Body-worn Hybrid-Sensors based Motion Patterns Detection via Bag-of-features and Fuzzy Logic Optimization

Detecting the motion patterns of workers is of significance in differentiating the human’s working capabilities among several personnel’s. Similarly, encouraging the key role in society as community dealing, violence recognition, robot’s perception and virtual businesses is also important in the present world. This paper proposes a novel system for worker’s motion patterns detection (WMPD) using a state-of-the-art filter and bag-of-features. Different human actions have been recognized from the sequences of hybrid-sensors including motion and physiological sensors. A three phased filter has been proposed for calibration, error correction, and optimization of inertial data. Further, the system segments the motion patterns using a sliding window of data points and extracts co-occurring robust bag-of-features belonging to cepstral and spectral feature domains. These features precisely deal with the frequency ranges, frame-to-frame frequency flux, cepstral based coefficients, and intensity dimensions. Subsequently, Fuzzy logic using particle swarm based optimization has also been utilized to get augmented data. Finally, hidden Markov models have been proposed to characterize different basic motion patterns taken from intelligent media-wearable smart home activities (IM-WSHA) and human activities database (HAD) datasets. The experimental results show that the proposed WMPD system achieved improved accuracies of 84.5% and 86.16% for IM-WSHA and HAD datasets, respectively.