Remote sensing video vehicle target detection and tracking method based on dynamic association model

Abstract

The invention discloses a remote sensing video vehicle target detection and tracking method based on a dynamic association model, and solves the problems of low tracking precision, poor stability andinflexible algorithm. The method comprises the following steps: intercepting an image frame by frame, performing moving target detection on a first frame image, and creating a storage space storage target; and for subsequent frame images, detecting candidate moving targets, selecting historical target estimation positions from the storage space and matched with the candidate moving targets, then updating historical target states, arranging the storage space and storing newly-appearing targets. Interference outside a road area is filtered by using a road mask, targets are flexibly added and deleted by using dynamic association, the state estimation of a disappearing moving target is optimized by using a group effect, and the tracking precision is improved by using a trajectory optimizationmethod. Simulation experiments also prove that the method reduces the calculation amount, improves the tracking precision and stability, and is used in the fields of traffic flow monitoring, driving route analysis and military intelligence acquisition.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to moving target detection and multiple moving target state prediction and matching, in particular to a remote sensing video vehicle target detection and tracking method based on a dynamic correlation model, which is used for detecting vehicle targets in remote sensing video with tracking. Background technique [0002] Remote sensing video satellite is a new type of earth observation satellite. Its biggest feature is that it can continuously observe a certain target area by "staring" and store it in the form of video, from which more time-space related information can be obtained. Ground object detection and tracking provides a new opportunity. Satellite video imaging provides important data support for remote sensing and earth observation. How to use satellite remote sensing video to realize the intelligent detection and tracking of important targets is an important resear...

AI Technical Summary

Problems solved by technology

However, due to the regularity of the target driving state of the vehicle, the target motion state changes greatly at intersections, overpass entrances, etc., and it is difficult for the Kalman filter to handle this change well.

The traditional method often judges the disappearance of the target at the boundary. This method conforms to common sense, and the amount of calculation is relatively small, which simplifies the algorithm process. However, when the tracker loses the target during the actual tracking process, the algorithm will not reset the lost will not delete trackers that have lost targets from the storage data, which will reduce the calculation speed of the algorithm and cause unreasonable use of storage resources

[0006] In practical applications, for remote sensing video moving vehicle tracking, it is difficult to obtain the accurate target position of the first frame, and because of the low resolution of the remote sensing video image and more noise interference, the target position is only detected in the first frame as the initial position of subsequent tracking, which is accurate low rate

When the vehicle is driving under the overpass, shaded by trees, etc., the vehicle target is within the monitoring range but cannot be detected in the first frame, and it will not be tracked using traditional methods

In the traditional method, the Kalman filter method is used to predict the position of the target, but the Kalman filter method is difficult to adapt to the situation where the target's motion state changes greatly, which will result in reduced tracking accuracy and poor track smoothness.

At the same time, the traditional method only judges the disappearance of the target at the boundary. If the target is lost, the tracker of the target will not be deleted, which will also cause a waste of memory.

Images