基于动态形状的红外目标提取算法

A novel dynamic shape based infrared object subtraction method

提出了一种基于动态形状的红外目标提取新算法。首先提出基于先验知识的动态形状概念以描述形变的时空相关性,用logistic主成分分析(PCA)将形状映射到低维度的潜变量子空间,并对潜变量变化趋势的建立自回归模型,通过模型预测指导分割过程。然后在马尔可夫随机场(MRF)的框架下将先验形状约束定义为从分割结果到先验形状间轮廓变换的距离,并结合目标的检测似然度信息统一为随机场势函数,再用GraphCut获取全局最优解分割。试验表明,该算法能够适应目标的外形变化,并能在目标被部分遮挡或难以从背景区分的情况下准确提取,鲁棒性较好,能够为机器视觉应用提供中级视觉的信息。

This paper introduces a novel segmentation method for deformable objects in monocular infrared videos.First,the concept of dynamic shape is proposed to implicitly represent spatial-temporal correlation.The logistic principal component analysis optimizes the mapping from high dimensional shape space to latent subspace,where an autoregressive model is established to predict the deformation.Then the shape prior term is formulated as a distance of deformation by explicitly estimating integral motion of the whol...