RadarCat for object recognition

The Soli radar, from Google ATAP, was designed to track micro finger motion for enabling gesture interaction with computing devices. First shown at Google I/O ’15, Soli introduces a new sensing technology that uses miniature radar to detect touchless gesture interactions.

As one of the few sites to receive the Google ATAP Soli AlphaKit in 2015 we have discovered, developed and tested a new and innovative use for Soli, namely, object and material recognition in a project we call RadarCat. To achieve this, we have used the Soli sensor, along with our recognition software to train and classify different materials and objects, in real time, with very high accuracy. A short snippet of this work was included in the Google ATAP Soli Alpha developers kit video shown at Google I/O ’16.

RadarCat (Radar Categorization for Input & Interaction) is a small, versatile radar-based system for material and object classification which enables new forms of everyday proximate interaction with digital devices. In this work we demonstrate that we can train and classify different types of objects which we can then recognize in real time. Our studies include everyday objects and materials, transparent materials and different body parts. Our videos demonstrate four working examples including a physical object dictionary, painting and photo editing application, body shortcuts and automatic refill based on RadarCat.

30 second teaser:

Full video:

Curated video by Futurism, with more than 1 million views!

Talk at UIST 2016:

Beyond human computer interaction, RadarCat also opens up new opportunities in areas such as navigation and world knowledge (e.g., low vision users), consumer interaction (e.g., checkout scales), industrial automation (e.g., recycling), or laboratory process control (e.g., traceability).

Our novel sensing approach exploits the multi-channel radar signals, emitted from a Project Soli sensor, that are highly characteristic when reflected from everyday objects.

RadarCat system diagram

Our current studies demonstrates that our approach of classification of radar signals using random forest classifier is robust and accurate.