Imagine if you could go into a grocery store, take what you need, and walk out. In fact, we don’t need to imagine; Amazon has already done that with their Amazon Go stores. This new form of retail is possible because of something called “indoor tracking.” Indoor tracking applications range from smart retail (e.g. Amazon Go) to finding a package out of thousands in a warehouse to tracking cars in a parking garage. The overall benefits of indoor tracking are vast: lower operational costs, better coverage, arrangement optimization, etc. However, tracking assets indoors is application specific and quite difficult. Let’s go through some technologies that could help you implement an indoor tracking solution in this overiew—an introduction to Leverege’s forthcoming series on indoor asset tracking and positioning systems.
Let’s first go over the Global Positioning System (GPS) technology. GPS has become ubiquitous. Everyone with a smartphone has GPS at their fingertips. Consumers can buy GPS modules for less than $50. Many IoT real-time asset tracking applications are now possible with the availability of GPS. However, GPS still struggles in one key application—indoor tracking. GPS technology struggles to track well indoors because GPS signals may not be able to penetrate through built structures. You can use an indoor gateway to boost signals and give more accurate positions, however, GPS is generally inaccurate for indoor tracking applications.
RF as an Indoor Tracking Technology
The next set of technologies is radio frequency (RF) indoor tracking, which uses beacons to send WiFi or Bluetooth signals depending on the application. WiFi and Bluetooth technologies use parameters such as the Received Signal Strength Indication (RSSI), Angle of Arrival (AoA), and Time of Arrival (ToA) to determine the location of a broadcasting device. The RSSI value is inversely proportional to distance, so a device’s distance from a given radio beacon can be approximated. AoA describes the angle of a wave at particular radio beacon. ToA refers to the time at which a wave arrives at a particular beacon. AoA and ToA are difficult to obtain indoors because radio travels at the speed of light. Given the obstacles an indoor scenario presents, multiple receivers are needed for these measurements. WiFi and Bluetooth receivers aren’t typically built to deal with this level of precision, so RSSI is the primary parameter used for RF indoor tracking.
Ultrasonic sensors could be used in an array to figure out where an asset is in space. Much like RF, we can use the sound waves to determine the RSSI, AoA, and ToA of a device. However, it’s much easier to determine distance because sound travels much slower than light, so the calculated distance using the ToA is typically much more accurate than RF. However, ultrasonic sensors are very limited in terms of the width of the beam pattern, which means they can detect assets far away, but neither too far left nor right of the ultrasonic beam.
Indoor tracking uses computer vision much like it’s normally applied to vision tracking tasks. The computer vision algorithm will identify the asset that needs to be tracked and follow it as it moves through the environment. This is part of the technology that Amazon uses in its Amazon Go stores to track customers as well as items for sale. However, this technology is very use case specific. For an RF application, it doesn’t matter if a person or your pet dog is wearing the tracker. However, this makes a huge difference for a computer vision system.
Algorithms & Tuning
There are numerous computer vision algorithms, so I’ll just go over those that are relevant to RF and ultrasonic indoor tracking technologies. Trilateration and Triangulation are mathematical algorithms that use three beacons to determine the location of an asset. Trilateration and Triangulation use distance and angles, respectively. These algorithms aren’t foolproof because RF and sonic waves don’t travel unobstructed to beacons. To account for obstruction, we can use empirical models to take into account the time of day, number of assets, etc., in order to determine location more accurately. However, even using those aiding data, indoor positioning is still difficult. RF is only accurate to two or three meters, which is a significant area in a small office space, grocery store, or retail environment.
The final step here is to combine multiple sensor types using a technique called “sensor fusion.” An example of sensor fusion is using a BLE tag with an accelerometer and gyroscope. The accelerometer and gyroscope can give you the general movement and direction of a given asset, which can help you track it with RF technologies. With empirical models, you can get a better idea of where the asset is compared to traditional RF tracking.
Indoor tracking is quite difficult for two reasons: technology isn’t extremely accurate and the environment is constantly changing. With both of those factors influencing outcomes, indoor tracking solutions can run awry quickly. RF tracking is accurate up to a few meters, however, some indoor measure a mere few meters. Ultrasonic tracking is much more accurate, but the width of its beam pattern is quite constrained. Sensor fusion provides one promising solution by leveraging multiple sensor inputs to get more accurate data.
Regarding the issue of changing environments, RF and ultrasonic waves can absorb and reflect off of various surfaces, which will change key characteristics of these waves—RSSI, AoA, and ToA. In a static environment, you can create an empirical model that accounts for unique scenarios. However, we want to track people and things that are moving. As a result, the surfaces are moving and changing. Therefore, the empirical model has to be more complex to deal with these changes.
There are numerous technologies such as WiFi, Bluetooth Low-Energy (BLE), Ultrasonic, Computer Vision, etc., which can be deployed to track assets indoors. However, indoor tracking is difficult because of accuracy limitations due to the underlying technology and also because of shifting indoor environments. Methods such as empirical models and sensor fusion can be used to account for these issues, but robust and flexible deployments remain a huge challenge. Indoor tracking has many exciting and important use cases. We hope to see more applications in the near future.
This post was originally published on Leverege’s blog.