Activities of Daily Living (ADL) are important indicators of both cognitive and physical well-being in healthy and ill humans. There is a range of methods to recognise ADLs, each with its own limitations. The focus of this research was on sensing location-driven activities, in which ADLs are derived from location sensed using Radio Frequency (RF, e.g., WiFi or BLE), Magnetic Field (MF) and light (e.g., Lidar) measurements in three different environments. This research discovered that different environments can have different constraints and requirements. It investigated how to improve the positioning accuracy and hence how to improve the ADL recognition accuracy. There are several challenges that need to be addressed in order to do this. First, RF location fingerprinting is affected by the heterogeneity smartphones and their orientation with respect to transmitters, increasing the location determination error. To solve this, a novel Received Signal Strength Indication (RSSI) ranking based location fingerprinting methods that use Kendall Tau Correlation Coefficient (KTCC) and Convolutional Neural Networks (CNN) are proposed to correlate a signal position to pre-defined Reference Points (RPs) or fingerprints, more accurately, The accuracy has increased by up to 25.8% when compared to using Euclidean Distance (ED) based Weighted K-Nearest Neighbours Algorithm (WKNN). Second, the use of MF measurements as fingerprints can overcome some additional RF fingerprinting challenges, as MF measurements are far more invariant to static and dynamic physical objects that affect RF transmissions. Hence, a novel fast path matching data algorithm for an MF sensor combined with an Inertial Measurement Unit (IMU) to determine direction was researched and developed. It can achieve an average of 1.72 m positioning accuracy when the user walks far fewer (5) steps. Third, a device-free or off-body novel location-driven ADL method based upon 2D Lidar was investigated. An innovative method for recognising daily activities using a Seq2Seq model to analyse location data from a low-cost rotating 2D Lidar is proposed. It provides an accuracy of 88% when recognising 17 targeted ADLs. These proposed methods in this thesis have been validated in real environments.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:766267 |
| Date | January 2018 |
| Creators | Ma, Zixiang |
| Publisher | Queen Mary, University of London |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://qmro.qmul.ac.uk/xmlui/handle/123456789/54466 |
Page generated in 0.0016 seconds