwater vehicles, due to the complex and variable underwater environment, traditional obstacle avoidance detection algorithms process sonar images through image processing methods, but in a strong reverberation environment, target signals are masked by the reverberation environment, and traditional methods have difficulty detecting target signals; using a fixed detection threshold detection method is difficult to balance high accuracy and low false alarm rate.

[0005] Therefore, in order to adapt to complex underwater environments and complete the obstacle target detection task in the unmanned vehicle collision avoidance system, an obstacle target detection algorithm with background environment suppression and high accuracy is urgently needed to further enhance the robustness of the detection algorithm in complex environments, improve the efficiency of obstacle target detection of the collision avoidance system, and ensure the warning ability and target existence confidence of the collision avoidance system. SUMMARY

[0006] The purpose of the present application is to overcome the defects of the prior art and provide an obstacle target detection method for an unmanned underwater vehicle collision avoidance system.

[0007] Therefore, the present application provides an obstacle target detection method for an unmanned underwater vehicle collision avoidance system, which comprises:

[0008] Step 1) receiving beam data collected by the collision avoidance sonar of the unmanned underwater vehicle;

[0009] Step 2) performing background cancellation on adjacent N frames of beam data by using inter-frame difference method;

[0010] Step 3) obtaining scale information and position information of the target to be detected by using clustering method on the data after background cancellation;

[0011] Step 4) performing data association, performing tracking filtering processing according to the result of data association and outputting target detection result.

[0012] Preferably, the value of N frames in step 2) is determined according to the relative motion speed of the unmanned vehicle and the obstacle target to be detected.

[0013] Preferably, the step 2) comprises:

[0014] Step 2-1) taking adjacent N frames of beam data as a group, performing difference processing on the sonar beam data at time T and the sonar beam data at time T-1 to obtain inter-frame sonar beam difference data;

[0015] Step 2-2) inputting the sonar beam difference data into a constant false alarm detector;

[0016] Step 2-3) inputting the sonar beam data at time T into a constant false alarm detector;

[0017] Step 2-4) combining the output of the constant false alarm detector in step 2-2) with the output of the constant false alarm detector in step 2-3) to obtain the output result of the inter-frame difference constant false alarm detector.

[0018] Preferably, the step 3) comprises:

[0019] Using clustering method on the output result of the inter-frame difference constant false alarm detector obtained in step 2), obtaining scale information of the target to be detected according to the number of detection cells, and obtaining position information of the target to be detected from the row and column indexes of the beam data.

[0020] Preferably, the step 4) comprises:

[0021] Performing data association on the clustering result obtained in step 3), updating the measurement value of the tracking target, and performing Kalman filtering on the tracking target after data association.

[0022] Preferably, the strategy of data association is:

[0023] Step S1) calculating all target positions {(x 1,k ,y 1,k ),…,(x n,k ,y n,k )} in the clustering result at current time k and the jth target position in the tracking trajectory at time k-1Distance of Wherein, n represents the number of targets in the clustering result at the current time, j = 1, 2, …, m, and m represents the number of tracking trajectories at the current time;

[0024] Step S2) threshold judgment is performed: if there is an element less than or equal to a set radius r in the distance , the smallest element d in the distance i is taken as the measurement value of the target position (x i,k , y i,k ) at the current time of the tracking trajectory j, and the target position (x i,k , y i,k ) is marked and does not participate in the calculation of subsequent new target trajectory generation; wherein i = 1, 2, … n;

[0025] If there is no element less than or equal to a set radius r in the distance , it is further judged whether the tracking target j is dead: if the tracking target j does not update the measurement value for more than 4 frames, it is determined that the tracking target is dead, and the corresponding tracking trajectory is terminated;

[0026] Step S3) j is incremented by one and then goes back to step S1), until j > m, that is, the measurement value of each tracking trajectory is updated, and goes to step S4);

[0027] Step S4) after the measurement value of all tracking trajectories is updated, it is judged whether new tracking target trajectory generation is performed for all targets in the clustering result that have not been marked, and the distance t,k between all unmarked target positions (x t,k , y t,k-1 ) at the current time k and all target positions (x t,k-1 , y t,k-2 ), (x t,k-2 , y t,k-5 ), …, (x t,k-5 , y t,k ) at k-1, k-2, …, k-5 is calculated Wherein represents the distance between all targets at k-1, k-2, …, k-5 and the unmarked target position (x t,k , y i,k ), if there are more than 3 frames with less than or equal to a set radius r, it is determined that a new target appears, and a new tracking trajectory is established.

