A Platform and Methodology Enabling Real-Time Motion Pattern Recognition on Low-Power Smart Devices
O. Sarbishei · 2019
This article presents a low-power platform, i.e., NeblinaTMsystem-on-module, with extensive hardware variants targeting Internet of Things applications. The base hardware utilizes a sensor fusion algorithm for 3D orientation tracking alongside a novel configurable framework for real-time Motion Pattern Recognition (MPR) on ARM Cortex M4F using inertial sensors. The MPR engine consists of configurable blocks performing shock-aware segmentation, histogram feature extraction, and classification using a single-hidden-layer Feedforward Neural Network. The framework can be used for human fitness/daily activity tracking or shock pattern recognition, e.g., in sports, such as tennis, golf, hockey, etc. Our platform can deliver multimodal user feedback as well. Experimental results have evaluated Neblina's MPR framework in terms of A) accuracy in human fitness activity recognition using some existing datasets, B) memory usage and latency for real-time execution on Cortex M4F and C) low power consumption for a longer lasting battery.