Multiple human upper bodies detection via deep deformable part model
Aichun Zhu, Jing Jin, Tian Wang, Xili Wan, Xinjie Guan · 2017
Upper body detection is a challenging problem in practical application scenarios and shares all the difficulties of object detection. This paper focuses on the problems of multiple upper bodies detection in still images, including the diversity of appearances and a non-rigid human body. We present a new architecture for upper body detection using a Convolutional Neural Network (CNN). In this architecture, it contains the appearance model and deformable model. The appearance model is built by 8 upper body parts, and the deformable model uses a Relative Mixture Deformable Model (RMDM). RMDM is defined by each pair of connected parts to compute the relative spatial information in the graphical model. This model is compared with the state of the art on the TV Human Interaction (TVHI) dataset. The experimental results demonstrate the effectiveness of the proposed method.