Human Posture Recognition using Artificial Neural Networks

This paper proposes the use of artificial neural networks(ANNs) to classify human postures, using an invasive(intrusive) approach, into 6 categories namely standing, sitting, sleeping and bending - forward and backward. Human posture recognition has numerous applications in the field of healthcare analysis like patient monitoring, lifestyle analysis, elderly care etc. Most importantly, our solution is capable of classifying the aforementioned postures in real-time, by wirelessly(Wi-Fi) acquiring and processing the sensor data on a Raspberry-Pi device with minimal lag. A data-set of 44,800 samples was collected - from 3 subjects - which was used to train and test the neural network. After experimenting and testing with a plethora of network architectures, an optimal neural network architecture(6-9-6) with suitable hyper-parameters was determined which gave an overall accuracy of 97.589%.