RECOGNIZING HUMAN BEHAVIOUR USING MULTISCALE CONVOLUTIONAL NEURAL NETWORKS

Abstract

Human behavior recognition is a critical field in computer vision with applications in surveillance, healthcare, and human-computer interaction. Traditional deep learning models often struggle with capturing both spatial and temporal variations in human activities. To address this challenge, we propose a Multiscale Convolutional Neural Network (MCNN) framework that efficiently extracts hierarchical features for accurate behavior recognition. The model incorporates multiscale convolutional layers to enhance feature representation across different receptive fields while integrating an improved channel attention mechanism to refine spatial and temporal dependencies. Additionally, a depth-separable convolution module reduces computational complexity, making the model more efficient for real-time applications. Experimental results on benchmark datasets demonstrate that our approach significantly improves classification accuracy while maintaining a compact model structure. The proposed MCNN framework provides a robust solution for human behavior recognition with superior performance in dynamic and complex environments.