[0028] Preferably, the set radius r is determined by the size of the obstacle target and the motion speed of the vehicle.

[0029] Compared with the prior art, the advantages of the present application are:

[0030] 1. The target detection method of detection first and tracking later is applied to the obstacle target detection of the underwater unmanned vehicle collision avoidance system, wherein the detection part and the tracking part can be optimized independently.

[0031] 2. The constant false alarm detector using the background cancellation method has higher detection rate than the constant false alarm detector of the conventional method, and can detect the target at a farther distance, thereby providing more operation space for the underwater vehicle to avoid obstacles and perform corresponding avoidance actions.

[0032] 3. The method further enhances the robustness of the detection algorithm in a complex environment, improves the efficiency of the obstacle target detection of the collision avoidance system, and guarantees the early warning ability and target existence confidence of the collision avoidance system. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is a detection first and tracking later framework of the obstacle target detection algorithm;

[0034] Figure 2 is an inter-frame "difference" constant false alarm detector;

[0035] Fig. 3 is the effect of the inter-frame "difference" operation, wherein Fig. 3(a) is an acoustic image, Fig. 3(b) is a target detection result image of a conventional constant false alarm detector, and Fig. 3(c) is an inter-frame "difference" constant false alarm detector result image;

[0036] Figure 4 is a processing flowchart of the target detection module;

[0037] Figure 5 is a processing flowchart of the tracking filter module;

[0038] Fig. 6 is a simulation experiment processing result, wherein Fig. 6(a) is an acoustic image, Fig. 6(b) is a target detection result image, Fig. 6(c) is a tracking filter result image, Fig. 6(d) is a comparison image of the real trajectory and the target tracking trajectory, Fig. 6(e) is a comparison image of the horizontal direction and the vertical direction target tracking trajectory and the actual trajectory, and Fig. 6(f) is an OSPA index;

[0039] Fig. 7 is a pool experiment result, wherein Fig. 7(a) is an acoustic image, Fig. 7(b) is a target detection result image, Fig. 7(c) is a tracking filter result image, Fig. 7(d) is a comparison image of the real trajectory and the target tracking trajectory, Fig. 7(e) is a comparison image of the horizontal direction and the vertical direction target tracking trajectory and the actual trajectory, and Fig. 7(f) is an OSPA index. DETAILED DESCRIPTION

[0040] Under the background of the increasing use of underwater vehicles, in order to protect the navigation safety of the vehicle, the application of the collision avoidance system is very important. The obstacle target detection part provides important information reference for the early warning and maneuvering of the collision avoidance system. In order to suppress reverberation and improve target detection probability, the method of detecting first and tracking later is used in the target detection process of the collision avoidance system. The method has high execution efficiency while taking into account the detection performance, and can meet the real-time requirements of mobile devices. In order to solve the problem that the echo signal of the obstacle target is masked by the reverberation, the background cancellation method is introduced to suppress the reverberation interference and improve the detection probability of the obstacle target in the reverberation background environment. In the detection part, the OS-CFAR constant false alarm detector is used to adaptively adjust the threshold value, and the target detection rate is improved. In the tracking part, the Kalman filter is introduced to remove the reverberation interference, filter out the obstacle target, and stably monitor the obstacle target, so as to form a continuous and stable target tracking trajectory relative to the sonar device.

[0041] The technical solutions of the present application will be described in detail below in combination with the drawings and embodiments.

[0042] Embodiments

[0043] Therefore, for the obstacle target detection problem of the underwater unmanned vehicle collision avoidance system, the embodiment of the present application proposes an obstacle target detection method for the unmanned underwater vehicle collision avoidance system. The target detection algorithm of detecting first and tracking later is introduced. The detection part and the tracking part can be independently optimized to adapt to different scenes, and are suitable for the demand of the underwater vehicle collision avoidance system working in complex environment. In the target detection framework of detecting first and tracking later, the constant false alarm detector with anti-interference ability is used in the detection part to improve the detection probability of the target, and the Kalman filter is used in the tracking part to remove the reverberation interference and reduce false alarm.

[0044] The method comprises the following steps:

[0045] Step 1) receiving the beam data collected by the collision avoidance sonar of the unmanned underwater vehicle;

[0046] Step 2) performing background cancellation processing on adjacent N frames of beam data by using the inter-frame difference method;

[0047] Step 3) obtaining the size information and position information of the to-be-detected target by using a clustering method on the data after the background cancellation processing;

[0048] Step 4) performing data association, performing tracking filtering processing according to the result of the data association, and outputting the target detection result.

