Work

Intelligent Fine-Grained Daily Living Locomotion Prediction Based on Skeleton Modeling and CNN

Year: 2024

Type: article

Abstract: Activities of daily living are important for human locomotion prediction.Humans perform several activities while moving from one place to another.Some of these activities are coarse-grained while others are fine-grained.However, the challenges remain for efficient locomotion prediction for daily living activities.A couple of important challenges faced are modeling of the human skeleton along with extraction of the relevant feature.Therefore, this research focuses on locomotion prediction while performing daily tasks.For this, a benchmarked publicly available dataset known as Opportunity++ has been considered to conduct experiments.The paper proposed a system that converted the videos from dataset into sequences of frames and filtered them.Next, background of each frame is removed using background frame subtraction technique.Then, human is detected via the skeleton modeling method based on five skeleton points.Furthermore, stochastic features have been extracted from skeleton model.Using the quadratic discriminant analysis method, these extracted features are then optimized due to the large vector size.Lastly, locomotion detected has been validated via convolutional neural network.Moreover, the experimental results show that the proposed methodology has achieved improvement in terms of accuracy rate of 82.94%.Compared to other state-of-the-art systems in literature, the proposed system outperformed.

Source: Traitement du signal

DOI: 10.18280/ts.410525