CN-121982068-A - Cross-satellite vehicle target tracking method combining road network and target information

CN121982068ACN 121982068 ACN121982068 ACN 121982068ACN-121982068-A

Abstract

The invention discloses a cross-satellite vehicle target tracking method combining road network and target information, which comprises the following steps of S1, performing relay earth observation on a target area by utilizing a low-orbit large-scale remote sensing satellite constellation to obtain a satellite image sequence, S2, selecting a vehicle target from satellite images at an initial moment, taking a target slice as the start of target tracking, S3, transforming road network data from geographic coordinates to image coordinates of the satellite images according to the mapping relation of the satellite image object space to the image space, predicting the motion range of the vehicle target at the current moment by utilizing road network information, S4, matching the target slice with all image blocks to be matched at the current moment by utilizing a deep learning target tracking network, selecting a regression area with the maximum confidence degree as the tracking result of the vehicle target at the current moment, and optimizing the vehicle target tracking result at different moments of adjacent satellites by combining the road network information to finally generate a vehicle target track.

Inventors

LIU YONG
CHEN JIAJIAN
GUO PENGYU
CAO LU
MENG LING

Assignees

中国人民解放军军事科学院国防科技创新研究院

Dates

Publication Date: 20260505
Application Date: 20260402

Claims (10)

1. A method for tracking a target of a vehicle across a satellite by combining road network and target information, the method comprising the steps of: S1, performing relay earth observation on a target area by using a low-orbit large-scale remote sensing satellite constellation to obtain a satellite image sequence; s2, selecting a vehicle target from satellite images at an initial moment, taking a target slice as the start of target tracking; S3, transforming road network data from geographic coordinates to image side coordinates of the satellite image according to the mapping relation from the satellite image object side to the image side, and predicting the movement range of the vehicle target at the current moment by utilizing road network information; And S4, matching the target slice with all the image blocks to be matched at the current moment by utilizing a deep learning target tracking network, selecting a regression area with the maximum confidence as a tracking result of the vehicle target at the current moment, and optimizing the vehicle target tracking results of adjacent satellites at different observation moments by combining road network information to finally generate a vehicle target track.
2. The method of claim 1, wherein the deep learning target tracking network is a twin zone nomadic network SiamRPN.
3. The method for tracking a vehicle target across satellites by combining road network and target information according to claim 1 wherein the relay earth observation means that after a previous satellite observes the region, a next satellite continues to observe the region.
4. The method for tracking a vehicle target across satellites by combining road network and target information according to claim 1, wherein in step S1, roads in a satellite imaging area are modeled, and road network data of a corresponding area is constructed by using existing road network data or detecting roads by using satellite images.
5. The method for tracking the vehicle target across satellites by combining the road network and the target information according to claim 1, wherein in step S2, the deep learning target tracking network is trained in advance, random rotation within ±180°, noise increase and contrast change modes are applied to training samples of the vehicle target, and generalization capability of the network for different movement directions of the same vehicle target at different moments is enhanced.
6. The method for tracking the vehicle target across satellites by combining the road network and the target information according to claim 1, wherein in step S3, the movement range of the vehicle target at the current moment is predicted by using the road network information, and the road area image block of the vehicle target is obtained by using the road network data within the movement range of the vehicle target.
7. The method for tracking the vehicle target across satellites by combining the road network and the target information according to claim 6, wherein the prediction of the motion range of the vehicle target at the current moment is performed by using the road network information, and the specific flow is as follows: S3.1, calculating an imaging time interval according to the imaging time of the satellite images at the previous moment and the current moment; S3.2, obtaining a maximum movement range of the vehicle target based on the maximum movement speed of the vehicle target and the movement direction at the last moment; S3.2, sliding window blocking is carried out on the satellite image at the current moment along the road network information in the constraint range, the center of the image block is a point on the road center line, and meanwhile, in order to prevent the vehicle target from being segmented by the adjacent image blocks, a certain overlapping area is reserved among the image blocks, so that the target tracking problem is converted into the image matching problem between the target slice and the plurality of image blocks at the current moment.
8. The method for tracking the vehicle target across satellites by combining road network and target information according to claim 1, wherein in step S4, after the tracking result of the vehicle target at the current moment is obtained, the motion track of the vehicle target is further optimized and used as guiding information for planning the task of the following observation satellite.
9. The method for tracking a vehicle target across satellites by combining road network and target information according to claim 8, wherein the specific flow for optimizing the motion trail of the vehicle target comprises: S4.1, screening out the shortest communicated road net piece section based on the vehicle target positions at the previous moment and the current moment; and S4.2, carrying out interpolation optimization processing on the vehicle target track between adjacent moments on each road based on the screened road network segments, and finally generating the vehicle target track which is more in line with the actual road condition.
10. The method for tracking a target of a vehicle across a satellite combining road network and target information according to claim 7, wherein in step S3.2, the window size is set to 2 times the size of the target slice.

