Daily life Log Recognition based on Automatic Features for Health care Physical Exercise via IMU Sensors

Wearable inertial based sensors are strong enablers for the acquisition of human daily life-log data. Eventually, many motion devices have often degraded the performance of wearable sensors due to inner/outer environmental effects. In addition, key decisions are made based on human life-log recognition results and precise recognition of human life-logs with lower limits of uncertainty is significantly important. For this purpose, many motion devices have been used in last decade, in order to recognize daily life activities. In this paper, we proposed an efficient model for better recognition results for healthcare patient's daily life-log patterns. We designed a 1D Haar based extraction algorithm and different statistical features to extract valuable features. For activity classification, we used Quadratic Discrimination Analysis (QDA) optimized by Artificial Neural Network (ANN) on two benchmarks PAMAP2 dataset and our self-annotated IM-SB database. The outcome of our system illustrates that our proposed model competes with other advanced methods in term of exactness and effectiveness.