CN113283330A - Video SAR moving target detection method based on deep learning and multi-target tracking algorithm - Google Patents
Video SAR moving target detection method based on deep learning and multi-target tracking algorithm Download PDFInfo
- Publication number
- CN113283330A CN113283330A CN202110551104.1A CN202110551104A CN113283330A CN 113283330 A CN113283330 A CN 113283330A CN 202110551104 A CN202110551104 A CN 202110551104A CN 113283330 A CN113283330 A CN 113283330A
- Authority
- CN
- China
- Prior art keywords
- tracking
- candidate
- detection result
- target
- moving target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 46
- 238000013135 deep learning Methods 0.000 title claims abstract description 14
- 238000000034 method Methods 0.000 claims abstract description 10
- 230000000694 effects Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The invention discloses a video SAR moving target detection method based on deep learning and multi-target tracking algorithm, and belongs to the technical field of radar. The method comprises the following steps: 1: detecting a moving target shadow in a single-frame video SAR image by adopting a fast R-CNN network based on a characteristic pyramid structure; 2: initializing a tentative tracker by using the detection result; 3: updating the tracker with the associated detection result; 4: changing tentative tracking of continuously updated three frames into stable tracking and deleting the rest tentative tracking; 5: and converting the stable tracker which is not updated by three continuous frames into a candidate tracking. 6: and if the candidate tracking is updated according to the detection result, converting the candidate tracking into stable tracking. 7: and if the number of times that the candidate tracker is not updated exceeds the threshold value, deleting the candidate tracker. The invention has the characteristics of high detection precision and strong real-time property.
Description
Technical Field
The invention relates to a video SAR moving target detection method based on deep learning and multi-target tracking algorithm, belonging to the technical field of radar.
Background
Synthetic Aperture Radars (SAR) have important significance in the field of modern remote sensing due to their ability to image a target area around the clock and all day long. However, the conventional SAR system has a low imaging frame rate and cannot provide reliable moving target positioning. The video SAR is used as a new imaging mode, can continuously image a target scene at a high resolution at a high frame rate, continuously monitors a target area in a dynamic mode, and visually reflects dynamic information such as motion of a target, scene change and the like. Due to the high working frequency of the video SAR, the Doppler modulation of the echo of a moving target is sensitive to the movement of the moving target, and the movement of the target can cause the moving target to generate defocusing and shifting in an image and leave a shadow at the real position of the moving target. Therefore, the moving target can be monitored by detecting the shadow.
In recent years, the method based on deep learning has achieved the most excellent detection result in target detection. Numerous scholars have explored the feasibility of deep learning in SAR image target detection. However, because the features of the SAR image are simple, a deep neural network is only used for detecting the shadow of the moving target, so that more false alarms and false-alarms are generated, and a detection method based on deep learning needs to be improved urgently.
Disclosure of Invention
Aiming at the problems mentioned in the background technology, the invention provides a video SAR moving target detection method based on deep learning and multi-target tracking algorithm.
The invention adopts the following technical scheme for solving the technical problems:
a video SAR moving target detection method based on deep learning and multi-target tracking algorithm comprises the following steps:
step 1: detecting a moving target shadow in a single-frame video SAR image by adopting a fast R-CNN network based on a characteristic pyramid structure;
step 2: initializing a tentative tracker by using the detection result in the step 1;
and step 3: associating the detection result with the trace and updating the tracker using the associated detection result;
and 4, step 4: changing tentative tracking of continuously updated three frames into stable tracking and deleting the rest tentative tracking;
and 5: converting the stable tracking which is not updated in three continuous frames into candidate tracking;
step 6: if the associated detection result updates the candidate tracking in the step 5, the candidate tracking is changed into stable tracking;
and 7: and if the number of times of the candidate tracks which are not updated in the step 5 exceeds a threshold value, deleting the candidate tracks.
The detection result in the step 2 is described by using the following eight-dimensional state space:
wherein x and y represent the horizontal and vertical coordinates of the center of the object, a and h represent the width-height ratio and height of the bounding box of the object,representing the rate of change of x, y, a, and h, respectively.
And the detection result and the tracking association in the step 3 are realized by Hungarian algorithm.
The invention has the following beneficial effects:
1. the method firstly utilizes a fast R-CNN (fast regional convolutional neural network) target detection network based on a characteristic pyramid structure to detect the moving target shadow in the single-frame video SAR image. And then, a multi-target tracking algorithm based on detection is adopted to track the moving target shadow, so that the detection probability is improved, and the false alarm rate is reduced.
2. The method applies a multi-target tracking algorithm based on detection to the detection of the moving target shadow of the video SAR, and has the characteristics of higher detection precision, simplicity and easiness in implementation and better real-time property.
Drawings
Fig. 1 is a tracking processing flow chart of a video SAR moving target shadow detection method based on deep learning and multi-target tracking algorithm of the present invention.
