Enhanced human lower-limb motion recognition using flexible sensor array and relative position image

Accurate lower-limb motion recognition is crucial for fields such as sports rehabilitation, smart healthcare, and intelligent monitoring, with wearable sensors being widely applied as a key technological approach. However, current mainstream lower-limb motion recognition methods mainly rely on single-node wearable sensors and improvements in classification models, leading to significant deficiencies in terms of data dimensionality and feature extraction capabilities. To address these issues, this paper proposes a motion recognition method that integrates a flexible sensor array and relative position images. First, a motion capture system using a flexible array sensor was developed, capable of collecting muscle activity data from 16 channels during lower-limb movements, providing rich data support for lower-limb motion recognition. Secondly, a novel relative position image generation method is proposed, which converts array data into grayscale images containing both time and spatial location information, thereby enhancing the feature expression ability of the raw data. Finally, a new network named MCRANet is designed, incorporating multi-scale cascaded residual attention module and cross-feature interaction module, significantly enhancing the feature extraction capability. Experimental results show that the proposed method performs excellently, achieving an accuracy of 97.88 % for lower-limb motion recognition, significantly outperforming existing advanced methods. © 2025 Elsevier Ltd