Introduction

Modern localization technologies show poor performances in environments that are hostile to radio waves, such as building sites and hospitals. Moreover, many indoor solutions rely on a large number of devices resulting in complex setups. Our patent describes a system based on an inertial measurement unit and machine learning techniques that is able to overcome said limitations.

Technical features

The system’s main hardware component is a differential motion sensor, such as an Inertial Measurement Unit. This device communicates with a computing unit where the differential motion data get processed by inductive algorithms such as neural networks. The first step of the analysis consists in the filtering any involuntary movement, as this is particularly important in some scenarios like tracking people. Whenever the system recognizes voluntary movements, it then identifies a path in the environment that the tracked entity follows, by comparing motion parameters such as velocity, change of orientation and duration of motion, with a number of motion models that it previously learnt during a training phase. The models can be further refined and personalized to fit a specific person or object, moving or being moved in the environment, through an initial learning and calibration phase.

Possible Applications

• Indoor positioning of people and objects;
• Safety in building sites and hospitals;
• Internal logistics;
• Retail localization;
• Indoor navigation.

Advantages

• Can be employed in any kind of environment;
• Does not depend on absolute reference systems;
• Installation and maintenance are simple;
• Robust to involuntary movements.