基于相关滤波的抗遮挡单目标跟踪方法研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）24-0086-06

Abstract：Aiming atthe problem that traditionalsingle-target tracking algorithms tend tofaildue to targetocclusion incomplexenvironments，animprovedanti-occlusionsingletarget trackingmethodisdesigned.Thealgorithmreplaces the singleHOG featureintheoriginalKCFalgorithmby fusing HOG features and CNfeaturesat thefeature level toachieve complementary detectionof the tracked target in terms ofshape and color. The Average Peak Corelation Energy （APCE） is introducedtojudgethetargetolusionsituation，andthelearningrateisdynamicallyadjustedbysettingathresholdand the Kalman filteralgorithm is fused topredictthe targetposition，which enhances theoccusionresistanceofthealgorithm. Experiments on the dataset show that the improved algorithm improves the accuracy by 7.3% and the success rate by 15.2% especially inoccusionscenarioswhere itexhibitssignificantperformanceenhancement.This methodissuitableforsourceconstrained smallplatforms， thus meeting the target tracking requirements of more application scenarios.

Keywords：single target tracking; feature fusion;Kalman;anti-occlusio1

0 引言

随着人工智能技术在各类无人平台上的广泛应用，目标跟踪技术作为其基础且主要的研究方向之一，吸引了国内外众多研究人员的高度关注，在安全监控、智能交通、无人机航拍等多个领域展现出重要的应用价值。（剩余7840字）