Human Motion Trajectory Perception and Recognition Based on Multi Sensor Network Fusion with Conditional Random Fields

In recent years, effective modeling and capturing realistic human motion in 3Dimensional (3D) environments was crucial for applications including game character animation, virtual reality and human-robot interaction. Traditional approaches for human motion recognition had faced several challenges which include limited generalizability, noise and occlusion, variability in motion patterns. Therefore, this research proposes Multi Sensor Network Fusion with Conditional Random Fields (MSNFCRF) to recognize human motion trajectory perception. Initially, the data is taken from Human3.6M dataset which consists of human pose data and motion capture data. Then, this data is preprocessed by using Empirical Mode Decomposition (EMD) which effectively reduced noise and dimensionality. After that, the features are extracted by using Maximum Lyapunor Exponent (MLE) which extracted dynamic and chaotic behavior of human motions and identifies subtle patterns and variations in the motion data. Finally, recognition is done by using proposed MSNFCRF which enhanced situational awareness and better handling of complex scenarios and effectively integrates information from multiple scales and modalities by enabling it to capture subtle patterns and relationships in the data. The proposed MSNFCRF achieved better results of accuracy (98.91 %), precision (98.83 %), recall (98.82%) and F1-score (98.90%) when compared with existing Inception-Long Short-Term Memory.