[0049] 1. Obstacle target detection algorithm flow of underwater unmanned vehicle collision avoidance system

[0050] The processing flow of the present application is as followsFigure 1 As shown, the beam data in the collision avoidance sonar is taken as input, the input multi-frame sonar data is first processed by background cancellation, then the processed data is judged by threshold to determine the position of the target, and finally the tracking filtering processing is performed according to the result of threshold judgment and the target detection result is output.

[0051] The inter-frame "difference" method is used in the background cancellation part to suppress the background environment, and the background cancellation can improve the contrast of the target signal in the background environment, so that the constant false alarm detector can more easily detect the target and improve the detection probability of the system. The target detection part is mainly composed of an inter-frame difference constant false alarm detector and a Dbscan clustering algorithm. The inter-frame difference constant false alarm detector can adaptively adjust the threshold according to the background environment, has the ability to balance the background environment noise and keep the false alarm rate constant while improving the target detection. The tracking filtering part is mainly composed of data association and Kalman filter. In order to improve the confidence of the existence of the obstacle target, reduce the target detection false alarm rate, obtain the stable and continuous relative motion trace of the detected target, and eliminate the reverberation interference, a target tracking algorithm is used. Finally, the output of the tracking filtering and the output of the target detection module are used to identify the target, and the final detection result of the system is obtained.

[0052] Inter-frame "difference" constant false alarm detector

[0053] The processing flow of the inter-frame "difference" method is as follows Figure 2 As shown, several adjacent frames of data (need to be determined according to the relative motion speed of the unmanned vehicle and the obstacle target, usually 3-5 frames) are taken as a group, the sonar beam data at the current time is subtracted from the data at the last time to obtain the inter-frame "difference" sonar beam data, and then the inter-frame "difference" data is input into the OS-CFAR detector. Since the difference operation can eliminate the static target, in order to preserve the information of the static target in the sonar data, the sonar beam data at the current time is also input into the OS-CFAR detector for processing. The outputs of the two are combined to obtain the output result of the inter-frame "difference" constant false alarm detector.

[0054] Fig. 3 shows the benefits brought by the inter-frame difference operation, wherein Fig. 3(a) is a front-looking sonar beam intensity diagram, Fig. 3(b) is the result after conventional CFAR detection, and Fig. 3(c) is the result detected by the inter-frame "difference" constant false alarm detector. By comparison, it can be seen that the target intensity is not obvious under the reverberation background environment interference, and the target is submerged in the background environment. After processing by the inter-frame "difference" constant false alarm detector, the influence of the background environment can be obviously suppressed, and the detection of the obstacle target is completed.

[0055] Target detection module

[0056] The target detection module consists of two parts: an inter-frame differential constant false alarm rate detector and a Dbscan clustering algorithm. Figure 4 As shown, the output of the inter-frame differential constant false alarm rate (CFAR) detector is used as the input to the clustering algorithm (Dbscan). Through processing by the inter-frame differential CFAR detector and the clustering algorithm, the target detection module completes the detection of sonar data, and its output includes the target's location and scale information. The reason for introducing the clustering algorithm is that obstacle targets are usually extended targets, which typically occupy multiple detection cells in the inter-frame differential CFAR detector. Therefore, it is necessary to group these detection cells into a class. The scale information of the target can be determined based on the number of detection cells occupied by the clustering result. Meanwhile, the row and column indices of the beam data (usually a matrix) contain real-world location information. Therefore, combining the two can determine the location and scale information of the sonar device relative to the obstacle target.

[0057] Tracking filter module

[0058] In actual sonar images, the sound waves returned by scatterers undergo multiple reflections from the seabed and sea surface, resulting in a "flickering" characteristic similar to targets on the sonar image. This causes many suspected target bright spots to still exist after processing by the target detection module. To obtain a higher detection probability and confidence level for obstacle targets, reduce the false alarm rate of target detection, and eliminate reverberation interference, a tracking algorithm needs to be introduced.

[0059] The tracking filter module takes the target position information output by the target detection module as input and outputs the target's state information as output. The processing flow of the tracking filter module is as follows: Figure 5 As shown, it consists of two parts: data association and Kalman filter. The data association part includes distance threshold decision, target birth, target death, and target measurement update. The distance decision uses formula (1), which means that only the measurement value at a certain distance from the tracked target can be included in the measurement value of the corresponding trajectory. The threshold value r needs to be set according to the motion state of the target and the sonar equipment.

