摘要
为提高视觉跟踪中目标模型的鲁棒性,提出一种基于稀疏加权的局部判别跟踪方法,在贝叶斯推论框架下进行目标跟踪。利用多个局部判别稀疏模型表示目标,根据每个局部模型在表达目标表观时的重要程度分配权重,将目标建模为多个局部模型的加权组合以减弱表观变化对模型的影响。在跟踪中选择与目标模型最相似的候选区域作为跟踪结果,通过遮挡检测减轻遮挡对跟踪的影响,并对目标模型进行在线更新以避免漂移。实验结果表明,该方法能在目标表观发生变化的情况下保证跟踪鲁棒性。
In order to enhance the robustness of the target model in visual tracking, a local discriminant tracking method based on sparse weighing is proposed within Bayesian inference framework. The target is represented as a combination of multiple local discriminative sparse models. The weight value is assigned for each local model according to its significance of expressing the target. The target is modeled as a weighted combination of local models to alleviate the influence of appearance changes and improve the robustness of the model. During tracking, the candidate which is the most similar to the target model is chosen as the tracking result, and occlusion detection is added to reduce the influence of occlusion. Besides, the target model is updated automatically to avoid drifting. Experimental results show that the proposed method can maintain a robust tracking when the target appearance changes.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第9期226-234,共9页
Computer Engineering
关键词
视觉跟踪
目标建模
表观变化
模型更新
稀疏表示
贝叶斯推论
visual tracking
target modeling
appearance change
model update
sparse representation
Bayesian inference
