GPS and IMU Fusion for Human Gait Estimation

Walking Trajectory Estimation Using Multi-Sensor Fusion and a Probabilistic Step Model

“…To overcome this drawback, IMU algorithms often implement strap-down integration, which leverages outside information to mitigate drift. To integrate acceleration data twice to obtain the position of a walking individual, the zero-velocity update (ZUPT) method [ 33 ] is commonly used. The ZUPT method uses the knowledge that the foot must have zero velocity at regular stance phases in the gait cycle [ 34 , 35 ].…”

“…Fusing the sensor data from both these types of sensors yields a much higher accuracy, as shown in [ 33 ]. Let be the estimate of an individual’s 2D (i.e., lat./long.)…”

“…Under the assumption that the IMU and GPS readings are drawn from Gaussian distributions, they can be fused online according to where is the covariance of the fused GPS/IMU measurement and is the mean of the fused estimate. It was shown in [ 33 ] that, if the sensors’ statistics are accurately characterized, the fusion of their measurements will provide an improved estimate of a human’s location.…”

“…This accuracy difference is illustrated in Figure 1, where an IMU is initially more precise than a GPS but eventually IMU uncertainty diverges cubically whereas GPS uncertainty is stationary. Fusing the sensor data from both these types of sensors yields a much higher accuracy, as shown in [33]. Let x I k be the estimate of an individual's 2D (i.e., lat./long.)…”

“…where Σ GI x k is the covariance of the fused GPS/IMU measurement and x GI k is the mean of the fused estimate. It was shown in [33] that, if the sensors' statistics are accurately characterized, the fusion of their measurements will provide an improved estimate of a human's location.…”