Tracking Generic Human Motion via Fusion of Low- and High-Dimensional Approaches

Tracking generic human motion is significantly challenging because of the high-dimensional state space as well as various motion types. In order to deal with the challenges, we propose a fusion formulation to integrate the low-and high-dimensional tracking approaches into one framework. The low-dimensional approach successfully overcomes the high-dimensional problem on tracking the motions with available training data by learning motion models. On the other hand, the high-dimensional approach is employed to recover the motions without learned models by sampling directly in the pose space. Within the framework, the two parallel approaches are fused by a set of criteria at each time step. The fusion criteria ensure that the overall performance of the system is improved by concentrating the advantages of the two approaches and avoiding their weak points. Experimental results with qualitative and quantitative comparisons demonstrated that the proposed formulation can fully display the advantages of different algorithms and effectively track generic human motion.Computer Science, Artificial IntelligenceCPCI-S(ISTP)