Aided Inertial Technology:
More Than IMU + GPS

Mike Zywiel

Over the last decade or so, inertial technology as applied to airborne surveying has progressed from being virtually unknown to something that almost everyone in the industry has contemplated using. Direct georeferencing—the ability to georeference remote sensing data without the need for ground control—enabled by inertial technology, is now an indispensable tool for many airborne remote sensing organizations, with aided inertial systems (often referred to as IMU + GPS) used on virtually every LiDAR system as well as on a great number of film and digital cameras.
Even though the existence of such systems is well known, much of the technology behind them is not understood quite as well. Although the industry has adopted the term IMU+GPS (or simply IMU) to describe these systems, is that really the best description? If you want to build a LiDAR system, can you just buy an IMU, bolt it to the sensor, develop an interface to display and log IMU data, and then go fly? Are such systems just an IMU with a survey-grade GPS receiver attached? These are some of the questions that will be addressed in this article.

What Is an IMU, and What Does it Do?
An IMU (Inertial Measurement Unit) is a self-contained sensor made up of one accelerometer and one gyro for each of three axes. The accelerometers sense linear accelerations, and the gyros sense angular rates or rate of rotation. Hence, when attached to an object (such as a camera or LIDAR), the IMU will sense and output the acceleration and angular rate experienced by the object as it moves over the earth. The outputs include not only the motion of the object, but also the acceleration due to gravity and the angular rates due to the earth’s rotation.
To compute the change in orientation and change in position of the object, earth rate and gravity must first be removed from the IMU outputs, then the remaining accelerations and angular rates are integrated (or summed). This can be quite challenging in itself, since it requires knowledge of where you are on the earth to account for variability in the earth rate and acceleration due to gravity.

Limitations
The IMU knows nothing about the absolute orientation or the position of the object on the earth. Significant software-based signal processing is required to convert the raw IMU data output into a complete position and orientation solution; long-term biases must be detected and compensated for, and incremental data must be integrated to derive displacement.
At this point, another significant limitation of IMUs comes into play, namely its lack of long-term accuracy. Like any electro-mechanical system, IMU data has inherent errors, and generally speaking, the more expensive the IMU, the smaller the errors. However, because inertial navigation solutions require the summing of past observations, the IMU errors stack up and transfer into the position and orientation solution. Therefore, given enough time, even an accurate IMU will result in position errors at the km level, and orientation errors of several degrees.
So, how to overcome the two significant limitations of IMU-based navigation systems: lack of absolute data and poor long-term accuracy? The answer is to combine the inertial data with a complementary sensor, such as GPS, to aid the inertial solution—thus, the term “aided inertial technology.”
The function of the GPS is twofold: to seed the inertial system with position data, and to control drift by supplying position data that has stable accuracy over the long term (in contrast to IMU data, GPS accuracy does not degrade over time). Sounds simple enough, but what exactly does such an integration involve?

Navigation Solution
The first requirement is navigator software. The IMU outputs are seeded with GPS-derived position and velocity, then integrated to generate a navigation solution. However, before the integration can begin, the navigator software must first determine the initial heading and tilt of the system with respect to the earth. It does so based on observations of the rate of rotation of the earth in the gyros, and gravity in the accelerometers. Then it must remove gravity and the earth rate from the incremental velocity and angular rate data, and integrate the data to compute the IMU position and orientation solution of the object relative to the earth. However, as previously described, because this process is based on the integration of past observations, the sensor's inherent errors stack up over time, and the solution will drift.

The Noise Factor
A second issue that must be overcome is that of noise. GPS position data is constrained in its error, but is quite noisy; the magnitude of this noise is a factor of the quality of equipment used. It is reasonable to follow the adage that “you get what you pay for,” and generally speaking, the more expensive the GPS receiver, the less noisy it will be. However, even the quietest survey-grade receivers are extremely noisy as compared to the inertial solution. Thus, the sensor outputs must be cleaned of noise before use for navigation purposes, otherwise, the navigation solution will be prone to large, unpredictable errors.

Error Estimation and Control
Error estimation is performed using a set of software algorithms called a Kalman filter. The Kalman filter is able to model IMU and GPS errors due to sensor bias and noise in real-time, and these errors can then be reset in the navigation solution. With a carefully tuned Kalman filter, complementary errors (errors that are uniquely separated from each other) can be all but eliminated, producing a stable, robust low-noise position and orientation solution when used to aid the inertial solution.
A third significant component of an aided inertial system is the error control that is used to apply the resets estimated by the Kalman filter to the navigation solution.
All three of the above steps—the generation of a navigation solution from raw IMU data, the filtering of noise from the GPS position solution, and the estimation and error control of the inertial solution—are achieved through signal processing in the aided inertial system's firmware. All three steps are critical to the final accuracy of the inertially-derived position and orientation solution.
The accuracy of the final solution depends on several factors: the accuracy of the IMU and GPS receiver, the design of the Kalman filter software, and the nature of the error control algorithms, to name a few. Different applications will have a different combination of these elements that will produce the optimal solution.
In short, what is referred to by many as IMU+GPS is much more than just an IMU and GPS receiver. It is true that a tactical grade IMU can be purchased for $25,000, a survey-grade GPS for $10,000 and a processing computer for $5000, far less than the cost of an aided inertial system. However, it is important to understand that these costs do not include the development of the software to handle the IMU and GPS data outputs, or converting them from raw incremental values into a robust, accurate and reliable position and orientation solution. Such costs can quickly eclipse the cost of the hardware several times over.

About the Author
Mike Zywiel manages the marketing and communications department at Applanix.

Back