Privacy-preserving indoor monitoring using vision sensors

Abstract

Indoor monitoring is the process of continuous observation and analysis of indoor spaces using sensors, cameras, and similar sensing deployments. These systems typically monitor: environmental parameters like temperature, air quality; disaster scenarios like fire, earthquakes; and human behaviour such as activities of daily living, unnatural and abnormal activities, occupancy of spaces, indoor localization, and marked behavioural changes. Indoor monitoring has a wide range of applications in elderly-care, health-care, and in systems facilitating smart-homes. The monitoring devices used in indoor monitoring systems largely comprise ambient sensors, wearable sensors (useful mostly in human monitoring) and vision sensors (typically cameras). Ambient and wearable sensors for indoor monitoring have limited effectiveness and are constrained largely in terms of performance and convenience respectively. These constraints can largely be overcome by the use of vision sensors largely comprising visible cameras. The issue with visible cameras, however, is compromise of privacy in such indoor spaces. The main objective of this thesis, therefore, is to devise privacy-preserving mechanisms for indoor monitoring utilizing vision sensors. These mechanisms harness the benefits of vision sensors in indoor monitoring whilst overcoming the limitations of privacy preservation. The thesis also conceives the use of appropriate learning algorithms for analyzing privacy-preserving visual data to enable automated monitoring.