Posture Estimation of a Human Body from Thermal Images of 2D Appearance Models of 3D Ellipsoidal Model

摘要

We propose a new model-based recognition method that involves the use of three-dimensional (3D) ellipsoidal models in various sizes and proportions as well as their two-dimensional (2D) appearance models. Most model-based vision is intended to recognize specified objects, and the model is specific to the object. However, our method can recognize various proportions of objects and was applied in posture estimation of the human body from thermal images.