Unmanned plane target tracking method combining mean-shift algorithm and particle-filter algorithm

Abstract

The invention provides an unmanned plane target tracking method combining a mean-shift algorithm and a particle-filter algorithm. The unmanned plane target tracking method comprises the following steps of: constructing a mean-shift tracking algorithm based on a bandwidth matrix according to the bandwidth matrix during a mean-shift tracking process, and updating a target-scale window in a self-adaptive manner during the tracking process; establishing a weighting and data-fusion target positioning method according to detected results of the mean-shift tracking algorithm and the particle-filter algorithm; determining an unmanned plane target position according to the weighting and data-fusion target positioning method; sampling particles in the particle-filter algorithm according to a target re-convergence method to generate the particle-filter algorithm based on the target re-convergence; and obtaining a target expanding and searching strategy according to the target re-convergence particle-filter algorithm and tracking the target. The embodiment of the invention can realize real-time positioning and tracking of the target under complex conditions including a dynamic scene, illumination change, scale change, shielding and the like, and therefore, the unmanned plane target tracking method has the advantages of being good in real-time performance, strong in adaptability, good in expandability and the like.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to an unmanned aerial vehicle target tracking method combining mean shift and particle filter. Background technique [0002] Object tracking has important research value in science and engineering. During the flight of the UAV tracking the target, due to the relative motion between the camera and the target, the application scene is complex and changeable, and the collected video images generally have obvious changes in illumination, obvious debris or noise in the image, and the target is partially blocked. Or complete occlusion, large target pose changes and other characteristics make it difficult to achieve target tracking based on sequence images. [0003] Target tracking algorithms can be divided into deterministic tracking algorithms and random tracking algorithms. The mean shifting algorithm is a deterministic tracking algorithm. This tracking algorithm can usually be tran...

AI Technical Summary

Problems solved by technology

The target tracking algorithm based on mean moving is simple and has good real-time performance, but it is easy to converge to the local extreme point, and the tracking window cannot be adaptively adjusted. When the target is highly mobile, the scale changes significantly, and there are different degrees of occlusion or illumination. Poor tracking when strong changes occur

[0004] The particle filter tracking algorithm is a random tracking algorithm. It uses multiple particles to effectively express the uncertainty of tracking. It shows strong robustness for tracking non-rigid objects and tracking under partial occlusion. There is particle degeneration phenomenon, the prediction accuracy is affected by the effect of cumulative error, and the calculation amount is relatively large, and the real-time performance is poor

Images