基于改进SMOG模型的目标跟踪算法

Object Tracking with Revised SMOG Model

空间颜色混合高斯模型(SMOG)是一种优于经典颜色直方图的目标颜色表示模型。然而,SMOG模型初始化时不可避免的会引入背景象素,且极有可能被误选为一个有效的目标分布,严重影响后续的相似性度量,且各目标分布鉴别性能的发挥会受到自身权重的严重制约,当背景中出现与目标颜色相似的干扰物体时,算法的跟踪准确性仍会有所降低。针对这些不足,提出了一种改进的SMOG模型,通过衡量背景与目标分布在空间颜色域上的联合距离来剔除误引入的背景分布,同时将联合距离作为目标分布鉴别能力高低的一种度量,引人到相似性度量函数中,并在跟踪过程中根据局部背景的变化动态的进行更新,充分根据每个高斯分布的鉴别性能调整其匹配权重。实验证明,改进后的SMOG模型能有效提高目标跟踪的鲁棒性。

Spatial-color Mixture of Gaussians （SMOG） is a new object appearance model which has been proven to be superior to classic color histogram appearance model. However,in the initialization of SMOG,some background pixels are inevitably introduced and more likely selected as an object mode,which often affects the performance of similarity measure,and the weight of each Gaussian mode is restricted by the probability of the pixels belonging to it,when something with similar color appears in the vicinity of the object,the performance of SMOG based tracking algorithm often degenerates. A revised SMOG model was proposed,which could effectively recognize and remove the introduced background modes by calculating the spatial-color joint distance between each Gaussian mode and the object local background,it also considers the joint distance as a confidence of the discriminative power of each mode,and introduces the confidence into the similarity measure function,and dynamically updates these confidences in the tracking process based on changing background,then modifies the weight of each Gaussian mode in similarity measure based on its confidence. Experiment results show that the revised SMOG model can efficiently improve the robustness of tracking.