Human Activity Recognition with Dual-Stream Residual Network based on Wearable Devices

Human activity recognition (HAR) is utilized to monitor human postures and actions in various fields such as human-computer interaction, motion detection and analysis, and medical rehabilitation. Currently, there is a growing interest in human activity recognition using wearable devices. Traditional convolutional neural networks (CNNs) typically use one-dimensional convolution for time-series signals, which makes it difficult to fully exploit high-dimensional features. This paper proposes a dual-stream residual network model for human activity recognition based on Gramian Angular Field (GAF). GAF converts a one-dimensional time-series signal into a two-dimensional image, enabling better observation of weak amplitude changes in the signal and providing better interpretability than traditional conversion algorithms. The experimental com-parison between the proposed model and several mainstream depth models on a local dataset shows that the proposed model outperforms all other models in all evaluation indicators.