verisense_count_steps: Count Steps According to Gu et al, 2017 Method

In walking: Segments Accelerometry Data into Walking Bouts using Open Source Methods

Count Steps According to Gu et al, 2017 Method

Description

This method is based off finding peaks in the summed and squared acceleration signal and then using multiple thresholds to determine if each peak is a step or an artifact. An additional magnitude threshold was added to the algorithm to prevent false positives in free living data.

Usage

verisense_count_steps(
  data,
  sample_rate,
  k = 3,
  periodicity_range = c(5, 15),
  similarity_threshold = -0.5,
  continuity_window_size = 4,
  continuity_threshold = 4,
  variance_threshold = 0.001,
  vm_threshold = 1.2,
  peak_finder = c("fast", "original"),
  verbose = TRUE,
  global_vm_threshold = 0.025
)

verisense_count_steps_revised(
  ...,
  k = 4,
  periodicity_range = c(4, 20),
  similarity_threshold = -1,
  continuity_window_size = 4,
  continuity_threshold = 4,
  variance_threshold = 0.01,
  vm_threshold = 1.25
)

Arguments

data	A data.frame with a column for time in POSIXct (usually HEADER_TIMESTAMP, not required), and X, Y, Z
sample_rate	sampling frequency of the input data
k	window size for controlling peak finding.
periodicity_range	a length-2 vector of the range of periodicity. These are integers that represent samples not seconds.
similarity_threshold	threshold (in g) for similarity between magnitude of peaks
continuity_window_size	Window size for continuity
continuity_threshold	Threshold for continuity
variance_threshold	Variance threshold for the signal
vm_threshold	vector magnitude threshold for a peak to be called a peak
peak_finder	function to find peaks, either the "original" from the code, or the optimized "fast" version.
verbose	print diagnostic messages
global_vm_threshold	Global acceleration VM threshold (in standard deviation) for the total vector. If sd(acc) < thresh no steps are estimated. Set to 0 to run estimation regardless.
...	not used, used to passes arguments from verisense_count_steps_revised to verisense_count_steps

Value

A vector of length round(nrow(input_data) / sample_rate) of the estimated steps, where the data is rounded to seconds

Note

the ⁠_revised⁠ version is the same algorithm with different defaults for the parameters as based on doi:10.3390/s22249984.

Author(s)

Matthew R Patterson [e-mail], MIT license, Copyright (c) 2020 Shimmer

Examples

csv_file = system.file("test_data_bout.csv", package = "walking")
if (requireNamespace("readr", quietly = TRUE)) {
  x = readr::read_csv(csv_file)
  colnames(x)[colnames(x) == "UTC time"] = "time"
  out = verisense_count_steps(x, sample_rate = 10L)
}
input_data <- matrix(runif(500 * 3, min = -1.5, max = 1.5), ncol = 3)
verisense_count_steps(input_data, sample_rate = 15L)
verisense_count_steps(input_data, sample_rate = 15L, peak_finder = "fast")
acc = sqrt(rowSums(input_data^2))
verisense_count_steps(acc, sample_rate = 15L, peak_finder = "fast")

walking documentation built on Aug. 4, 2026, 1:07 a.m.