Privacy-preserving human behaviour monitoring through thermal vision

Tools

Naser, A.F.N. ORCID: 0000-0001-5969-1756, 2022. Privacy-preserving human behaviour monitoring through thermal vision. PhD, Nottingham Trent University.

Preview

Text
Abdallah Naser 2022.pdf - Published version
Download (12MB) | Preview

Abstract

Despite the abundance of human-centred research to support domestic human behaviour monitoring in various vital applications, there are still notable limitations to deploying such systems on a broader scale. The main challenge is the trade-off between privacy, performance, and cost of assistive technologies to support older adults to live independently in their own homes. For example, the traditional vision-based sensing approach provides excellent performance while violating human privacy in domestic environments. In contrast, the ambient sensing approach, e.g., employing Passive Infra-Red (PIR) sensors, maintains human privacy but suffers significant performance hindrances in realistic scenarios such as multi-occupancy environments.

This research proposes to utilise the Thermal Sensor Array (TSA) to adjust the trade-off between privacy and performance in domestic environment applications. The rationale behind proposing this sensor for human behaviour monitoring applications is its claimed advantages to perform well while maintaining human privacy, low-cost, and noncontact capabilities. Nevertheless, there has not been sufficient related work to empirically validate the hypothesis of using this low-resolution imager in domestic monitoring. Furthermore, most published works that use the TSA have not yet reached the deployment stage due to the TSA sensing constraints. In particular, TSA is sensitive to environmental thermal noise, and its Field of View (FoV) is not wide enough to cover a large inspection area. Intelligent algorithms should be employed in order to avoid these limitations.

The focus of this thesis is to investigate the human physiological and behavioural thermal patterns for privacy-preserving human behaviour monitoring to support the independent living of older adults in a multi-occupancy environment by using TSA. This will be achieved through signal processing and machine learning techniques. To achieve this aim, the research methodology is drawn into two main directions. First, human physiological processing of the human thermal signal. Second, human behavioural processing of the human motion signal. This drawn methodology resulted in four main novel contributions.

The first novel contribution of this research is to propose an adaptive segmentation of the human physiological presence and count the number of people from different sensor placements, indoor environments, and human-to-sensor distance. The second contribution is to extract localisation knowledge of the human physiological appearance in terms of human-to-sensor distance and human-to-human distance. Extracting human localisation knowledge is also applicable in other applications such as caregivers and care time monitoring. The third contribution is to fuse multiple TSAs to cover a wide inspection area, e.g., private or care homes. Hence, objects that appear in the low-resolution thermal images acquired from TSA have low intra-class variations and high inter-class similarities, making the identification of the overlapping regions through matching a comparable template image in multiple images very difficult. This research proposes a motion-based approach to fuse multiple TSAs and learn the domestic environment layout with a privacy improvement of utilising TSA in potential monitoring applications running in the cloud. Inspired by the results from this stage of the research, the fourth contribution of the research presented in this thesis is a human-in-the-loop fall detection approach in the Activities of Daily Living (ADLs) that reduces the false-positive alerts while keeping the false-negative fall predictions as low as possible. The novel solutions and the results presented in this thesis demonstrate a significant contribution toward enabling privacy-preserving human behaviour monitoring.

Item Type:	Thesis
Creators:	Naser, A.F.N.
Date:	April 2022
Divisions:	Schools > School of Science and Technology
Record created by:	Linda Sullivan
Date Added:	15 Nov 2022 12:22
Last Modified:	15 Nov 2022 12:23
URI:	https://irep.ntu.ac.uk/id/eprint/47409