A group from Tohoku University and Shinshu University has succeeded in developing an image recognition system that supports the discovery of objects from images taken by remote-controlled robots used in search activities.It was being promoted as part of joint research and development in the "ImPACT Tough Robotics Challenge" led by the Council for Science, Technology and Innovation of the Cabinet Office, and can be installed in various types of robots working in disaster environments.
Part of this achievement was presented at the 2016th Robotics Society of Japan Academic Lecture held on September 9-7, 9.
In search activities in disaster areas, it is important to analyze images taken by robots to find people and objects in the rubble.In recent years, while remarkable progress has been made in the field of image recognition due to the development of artificial intelligence technology, it takes time and effort to build a function (image recognizer) for obtaining useful information from images, and it takes time and effort to construct a disaster site. It was difficult to apply it in such an environment.
To address this issue, the research group has developed a technology to create an image recognizer from a small amount of information by streamlining the process of obtaining recognition ability from video.As a result, it will be effective quickly even at the first disaster site, leading to video analysis and discovery of important information.
At the same time, we have also developed a technology for estimating the materials that make up rubble and a technology for quantifying the surface condition such as being wet or soiled.It is also expected to find a place that is easily collapsed or slippery in advance and prevent the disaster site from collapsing.
As a result of mounting this system on a snake-type robot and conducting verification in a test field simulating a damaged wooden house and in a forest environment, a certain level of performance was confirmed.
The technology developed this time is also suitable for dealing with tasks for which it is difficult to prepare data in advance, and it was devised to function even in a messy environment, and it was difficult to obtain sufficient performance with the conventional method. It can also be used for the target.In the future, it is expected that not only disaster response but also infrastructure inspection and expansion into agriculture, forestry and fisheries will be expanded.
