Sensors for Gait Analysis

• Sensors for Gait Analysis
  • Sensors
    • Inertial Measurement Units
      • Sensor placement
        • Preference in literature
        • (inter- and intra=) Segments to choose
      • Filter
      • Application
    • Insole pressure sensors

Sensors

Distribution of studies based on the type of wearable sensors used	Distribution of studies based on the type of sensors used for ground-truth validation of IMU-based gait analysis. Absolute number of studies in each category is listed within parentheses.

IMU—inertial measurement unit, IPS—insole pressure sensor, EMG—electromyography sensor.

Only 11% of studies using a rule-based method required the placement of more than one IMU per leg, in contrast to 53% of machine learning approaches. This means applications using the rule-based methods typically require a less complex sensor setup, which would be preferable to the end-user. Furthermore, in 84% of studies using rule-based methods, the approach could be used independently for one leg or the other, compared to 47% for machine learning.

Inertial Measurement Units

Sensor placement

Preference in literature

Number of studies using inertial measurement units (IMUs) that placed the sensor(s) on specific anatomical locations. Single placement contains studies where sensor(s) were placed only in one anatomical location. Placement combinations’ columns indicate studies where sensor(s) were placed in more than one location. Each relevant location is marked by a shaded cell and the number of studies using this combination is indicated at the bottom of the column. The total indicates the sum of studies where the sensor(s) were placed on that given anatomical location.

(inter- and intra=) Segments to choose

Foot

• Li et al., 2016¹: compared IMU signals from the thigh, the shank and the foot, based on the stability of “energy of acceleration” which is the norm of raw acceleration minus gravity
• Jasiewicz et al., 2006²: significantly more accuracy in gait event detection methods
• Taborri et al., 2014³: better accuracy of a HMM-based classifier for gait event detection when using data of the foot.
• Wenger, N. et al., 2015⁴: neuro-behavioral experiments suggesting that limb endpoints are the primary variables used to coordinate locomotion in animals and humans

Shank

• Bejarano et al., 2013⁵: the root mean square error between each cycle (as well as the average)

Irrelevant (in some tasks)

• Mazilu et al., 2012⁶: in the context of freezing of gait

Intra-segment

​ mount/socket; Anwary et al.⁷: the medial arch followed by the Achilles tendon(also the centre point of foot length; location 1 🔽); based on accuracy of extracted features: Number of Strides, Distance, Speed, Stride Length, Stride Period, Stance Length, Stance Period, Swing Length, Swing Period and ratio of stance and swing

Filter

39 out of 69 studies at least used a preprocessing filter, 31 of which used a low-pass filter among which 15 used the Butterworth low-pass filter. Note that preprocessing filters add to the latency in data processing, which is undesirable in a real-time system.

Application

Insole pressure sensors

Only six out of 28 studies related to IPS used any sort of preprocessing/filtering. Instead, these studies relied directly on the raw signal from the IPS, likely contributing to shorter latency, an advantage when it comes to real-time systems.