Real Time Transfer Technique Assessment Using the Kinect 2 Sensor

Approximately 3.3 million Americans over the age of 15 use wheelchairs daily (Census, 2010). Wheelchair users rely heavily on their upper extremities to complete common but essential activities of daily living such as getting in and out of bed, transferring to a toilet or a shower, and transferring in and out of a car. Manual wheelchair users will perform on average 14 to 18 transfers a day, which are extremely physically demanding and can lead to upper extremity pain and injury (Hogaboom, Worobey, & Boninger, 2016; Tsai, Hogaboom, Boninger, & Koontz, 2014). Research shows that the prevalence of upper extremity pain, specifically shoulder pain, in wheelchair users ranges between 31 and 73 percent (Cooper R, 1998). Unfortunately, shoulder pain leads to decreased quality of life and participation in physical activity (Gutierrez et al., 2007). The use of good transfer mechanics to avoid pain and injury is important for wheelchair users when performing transfers.
The Transfer Assessment Instrument (TAI) is a tool used by clinicians and therapists to assess transfer quality and identify problems in wheelchair transfers which can cause increased forces on upper extremity joints (Tsai, Rice, Hoelmer, Boninger, & Koontz, 2013) (Tsai et al., 2014). The TAI is a 2 part assessment, with the first part consisting of specific transfer issues rated on a “yes”, “no”, or “not applicable” scale. Part 2 consists of global performance of transfer quality, techniques and indication of assistance, which is scored from 0 to 4. The final TAI score is the average of both part 1 and part 2 from 0 to 10 (Koontz, Tsai, Hogaboom, & Boninger, 2016), where a 10 is perfect transfer technique and a 0 is very poor transfer technique. Some items of the TAI, particularly those related to the body mechanics can be subjective, and are open to different interpertations.
The Microsoft Kinect is a motion sensing device originally designed for use with video game systems. The Kinect sensor has been shown to be accurate at sensing joint centers and has found applications in various clinical and research settings (Xu, McGorry, Chou, Lin, & Chang, 2015). In preliminary work, we developed algorithms that combined selected biomechanical variables meaured by the Kinect and manual measurements of the body’s segments lengths to discern proper from improper technique (Lin Wei, 2017). The approach used Kinect’s developers software module, a separate post-processing module (Matlab and Mathmatica), and a statistical analysis module (SPSS). The requirement of having to take physical measurements of the person’s dimensions and the use of multiple applications to obtain results makes it an impractical solution to implement in a clinical setting.
The aim of this study was to 1) create an integrated software program that can automate the TAI scoring using the Kinect in real-time, 2) evaluate the performance of the program in predicting the TAI items scores correctly with manually-measured body measurements, and 3) evalute the performance of the program to predict TAI scores when Kinect-measured body measurements are used.