We are researching technology to recognize environments not only indoors but entire cities using multimodal data such as images, sounds, and text data. For example, we are researching technology to recognize indoor and city environments and grasp the extent of damage when an earthquake occurs using commonly available cameras.
Research into indoor environment recognition technology using AI
Using image data and acoustic data on the collapse of structures, including non-structural components, during an earthquake, we will calculate the amount of movement of each object and research methods for understanding the environment and assessing damage using AI. We will use image and acoustic data to classify and quantify the amount and state of movement of objects, and develop a damage assessment method to grasp the extent of damage. By clearly indicating the damage assessment results immediately after a disaster occurs, we aim to optimize emergency response measures, minimize loss of urban functions, and enable early recovery.
Research into indoor environment simulation technology using AI
We will research methods for predicting damage before a disaster occurs by capturing images of indoor spaces using cameras and performing physical simulations based on the captured data. This system recognizes non-structural elements such as furniture and recreates their spatial arrangement in a 3D model. It automatically generates a 3D model of the furniture and uses physical simulation to recreate the furniture's movement, including tipping over. We will research detailed simulation models to prevent furniture from tipping over or moving by classifying and quantifying the amount and state of movement of objects before a disaster occurs. We aim to reduce human and functional damage by promoting measures such as securing facilities, equipment, fixtures, and furniture before a disaster occurs.
Research into AI-based environmental recognition technology for entire towns
We will research methods for determining the extent of damage in the event of a disaster from image data using general-purpose cameras that can be installed over a wide area and are widely used for surveillance purposes. We will research methods for automatically calculating the extent of damage using smoke recognition and other methods based on footage of the damage caused by the Noto Peninsula earthquake. Based on the damage information available on the cloud at the time of a disaster, we will use this information to research damage estimation technology for areas that are difficult to access and optimal planning methods for recovery and reconstruction.
Papers and Conferences Presentation
Paper presentation
Title
Laboratory
Contents
Spatial Recognition Technology for Embedded Devices - Detecting Floor Plans and Furniture from a Camera Mounted on a Home Air Conditioner -
