Pose Estimation and Tracking of Eating Persons in Real-life Settings
mag(2011)
摘要
We present an approach to estimate and track 2D upper body poses of persons who are having a meal in videos with highly challenging uncontrolled imaging conditions. We employ a probabilistic model that represents the body as a kinematic tree, and perform inference in this kinematic tree model using particle ltering, and also estimates self-occlusions. Our approach is evaluated with 7 di erent videos, in which di erent persons with di erent types of clothing, orientation can be seen, su ering from various degrees of motion blur. The evaluation shows that our approach can deal with complex data, and successfully tracks upper bodies with a lot of self-occlusions.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络