Mobile crowd sensing by Android phones - I

Abstract

Recent advancement in technology have given rise to a new sensing paradigm, Mobile-Crowd Sensing (MCS), that taps on the multiple sensors available on a smartphone to retrieve information, and this has led to the use of smartphones for human activity recognition. This project taps on this sensing paradigm and applies it to the area of queuing analytics, and looks at the feasibility and challenges of this approach. The developed application specifically looks into the detection of a more complex queuing pattern, namely the Number Queue, using a hierarchical classification framework and the linear accelerometer and gravity sensor. In the framework, the application first looks to identify basic physical activities such as the likes of walking and sitting, by partitioning incoming sensor data into micro-activities of two seconds. These sequences of micro-activities are then partitioned into larger frames of 30 seconds, identified as higher-level activities, and smoothed to eliminate erroneous readings. Finally, the queue detection is performed based on the smoothed set of higher-level activities, by actively looking for the signature of a queuing pattern in real-time. Based on 30 tests conducted, the application was able to detect number queue patterns with lesser variability, with close to 85% accuracy. The approach of using MCS for queuing analytics provides a good foundation and can be further improved on by extending it to other queue patterns, or a combination with other approaches for a more robust application.