Gait Recognition Based on Two-scale Dynamic Bayesian Network and More Information Fusion

Gait recognition research gets rapid development as one of biometric. Now almost all the gait recognition researcher focus on gait recognition rate only in the single condition, but the gait recognition rate has rapid decline in the wearing coat and carrying bag condition. Based on analyzing the gait timing characteristics when human is moving, a novel gait recognition model that expressed two-scale dynamic Bayesian network and more information fusion is proposed. The model contains four levels of states and every time slice of the model is expressed by the fusion of large-scale information and small-scale information. This model can well express the timing characteristics of gait, that are the body posture and range of motion and other gait rhythmic changes characteristics. Experimental result show that the model recognizes gait with high rates and good robustness to the silhouette noise and lost of information and fuse the large-scale information and small-scale information well. The model can greatly reduce impact of gait recognition rate in the wearing coat and carrying bag condition.