Description

Cross-satellite vehicle target tracking method combining road network and target information Technical Field The invention relates to the field of satellite image processing, in particular to a cross-satellite vehicle target tracking method combining road network and target information, which realizes continuity, robustness and accuracy of vehicle target tracking in a complex environment. Background The satellite-based ground vehicle target wide-area monitoring has high application value. The prior art mainly focuses on researches such as vehicle target detection based on a single satellite image, vehicle target tracking based on a single optical video satellite and the like, and can only extract vehicle target static information or has shorter tracking time (generally tens of seconds), and long-time monitoring cannot be performed. In order to overcome the defect of the monitoring capability of a single satellite, relay observation can be carried out on the region of interest through a low-orbit remote sensing satellite constellation, and vehicle target tracking is carried out by utilizing a satellite image sequence obtained by high-frequency revisit observation, so that long-time and wide-area continuous monitoring on a vehicle target is realized. With the rapid development of low-orbit satellite constellation technology, the number of low-orbit remote sensing satellites is increased, and a large-scale deployment trend is presented, so that the tracking of the targets of the vehicle by the satellites based on different satellite images is possible, but no relevant solution to the problem of tracking the targets of the vehicle by the satellites exists at present. Compared with the target tracking based on satellite video, in the target tracking across satellites, the interval between adjacent satellite images is longer, the minimum interval can reach the order of minutes, the frame frequency of the satellite video can reach 10 frames per second, and the two frames are different by 1-2 orders of magnitude. Meanwhile, the observation side-sway angles of different satellites are different and discontinuous, and certain geometrical and radiation differences exist among different satellite images. The vehicle target has the characteristics of strong maneuverability, small size and the like, and has large movement displacement and few pixels in the satellite image, so that the tracking of the vehicle target crossing the satellite is very difficult. Disclosure of Invention Aiming at the problems existing in the prior art, the invention aims to provide a cross-satellite vehicle target tracking method combining road network and target information, which is used for carrying out relay observation on a region of interest through a low-orbit satellite constellation to obtain a high-frequency satellite image sequence, and combining characteristics such as road network information, target motion, attribute and the like as constraint matching conditions to realize vehicle target matching and tracking under a complex scene. In order to achieve the above object, the present invention provides a method for tracking a target of a vehicle across satellites by combining road network and target information, the method comprising the steps of: S1, performing relay earth observation on a target area by using a low-orbit large-scale remote sensing satellite constellation to obtain a satellite image sequence; s2, selecting a vehicle target from satellite images at an initial moment, taking a target slice as the start of target tracking; S3, transforming road network data from geographic coordinates to image side coordinates of the satellite image according to the mapping relation from the satellite image object side to the image side, and predicting the movement range of the vehicle target at the current moment by utilizing road network information; And S4, matching the target slice with all the image blocks to be matched at the current moment by utilizing a deep learning target tracking network, selecting a regression area with the maximum confidence as a tracking result of the vehicle target at the current moment, and optimizing the vehicle target tracking results of adjacent satellites at different observation moments by combining road network information to finally generate a vehicle target track. Further, the deep learning target tracking network is a twin zone nomadic network SiamRPN. Further, relay earth observation means that after the previous satellite observes the area, the next satellite continues to observe the area. Further, in step S1, the road in the satellite imaging area is modeled, and road network data of the corresponding area is constructed by using the existing road network data or detecting the road by using the satellite image. Further, in step S2, the deep learning target tracking network is trained in advance, and random rotation within ±180°, noise increase, contrast change and other modes are applied to