The Korean Institute of Advanced Scientific and Technological Research (KAIST) developed Knocker, a system for recognizing objects by analyzing the specific vibrations that occur when tapped. An ordinary smartphone acts as the percussion instrument, and for the analysis of the data, the gyroscope, accelerometer and microphone already available in it are used. Nothing else is required for Knocker to work, it is a purely software tool that can be used on any devices.
Knocker action is based on machine learning, the technology allows you to analyze the smallest nuances of vibrations and distinguish characteristic patterns from them. Initially, the system does not know anything, and it needs to be taught, for example, to distinguish a glass bottle from a plastic one. In theory, a smartphone with Knocker can recognize almost any object, but so far there are only 23 objects in its “memory”.
In silence, Knocker recognition accuracy reaches 98%, in the conditions of a cafe or bus stop noise it decreases to 83%. However, the main idea is not to identify the object, but in a new form of communication – when the smartphone finds out what is in front of it, it automatically turns on the desired service. In this way, for example, you can quickly order pizza by tapping on an empty box. Find out the schedule of transport by gently tapping on a road sign, or control a smart home by tapping on various household appliances.