[0060]

[0061] Where x i,k y i,k This represents the x-coordinate and y-coordinate positions of target i at time k in a Cartesian coordinate system. d represents the estimated x and y coordinates of the tracked target j at time k in a Cartesian coordinate system; d represents the distance between the two; and r represents the decision threshold, which is determined by the size of the obstacle target and the speed of the vehicle.

[0062] After the threshold decision, the measurement value of the tracking target is updated. If the tracking target is not updated for several times, it is determined that the target is dead and the tracking trajectory is terminated. If the measurement value is not included in any tracking target, it enters the new target library. If the new target appears for several times, it is determined that the target exists and a new trajectory is established.

[0063] The specific strategy of data association is:

[0064] The distance between all target positions {(x 1,k ,y 1,k ),…,(x n,k ,y n,k} in the clustering result at the current time k and the jth target position in the tracking trajectory at the time k-1 is calculated. Where n represents the number of targets in the clustering result at the current time, j=1,2,…,m, and m represents the number of tracking trajectories at the current time.

[0065] Threshold judgment is performed. If there is an element in d that is less than or equal to the set radius r, the smallest element d i in d is taken. The target position (x i,k ,y i,k ) corresponding to d i (i=1,2,…n) is taken as the measurement value of the tracking trajectory j at the current time, and the target position (x i,k ,y i,k ) is marked and does not participate in the calculation of subsequent new target trajectory generation.

[0066] If there is no element in d that is less than or equal to the set radius r, it is necessary to determine whether the tracking target j is dead. If the tracking target j has not been updated for more than 4 frames, it is determined that the tracking target is dead and the corresponding tracking trajectory is terminated.

[0067] After j is added, the above operation is repeated until j>m, that is, the measurement value of each tracking trajectory is updated.

[0068] After the measurement value of all tracking trajectories is updated, the judgment of new tracking target trajectory generation is performed on all unmarked targets in the clustering result. The distance between all unmarked target positions (x t,k ,y t,k ) at the current time k and all target positions (x t,k-1 ,y t,k-1 ),(x t,k-2 ,y t,k-2),…,(x t,k-5 ,y t,k-5 ) of all targets and unmarked target position (x wherein represent the distances of all targets and unmarked target position (x t,k ,y t,k ) at k-1, k-2,…k-5 time, if there are more than 3 frames with distances less than or equal to the set radius r, it is determined that a new target appears, and a new tracking trajectory is established.

[0069] Obstacle target detection algorithm results

[0070] To test the applicability of the method and the accuracy of the detected target, simulation experiments and pool experiments were carried out, and the performance of the algorithm was qualitatively analyzed.

[0071] As shown in Fig. 6(a) is a simulation sound chart under single target 10dB condition, target movement speed 0.4m / s, gradually close to the sonar from far to near, active sonar emits CW signal, frequency 200KHz, pulse width 30us, range 35m. Fig. 6(b) is the result of the simulation sound chart after differential CFAR processing, Fig. 6(c) is the trajectory and position result of the target after processing by the detection algorithm, Fig. 6(d) is the real trajectory of the target movement and the movement trajectory of the target detection, Fig. 6(e) is a schematic diagram of the real trajectory and the target state estimation value changing with time in X direction and Y direction, and Fig. 6(f) is the OSPA performance analysis of the detection algorithm under 10dB, wherein the calculation method of OSPA is shown in the following (2), and the smaller the value of OSPA, the higher the multi-target tracking accuracy.

[0072]

[0073] wherein X={x1,x2,…,x m} and Y={y1,y2,…,y n} represent the real target set and the estimated target set respectively; m and n represent the number of real targets and estimated targets respectively; c represents the truncation parameter, which is set to 30; p is the order parameter, which is set to 1; ||x i -y π(i) || represents the distance between the ith real target x i and the assigned estimated target y π(i) ; c p *|n-m| represents the number error caused by unmatched targets.