FIGS. 2(a) and 2(b) are both preliminary test result graphs of Faster R-CNN based on the feature pyramid structure.
Fig. 3(a) and fig. 3(b) are both two multi-target tracking algorithm processing result graphs.
Detailed Description
The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. The described embodiments of the present invention are only for explaining the present invention and do not constitute a limitation to the present invention.
As shown in fig. 1, the embodiment provides a method for detecting a moving target shadow of a video SAR based on a deep learning and multi-target tracking algorithm, which includes the following specific steps:
step 1: and detecting the moving target shadow in the single-frame video SAR image by adopting a fast R-CNN network based on a characteristic pyramid structure.
Step 2: and initializing the tentative tracker by using the detection result in the step 1.
The detection results are described using the following eight-dimensional state space:
in the formula, x and y respectively represent the horizontal and vertical coordinates of the center of the target, a and h respectively represent the width-height ratio and the height of the target boundary box, and the other four variables respectively represent the change rates of x, y, a and h. The tracker uses a kalman filtering process, using parameters (x, y, a, h) to initialize the tracking state, with the corresponding velocity component set to zero.
And step 3: the tracker is updated with the associated detection results.
And the association detection and tracking are realized by Hungarian algorithm, the distributed cost matrix is composed of the intersection and parallel ratio of the detection bounding box and the tracking bounding box, and finally, the association result with the intersection and parallel ratio smaller than 0.3 is removed.
And 4, step 4: the tentative track that is continuously updated for three frames is changed to a stable track and the rest tentative tracks are deleted.
Three frames of continuous updating of tentative tracking indicate successful tracking of the target, and deletion of the rest tentative tracking can effectively reduce false alarm.
And 5: and converting the stable tracking which is not updated by three continuous frames into the candidate tracking.
If the detection result of no matching of three continuous frames of stable tracking is updated, the state is converted into candidate tracking. Indicating that the tracking target temporarily disappeared in the scene.
Step 6: and if the associated detection result updates the candidate tracking in the step 5, converting the candidate tracking into stable tracking.
And once the detection result is matched with the candidate tracking, the candidate tracking is converted into stable tracking, and the target reappears in the scene.
And 7: and if the number of times of the candidate tracks which are not updated in the step 5 exceeds a threshold value, deleting the candidate tracks.
The threshold is set to 30-40 and candidate tracks that exceed the threshold are deleted indicating that the target has completely left the scene.
Comparing fig. 2 and fig. 3, it can be seen that there are more false alarms and false-misses in the preliminary detection result of the Faster R-CNN, and the detection effect of the moving target shadow is obviously improved after the detection result is tracked by using the multi-target tracking algorithm.
To better illustrate the performance of the present invention, a total of 140 images of video SAR (of which there are 779 moving object shadows) were tested, and the statistical results are given in table 1.
Table 1140 frame video SAR test results
779 of the shadows are shared in 140 frames of images, 745 false alarms and 34 false alarms are correctly detected by the method provided by the invention. The method provided by the invention can obtain a good effect on the video SAR moving target shadow detection.
The above embodiments are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modifications made on the basis of the technical scheme according to the technical idea of the present invention fall within the protection scope of the present invention.
Claims (3)
1. A video SAR moving target detection method based on deep learning and multi-target tracking algorithm is characterized by comprising the following steps:
step 1: detecting a moving target shadow in a single-frame video SAR image by adopting a fast R-CNN network based on a characteristic pyramid structure;
step 2: initializing a tentative tracker by using the detection result in the step 1;
and step 3: associating the detection result with the trace and updating the tracker using the associated detection result;
and 4, step 4: changing tentative tracking of continuously updated three frames into stable tracking and deleting the rest tentative tracking;
and 5: converting the stable tracking which is not updated in three continuous frames into candidate tracking;
step 6: if the associated detection result updates the candidate tracking in the step 5, the candidate tracking is changed into stable tracking;
and 7: and if the number of times of the candidate tracks which are not updated in the step 5 exceeds a threshold value, deleting the candidate tracks.
