Real time human activity recognition from accelerometer data using convolutional neural networks

The study of human regular tasks have become more prevalent and accessible as a result of the widespread use of different sensors integrated into mobile devices. This issue now exists in different vast real-world area applications some examples are “healthcare monitoring, fitness tracking, and user-adaptive systems”. Where a generic model capable of recognizing an arbitrary user's activity in real-time is necessary. We present in this paper, a user-independent deep learning technique for digital human activity categorization. We advocate using CNN in conjunction with basic statistical characteristics that keep info regarding the global shape of different time series for local feature extraction. We also look at how the duration of a time series affects recognition accuracy, restricting it to a single second to allow for time series analysis activity classification.