[0074] The processing result of the pool experiment is shown in Fig. 7. The data scanned by the forward-looking sonar after the target is shown in Fig. 7(a). The target is marked by a red box. The echo signal of the target is submerged in the strong reverberation environment, and it is difficult to identify the target in a single frame image. After the inter-frame "difference" constant false alarm detection, the detection result is shown in Fig. 7(b). The target is successfully detected, but the detection result contains a large amount of reverberation interference. It is difficult to determine the direction and distance of the target only by the constant false alarm detection result. The constant false alarm rate of the OS-CFAR is set to 10 -4 , N = 32. The result after the processing of the tracking filter module is shown in Fig. 7(c). By comparing with Fig. 7(a), it can be found that the target can be detected from the numerous reverberation interferences and the motion trajectory of the target can be obtained after the tracking processing. Fig. 7(d) is the real trajectory of the target motion and the estimated motion trajectory of the target state. Fig. 7(e) is a schematic diagram of the real trajectory and the estimated value of the tracking method in the X direction and the Y direction changing with time. Fig. 7(f) is an OSPA diagram of the improved method and the traditional method in the pool experiment. The horizontal coordinate is time, and the vertical coordinate is the OSPA index.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit it. Although the present application is described in detail with reference to the embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. An obstacle target detection method for a collision avoidance system of an unmanned underwater vehicle, comprising: Step 1) Receive beam data collected by the collision avoidance sonar of the unmanned underwater vehicle; Step 2) Background cancellation processing is performed on the adjacent N frames of beam data using the inter-frame difference method; Step 3) Use clustering to obtain the scale and location information of the target to be detected from the background cancellation data; Step 4) Perform data association, perform tracking filtering based on the data association results, and output the target detection results; Step 2) includes: Step 2-1) Take the adjacent N frames of beam data as a group, and perform subtraction on the sonar beam data at time T and the sonar beam data at time T-1 to obtain inter-frame sonar beam differential data. Step 2-2) Input the sonar beam differential data into the constant false alarm rate detector; Steps 2-3) Input the sonar beam data at time T into the constant false alarm rate detector; Step 2-4) Combine the output of the constant false alarm rate detector in step 2-2) with the output of the constant false alarm rate detector in step 2-3) to obtain the output result of the inter-frame differential constant false alarm rate detector.

2. The obstacle target detection method for an unmanned underwater vehicle collision avoidance system according to claim 1, characterized in that, The N-frame value in step 2) is determined based on the relative motion speed between the unmanned vehicle and the obstacle target to be detected.

3. The obstacle target detection method for an unmanned underwater vehicle collision avoidance system according to claim 1, characterized in that, Step 3) includes: The output of the inter-frame differential constant false alarm detector obtained in step 2) is clustered to obtain the scale information of the target to be detected based on the number of detection units occupied, and the position information of the target to be detected is obtained from the row and column index of the beam data.

4. The obstacle target detection method for an unmanned underwater vehicle collision avoidance system according to claim 1, characterized in that, Step 4) includes: The clustering results obtained in step 3) are correlated with data to update the measurement values ​​of the tracked target. After data correlation, Kalman filtering is applied to the tracked target.

5. The obstacle target detection method for an unmanned underwater vehicle collision avoidance system according to claim 4, characterized in that, The data association strategy is as follows: Step S1) Calculate all target locations {(x) in the clustering results at the current time k. 1,k ,y 1,k ),…,(x n,k ,y n,k The position of the j-th target in the tracking trajectory at time k-1. distance Where n represents the number of targets in the clustering results at the current time, j = 1, 2, ..., m, and m represents the number of tracking trajectories at the current time; Step S2) Perform threshold judgment: If If there exists an element in the set radius r that is less than or equal to the set radius r, then take it. The smallest element d in i Corresponding target position (x) i,k ,y i,k ) is used as the measurement value of the current moment of the tracking trajectory j, and the target position (x) is used as the measurement value. i,k ,y i,k The markers (i, 2, ..., n) are not used in the calculation of subsequent new target trajectory generation; where i = 1, 2, ..., n. like If there are no elements less than or equal to the set radius r, then it is further determined whether the tracked target j is dead: if the tracked target j has not updated its measurement value for more than 4 frames, then the tracked target is determined to be dead and the corresponding tracking trajectory is terminated. After incrementing j by one in step S3), proceed to step S1) until j > m, that is, the measurement value has been updated for each tracking trajectory, and then proceed to step S4). Step S4) After updating the measurement values ​​of all tracking trajectories, determine the generation of new tracking target trajectories for all unlabeled targets in the clustering results, and calculate the positions (x, k) of all unlabeled targets at the current time k. t,k ,y t,k ) and all target positions (x) in the previous 5 frames. t,k-1 ,y t,k-1 ),(x t,k-2 ,y t,k-2 ),…,(x t,k-5 ,y t,k-5 distance in These represent the positions (x, k) of all targets and unmarked targets at times k-1, k-2, ..., k-5, respectively. t,k ,y t,k If there are more than 3 frames at a distance of ) If the radius is less than or equal to the set radius r, a new target is identified and a new tracking trajectory is established.

6. The obstacle target detection method for an unmanned underwater vehicle collision avoidance system according to claim 5, characterized in that, The set radius r is determined by the size of the obstacle target and the speed of the vehicle.

Images