Computer Vision Based Safety Inspection in Public Housing

This paper presents using novel deep learning techniques for automated inspection of public housing common areas for detecting obstructions, blockages, and clutters. In Singapore, majority of the population resides in public housing. Although guidelines for public areas are well established, there are many cases where residents fill hallways, corridors, and common areas with clutter. This poses a risk in the case of a fire hazard or a medical emergency. Normally, safety inspections are conducted by town council and management corporations on-site to identify potential hazards and remedial actions. This manual approach is ineffective due to the low frequency of inspection. It is also a costly matter since it requires manpower. To address these challenges, we developed a computer vision-based solution for the continuous inspection of these common areas. We also employed an Autonomous Mobile Robot (AMR) equipped with camera to conduct inspections dynamically instead of using fixed surveillance cameras. We automated the inspection process by detecting the obstructions and clutters using the latest deep learning networks and reporting them to a database over the internet. This way a significant cost reduction is achieved.