Table 1 Summary of recent accelerometer-based step counting studies

Ying (2007) [20]

Treadmill

Lateral side of left and right foot

Dual axis accelerometers (200 Hz)

Pan Tompkins, template, dual axis peak detection

Accurate step detection

Qualitative comparison

Zijlstra (2003) [24]

Hallway

Trunk

Triaxial accelerometer (100 Hz)

Peaks preceding sign change in forward acceleration

Foot strike

Within 0.02 s (SD <0.03)

Huang (2012) [10]

Treadmill

5 locations

HTC smartphone (10 Hz)

Threshold from training period

Count steps

93-96% step count Accuracy

Naqvi (2012) [11]

Level ground

Near centre of mass (COM)

Smartphone (100 Hz)

Adaptable threshold

Count steps

1-2 step error (of 15-40 steps)

Kim (2004) [22]

Hallway

Ankle

MEMS accelerometer, vertical and forward (100 Hz)

Sequential thresholds to recognize swing phase, foot strike

Count steps, estimate distance

<1% step count error 5% distance error

Yang (2012) [3]

25 m, hallway

Lower back in belt

HTC smartphone (25 Hz)

Peaks preceding sign change in forward acceleration, manually verified

Foot strike, regularity, symmetry

Visually verified to 100% accuracy

Ayub (2012) [12]

Hallway

3 locations

HTC smartphone (25 Hz) interpolated 50 Hz

Zero crossing and threshold lengths, Variance detector

Step count, stride length

1.5-5% step count error

Derawi (2010) [23]

20 m, level ground

Left leg by hip

Accelerometer (100 Hz)

Neighborhood search for minimum peaks

Cycle detection, distance metric

EER = 5.7%

Martin (2011) [25]

Varying speeds

Varying locations

Accelerometer (30 Hz)

Continuous wavelet transform (CWT)

Stride length (step counting)

Not reported

Kim (2013) [21]

Treadmill varying speeds

Left waist

Triaxial accelerometer (32 Hz)

Heuristic, adaptive threshold, adaptive locking period

Step count and activity monitoring

97% Recognition rate