Deployment-Oriented Evaluation of Lightweight IMU-Based Human Activity Recognition: On-Device Efficiency and Streaming Feasibility

Abstract

On-device Human Activity Recognition (HAR) requires balancing accuracy and deployment efficiency on constrained hardware. We present Lightweight Human Activity Recognition (L-HAR), a controlled comparison of Baseline and Lightweight Temporal Convolutional Network (TCN), Transformer, and Long Short-Term Memory (LSTM) models using data sampled at 100Hz from three inertial measurement units (IMUs) worn by one participant. Evaluations covered classification, model footprint, multiply–accumulate operations (MACs), offline CPU latency, Desktop/Raspberry Pi streaming, 8-bit integer (INT8) quantization, and software-estimated Raspberry Pi energy efficiency. Lightweighting reduced parameters by up to 95.5%, model size by 94.0%, and MACs by 87.6–95.5%, with accuracy reductions of 0.8–2.2 percentage points (pp) and sub-millisecond offline latency for all Lightweight models. Baseline and Lightweight models sustained approximately 100Hz message/inference rates on Desktop, whereas no Raspberry Pi configuration reached 100Hz inference. Among Lightweight models, Transformer achieved 9.381ms End-to-End (E2E) latency and 66.925Hz inference. INT8 quantization changed accuracy and F1-score by less than 0.13 pp; Quantized TCN achieved the best Raspberry Pi streaming result (9.066ms E2E; 69.928Hz) with an estimated 49.933mJ per inference. These results demonstrate architecture- and backend-dependent deployment behavior and the need for direct target-platform evaluation.