多特征提取逐步求精的高速移动目标跟踪算法

Multi-feature Extraction and Stepwise Refinement Based High-speed Moving Target Tracking Algorithm

传统的高速移动目标跟踪通常使用图像特征描述,不能够根据跟踪场景自适应地选择最优跟踪特征,导致功能模板很容易产生漂移问题.为此,提出一种基于特征融合和逐步求精的高速移动目标跟踪算法.该算法主要包括3个阶段:第1阶段为自适应多特征融合阶段,通过计算跟踪目标每一特征的前景及背景的区分度,获取目标特征的融合模型;第2阶段是基于多特征内核跟踪阶段,在Mean-Shift框架下,引入Epanechnikov函数作为内核函数提升目标区域中心的像素权重比值;第3阶段为目标模型的自适应更新,通过设计一种模板更新策略提高跟踪结果的准确度.仿真实验结果表明,该算法适用于高速目标跟踪.

The traditional high-speed moving target tracking schemes can only use image features description constantly, but not adaptively choose the optimal tracking feature according to tracking scenes, which leads to the feature template drift easily. To solve the above problem, a new tracking algorithm of high-speed moving targets based on multi-feature fusion and stepwise refinement is presented. This algorithm consists of three stages, the first stage is the adaptive multi feature fusion stage, through the calculation of target tracking of foreground and background for each feature discrimination, fusion model acquisition target feature; the second stage is the feature tracking based on kernel stage, in the Mean-Shift framework, using Epanechnikov function as the kernel function of pixel weight lifting the ratio of target area center; adaptive updates for the third phase of the target model, through the design of a template updating strategy to improve the accuracy of tracking results. The simulation results show that the proposed algorithm is suitable to track a high-speed target.