Terrain Based Parameter Optimization for Zero-Velocity Update Inertial Based Navigation Solutions

Foot-mounted inertial navigation can achieve high accuracy from poor quality sensors by performing zero-velocity updates (ZVU) every step. ZVUs keep sensors calibrated and minimize drift in the position solution. To get the best performance, the ZVU algorithm needs to be carefully tuned. This paper demonstrates the benefit of adapting that tuning to terrain type. Four terrain types are considered: concrete; grass; pebble and sand.

Using inertial measurement unit (IMU) context detection as a foundation, this paper aims to determine optimum ZVU parameters for various terrains. Our previous work from 2023 [1] uses context detection with MEMS IMUs to identify terrains using a k-Nearest Neighbor (kNN) algorithm with 99.24% accuracy. Using an IMU attached to the right foot, data was collected across four terrains for ZVU specific parameters – gyroscope magnitude threshold value, zero velocity accelerometer magnitude threshold, zero-velocity interval duration, and time between each zero-velocity occurrence. Figures 1 - 2 show examples of the parameters during the walking cycle.

Figure 1 ZVU walking parameters in the pedestrian walking cycle as shown using accelerometer magnitude.

Figure 2 ZVU walking parameter in the walking cycle as shown using gyroscope magnitude.

The four parameters comprise the tuning for the ZVU algorithm used for pedestrian navigation. The gyroscope and accelerometer thresholds are the maximum values where the stance phase begins and identify zero-velocity intervals. The zero-velocity interval duration and time between occurrences identify when the ZVU should be applied. The terrains are also separated into two classes: hard (concrete and grass) and soft (pebble and sand). Hard terrains are classified by firm foundations whereas soft terrains, the ground gives way during walking.

Figure 3 contains a percent change comparison between hard and soft terrains. There is a 112.1% difference in accelerometer threshold values and 41.4% difference in interval durations between the two classes.

Figure 3 Percent change comparison of hard and soft terrain classes.

With the large differences between surface hardness it is expected that terrain-dependent ZVU algorithms will significantly improve on current fixed parameter algorithms. An assessment of position accuracy will be included in the final paper.