论文部分内容阅读
针对红外目标跟踪过程中目标纹理信息缺乏,与背景灰度呈现强耦合性,特别是在遮挡情况下目标特征信息链断裂,特征信息无法延续的实际跟踪问题,提出了基于显著特征空间的抗遮挡跟踪算法。首先通过分析红外目标特性,利用多尺度显著性、对比度和信息熵等信息生成显著特征向量空间,结合超像素特征距离和空间距离对区域进行聚类融合,突出目标区域,生成显著图。然后融合显著区域和原图,生成多个目标候选区作为跟踪算法输入。最后通过目标的空间分布场矩阵对全局的候选区域进行匹配,同时建立遮挡检测机制,基于显著区连通区变化和特征相似度变化曲线对遮挡的起始进行判断,结合遮挡判定设置模型更新策略。在不同红外测试集上的实验结果表明:所提算法在遮挡情况下也能达到较好的跟踪效果,有效增强了跟踪算法的鲁棒性。
Aiming at the lack of target texture information in the infrared target tracking process and strong coupling with the background gray level, especially the tracking problem that the target information chain breaks down and the feature information can not be extended in the case of occlusion, an anti-block based on significant feature space Tracking algorithm. Firstly, the characteristic eigenvector space is generated by analyzing the characteristics of the infrared target, utilizing the information of multi-scale saliency, contrast and information entropy, and then combining with the super-pixel feature distance and spatial distance to cluster the regions, highlighting the target region and generating the saliency map. Then, the salient regions and the original image are fused to generate a plurality of target candidate regions as the tracking algorithm input. At last, the global candidate regions are matched through the matrix of the spatial distribution of the target, meanwhile the detection mechanism of the occlusion is established. The start of the occlusion is judged based on the change of the connectivity of the salient regions and the curve of the characteristic similarity, and the model updating strategy is set up in combination with the occlusion decision. The experimental results on different infrared test sets show that the proposed algorithm can achieve better tracking performance even under occlusion, which effectively enhances the robustness of the tracking algorithm.