Human Motion Reconstruction Based On Duplex Kinects

With the rapid developments and wide applications of virtual reality as well as augmented reality, human motion reconstruction which consists of human body geometry and movement information has been used in research, medical rehabilitation and business. The reconstruction of human motion based on Kinect depth camera is one of the hottest research areas. On the one hand, the real-time depth images of Kinect can be used in inferring human skeleton and acquiring human movement information. On the other hand, human body geometry can be reflected by 3D model reconstructed from depth information. This paper obtains human skeleton with movement information and individual human model with body geometry from duplex Kinects, and drives the 3D human model with skeleton to reconstruct human motion. The main contribution of the dissertation can be noted as following:(1) Kinect has a limited view angle and its inferred skeleton can be inaccurate because of self-occlusion of body parts. The employment of duplex Kinects increases the measurement range and eliminates the effect of self-occlusion to a certain extent. The interference between different Kinects can be reduced to a minimum by equipping vibrated interference eliminator.(2) Human skeleton movement optimization algorithm is proposed to improve the stability of skeleton directly inferred from Kinect. The constraint condition of bone-length preserving and motion smoothness of skeleton joint are employed in this process. A reasonable skeleton, which has constant bone-length and smooth trajectories of joints, can be acquired by using the optimization algorithm.(3) A naked individual human model is reconstructed from depth images, and human motion reconstruction is realized by driving the 3D human model with optimized skeletons. The individual human model is reconstructed to eliminate the influence of clothing shape. The driven 3D human model by optimized skeletons contains human body geometry and movement information, it can be used in many application scenarios including virtual fitting.