IoT sensors: fusion or confusion?
What keeps the Internet of Things from becoming a tangled jumble of incoming data from various connected devices? And within a single device that feeds information into the IoT, how does sensor fusion work?
Implemented in a wide variety of product categories, sensor fusion is a necessity for applications like wearable gadgets supporting health and fitness and the body motion tracking devices used in the production of advanced CGI movies and gaming. Even products as “simple” as your smartphone require multiple sensors with lots of degrees of freedom and a powerful MCU, to collect, coordinate, process, analyze, filter, and communicate data.
The latest versions of sensor fusion are a set of adaptive prediction and filtering algorithms based on extended Kalman Filter theory that uses quaternion concepts to avoid mathematical singularity and deliver more reliable results. These algorithms "make sense" of all of the complex information coming from multiple sensors, including accelerometers, gyroscopes, compasses, and pressure sensors, by taking each sensor's measurement data as input, and compensating for drift and other effects and limitations of each individual sensor, to output accurate and responsive dynamic results.
For one example, let's take a look at the application of pedestrian dead reckoning (PDR) and the fusion of four sensors and five inputs. You've got your GPS input as well as your accelerometer, gyro, magnetometer, and pressure-sensor data coupled to an MCU.
Working together, a three-axis accelerometer and three-axis gyroscope function as a strapdown inertial navigation or pedometer-based portable navigation device. The accelerometer provides step detection, and the tilt-compensated compass – a three-axis magnetometer – if disturbed, allows the gyro to make heading adjustments. The compass calculates magnetic field and compensates for the gyro's zero-rate drift over time. Meanwhile, the pressure sensor, working with the accelerometer, acts as an altimeter and conveys floor changes for indoor navigation.
So much data, so little time...
The MCU houses the Kalman Filter, which predicts sensor errors from the equations of inertial navigation. It estimates and compensates for the gyro's long-term bias drift, magnetic anomalies, and provides data for dead-reckoning applications when GPS info is unavailable.
Et voila! Sensor fusion.
To read more of this external content, go to "In the next few years," on Embedded.com's sister publication, EETimes Europe.