Frontal motion tracking based on image features analysis and particle filter

This work presents a method that integrates the image feature analysis into standard particle filter for monocular frontal human body motion tracking. The scaled prismatic models are taken as human body models and a state vector is used to represent human body pose. The image feature analysis applies the trained back propagation neural networks to locate some key joints such as elbow joints and knee joints with high precision. Unlike the standard particle filter, the state vector can be partly inferred from the key joints obtained by the image feature analysis in the proposed method. Thus, it reduces the number of sampled particles required by the standard particle filter. The performance analysis shows that this algorithm outperforms the standard particle filter since it reduces computation load and increases robustness.