LMSPNet: Improved Lightweight Network for Multi-Person Sitting Posture Recognition

Abstract

Incorrect sitting posture may lead to health problems. Therefore, effective sitting posture recognition can remind individuals to maintain correct sitting posture and reduce discomfort. Traditional methods for sitting posture recognition have limitations in terms of high cost and slow inference speed. To address these issues, we propose a novel model called LMSPNet for multi-person sitting posture recognition. This model first employs the Light Convolution Core (LCC) to reduce the complexity of the model and then introduces Convolutional Block Attention Module (CBAM) to adaptively adjust the receptive field in the neural network to capture global contextual information, thereby enabling the model to better learn relationships between different channels. We construct the first human sitting posture dataset to evaluate the performance of LMSPNet. Experimental results demonstrate that, compared to the baseline models, our LMSPNet achieves state-of-the-art results with an accuracy of 99.57%. Therefore, our model is expected to become a powerful tool for multi-person sitting posture recognition.