Browse All Categories
Jean-Frederic Gauvin
By Jean-Frederic Gauvin on February 03, 2022

Deploying indoor localization with artificial intelligence

Indoor localization is a challenge

Few of our competitors provide localization services. The reason is that indoor localization is a challenge, mainly due to noise and the general complexity of the electromagnetic environment.  Received Signal Strength Indication (RSSI) is commonly used in proximity applications and for localization purposes, but does not provide good distance approximations without proper calibration. The task of indoor localization is generally a tedeous task. 

The conventional approach has many downsides

Location fingerprinting is a well-established approach for indoor localization. It consists of two phases; the offline phase and the online phase. In the offline phase you must manually walk around and capture fingerprint data throughout the environment. Each fingerprint is then stored in a database. Later, in the online phase, a simple algorithm can guess the node's position by performing a lookup to the fingerprint database. 

Fingerprinting is both time-consuming and labour intensive. In addition, if the environment ever changes, the data collection process has to be redone. For this reason it's very difficult to maintain these systems over time. 

In addition, location fingerprinting is costly. The accuracy of the system is impacted by how many devices are placed in the environment. The more devices, the more accurate the system becomes. 

We have a better solution

Meshtech is currently developing a new indoor localization system that will provide high-performing, quick and easy localization of Meshtech sensors.  We utilize artificial intelligence (AI) to recognize patterns in the electromagnetic environment and enhance the positioning accuracy over conventional techniques. Our solution can predict the room or area where the node is with high accuracy, even without requiring new hardware. The service can also provide pin-point 3D positions with roughly 3m accuracy under ideal curcomstances. Our positioning system is capable of automatic calibration, meaning that you can sit back and relax while the system tunes its parameters. There is no need for manual fingerprinting. You can quickly deploy localization for your network. 

It can run basically anywhere

Our positioning system implements a web API and a back-end positioning engine. You can run the system locally or in the cloud inside Docker containers. The positioning engine computes the positions by using RSSI-data sent from Gateways to the positioning system via MQTT

It is fully scalable

Since the calibration process is automatic, and no manual fingerprinting is requried, it's much easier to scale our solution than the conventional approach. It even works across multiple sites and in multi-floor environments. All RSSI data is simply sent through the same MQTT broker from multiple gateways. The positioning engine predicts the position of all your sensors and make the positions accessible through its API. 

And it requires low manual effort

Our gateways and network extenders function as access points. The APs are usually statically positioned in the environment and are thus, used as anchors for the positioning algorithm. To provide accurate predictions, the AP positions, and ideally the floor plan should be defined. We have made this process as easy as possible, even for environments with multilple floor levels. If you don't want to draw the map yourself, we can do it for you. 

You decide how to present it

We provide the node positions and you decide how you want to present it to end users. Our system has no single front-end. It can be integrated with any top system. 

Published by Jean-Frederic Gauvin February 3, 2022
Jean-Frederic Gauvin