2. The method for detecting the moving target of the SAR based on the video with the deep learning and multi-target tracking algorithm as claimed in claim 1, wherein the detection result in the step 2 is described by using the following eight-dimensional state space:
3. The method for detecting the moving target of the SAR based on the video through the deep learning and multi-target tracking algorithm as claimed in claim 1, wherein the association between the detection result and the tracking in the step 3 is realized through the Hungarian algorithm.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110551104.1A CN113283330A (en) | 2021-05-20 | 2021-05-20 | Video SAR moving target detection method based on deep learning and multi-target tracking algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110551104.1A CN113283330A (en) | 2021-05-20 | 2021-05-20 | Video SAR moving target detection method based on deep learning and multi-target tracking algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113283330A true CN113283330A (en) | 2021-08-20 |
Family
ID=77280247
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110551104.1A Pending CN113283330A (en) | 2021-05-20 | 2021-05-20 | Video SAR moving target detection method based on deep learning and multi-target tracking algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113283330A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114609634A (en) * | 2022-03-21 | 2022-06-10 | 电子科技大学 | Shadow-based video SAR multi-target tracking method under interactive multi-model |
CN114708257A (en) * | 2022-05-18 | 2022-07-05 | 中国科学院空天信息创新研究院 | SAR moving ship target detection method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108734103A (en) * | 2018-04-20 | 2018-11-02 | 复旦大学 | The detection of moving target and tracking in satellite video |
CN109917378A (en) * | 2018-12-26 | 2019-06-21 | 西安电子科技大学 | Utilize the VideoSAR moving target detecting method of space time correlation |
CN110390292A (en) * | 2018-10-26 | 2019-10-29 | 西安电子科技大学 | Based on the remote sensing video frequency vehicle object detecting and tracking method for dynamically associating model |
CN111127513A (en) * | 2019-12-02 | 2020-05-08 | 北京交通大学 | Multi-target tracking method |
-
2021
- 2021-05-20 CN CN202110551104.1A patent/CN113283330A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108734103A (en) * | 2018-04-20 | 2018-11-02 | 复旦大学 | The detection of moving target and tracking in satellite video |
CN110390292A (en) * | 2018-10-26 | 2019-10-29 | 西安电子科技大学 | Based on the remote sensing video frequency vehicle object detecting and tracking method for dynamically associating model |
CN109917378A (en) * | 2018-12-26 | 2019-06-21 | 西安电子科技大学 | Utilize the VideoSAR moving target detecting method of space time correlation |
CN111127513A (en) * | 2019-12-02 | 2020-05-08 | 北京交通大学 | Multi-target tracking method |
Non-Patent Citations (1)
Title |
---|
闫贺等: "基于改进快速区域卷积神经网络的视频SAR运动目标检测算法研究", 电子与信息学报, 23 December 2020 (2020-12-23), pages 615 - 622 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114609634A (en) * | 2022-03-21 | 2022-06-10 | 电子科技大学 | Shadow-based video SAR multi-target tracking method under interactive multi-model |
CN114609634B (en) * | 2022-03-21 | 2023-04-25 | 电子科技大学 | Shadow-based video SAR multi-target tracking method under interactive multi-model |
CN114708257A (en) * | 2022-05-18 | 2022-07-05 | 中国科学院空天信息创新研究院 | SAR moving ship target detection method and device |
CN114708257B (en) * | 2022-05-18 | 2022-08-16 | 中国科学院空天信息创新研究院 | SAR moving ship target detection method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106875415B (en) | Continuous and stable tracking method for small and weak moving targets in dynamic background | |
CN109934844A (en) | A kind of multi-object tracking method and system merging geospatial information | |
CN110456320B (en) | Ultra-wideband radar identity recognition method based on free space gait time sequence characteristics | |
CN112991391A (en) | Vehicle detection and tracking method based on radar signal and vision fusion | |
CN111899568B (en) | Bridge anti-collision early warning system, method and device and storage medium | |
CN113283330A (en) | Video SAR moving target detection method based on deep learning and multi-target tracking algorithm | |
CN111781608A (en) | Moving target detection method and system based on FMCW laser radar | |
CN109239702B (en) | Airport low-altitude flying bird number statistical method based on target state set | |
US20220168900A1 (en) | Visual positioning method and system based on gaussian process, and storage medium | |
CN112782664A (en) | Toilet fall detection method based on millimeter wave radar | |
Sun et al. | Fusing Kalman filter with TLD algorithm for target tracking | |
CN112634368A (en) | Method and device for generating space and OR graph model of scene target and electronic equipment | |
CN111259332B (en) | Fuzzy data association method and multi-target tracking method in clutter environment | |
Chen et al. | A graph-based track-before-detect algorithm for automotive radar target detection | |
CN113253289A (en) | Unmanned aerial vehicle detection tracking system implementation method based on combination of laser radar and vision | |
CN113687429B (en) | Device and method for determining boundary of millimeter wave radar monitoring area | |
Zhu et al. | A rain-type adaptive optical flow method and its application in tropical cyclone rainfall nowcasting | |
CN111881837B (en) | Shadow extraction-based video SAR moving target detection method | |
杜绪伟 et al. | Real-time hand tracking based on YOLOv4 model and Kalman filter | |
CN116862832A (en) | Three-dimensional live-action model-based operator positioning method | |
Shang et al. | Moving target shadow detection and tracking based on THz video-SAR | |
CN113420594A (en) | SAR image ship detection method based on improved Faster R-CNN | |
CN107561506A (en) | A kind of multistage false-alarm probability joint recursion object detection method | |
Liu et al. | An Intelligent Signal Processing Method for Motional Vital Signs Detection System Based on Deep Learning | |
Bai et al. | Pedestrian Tracking and Trajectory Analysis for Security Monitoring |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |