Thursday, July 7, 2016 at 10:00 AM in Rice 504
Md Abu Sayeed Mondol
Advisor: John Stankovic
Committee Members: Alf Weaver (Committee Chair), Jack Davidson, Yanjun Qi and John Lach (ECE, Minor Representative).
Title: Toward Robust and Accurate Human Activity Recognition using Wearable Sensors
ABSTRACT: Human activity recognition is very important in many areas including healthcare, safety, behavior monitoring, energy management and manufacturing. Wearable devices enriched with sensors like accelerometers, gyroscopes and magnetometers can be used in recognizing wide range of human activities. However, activity recognition using these devices is challenging due to issues like confounding gestures present in different activities, diversity in performing the same activity, and limited resources of the wearable devices. Today, many solutions for wearables are focused on some particular activities, and they do not generalize to other activities. One challenge is to develop underlying algorithmic solutions for activity recognition that can be used in many different wearable based applications. We propose new directions for creating such basic results that we are calling (i) Direction Agnostic Modeling, (ii) Direction Aware Modeling, (iii) Orientation Reachability, (iv) Spatiotemporal Segmentation, and (vi) Dynamic Space Time Warping for Device Orientation. Proposed techniques are based on the orientations of the wearable devices where the orientations provide pivotal information regarding different human activities. We hypothesize that highly robust and accurate activity recognition models can be developed by integrating the proposed fundamental techniques with the state of the art. The proposed research aims at finding solutions for effective integration of the techniques toward robust and accurate human activity recognition, particularly for realistic settings.