CAPÍTULO V -DE LOS RECLAMOS-
ART.246:REQUISITOS PARA LA CABINA DE ASCENSORES
In order for ExerciseCheck to compare the reference and practice exercises, the joint
each exercise. For example, for the squat exercise, most of the joints are moving.
However, the range of motion for all these joints is not the same and some of them
carry more information than others. Thus, by specifying the “most important joint”
(or joints) for each exercise, ExerciseCheck can provide a more accurate analysis
result. A description of which joint is selected for the analysis of which exercise is
provided in Table 4.3. In addition to the most important joint, the direction of the
movement and the initial direction are specified when a new exercise is added to
the system. For the squat exercise, the most important joint is the SpineBase, the
movement is up and down, and it begins when the patient moves downward.
Preprocessing
In our analysis, first both the practice and reference trajectories are normalized using
the corresponding neck point captured in the beginning of the exercise, as well as
the “body width” and “body height” variables.The body width variable is defined as
the difference in “depthX” from ShoulderLeft to ShoulderRight, and the body height
variable is the difference in “depthY” from SpineBase to Neck. Then the trajectories,
both practice and reference, are smoothed using Gaussian smoothing.
Accuracy
ExerciseCheck employs Dynamic Time Warping (DTW) to compare the reference tra-
jectory obtained during PT-supervised exercising to the practice trajectory obtained
during unsupervised home-based exercising. It calculates an optimal match between
two given sequences (e.g. time series of joint positions) by aligning the sequences in
a non-linear fashion and finds the distance between the two sequences. The distance
or dissimilarity measure can be converted to an accuracy measure using the following
formula:
Accuracy = 1 − distance
The maximum distance is obtained using DTW between the reference trajectory and
a synthetic practice trajectory where the user is not moving at all. It is worth noting
that if the user performs a random practice exercise, the distance (output of DTW)
might exceed the maximum distance; in this case, we use a threshold to avoid negative
values for accuracy and set the accuracy to be zero.
Speed
For speed analysis, ExerciseCheck previously (Saraee et al., 2017) used the inverse
ratio of the time it took a user to perform reference or practice exercise. The as-
sumption in using this “movement completion time” was that since the same number
of repetitions had to be performed in the reference and practice exercises, the time
it takes to complete an exercise correlates with the speed of completion. In the new
version of ExerciseCheck, however, we have implemented repetition counting, which
complicates the speed analysis. The users are supposed to exercise until the expected
number of accepted repetitions is reached. Thus, the total number of repetitions
can be higher in practice compared to the reference and consequently, the former
assumption does not apply anymore.
ExerciseCheck estimates the speed of the user based on the most important joint,
i.e., its movement in the specified direction. The estimate is an average over the
speed measurements per frame above a given threshold. The movement speed in
each frame is defined as the absolute value of the movement of the joint compared to
the previous frame. We have also set a threshold for minimum speed to exclude the
frames for which a very low speed is measured. If the patient is resting between the
two repetitions and the movement speed is very low, we do not want to include speed
measurements during the resting period in our calculation. By setting this threshold,
Table 4.3: The exercises used in our experiment, including descrip- tions of the active motion and the joints used for quantitative analysis.
Exercise name Description Most important
joint
Shoulder Abduction
Shoulder abduction involves the rotation of the shoulder joint to move the arm straight out and away from midline of the body.
Wrist
Hip Abduction Hip abduction involves rotation of the hipjoint in such a way as to move the midline of the body.
Ankle
Squat
The squat involves translation of the hip , as the torso is repeatedly raised and low- ered during the exercise.
Hip
Marching in place This exercise involves the translation, ofthe knee joint as the leg is repeatedly raised and lowered during the exercise.
Knee
Range of motion, difficulty adjustment, repetition counting
ExerciseCheck can count the number of repetitions in each trial for an exercise. In
order to do this, ExerciseCheck monitors the position of the most important joint in
each frame and increases the number of repetitions as the joint reaches the specified
threshold based on the range of motion in the reference exercise.
ExerciseCheck also provides different levels of difficulty. A more difficult level
of exercise assigns a wider range of motion to the user and increases the repetition
counting more strictly. This feature is enabled by adjusting the threshold for the
expected active range of motion and tracking the position of the most important
joint.
To count repetitions when the user is practicing a certain exercise, ExerciseCheck
projects the current position of the important joint read from the Kinect sensor to the
reference space using the neck position and either body width (if moving in X, i.e.,
for an exercise that involves moving up and down, with “moving up” as the starting
direction, we take the “depthY” of the most important joint and denote its current
readout from the Kinect as “currP” and its correspondent in the reference space as
“currR.” We also denote the highest number that the participant reached during
his/her reference recording as “refMin” and the lowest as “refMax.” The baseline
AROM is therefore the difference between “refMin” and “refMax.” Then, according
to the difficulty level chosen by the participant, we set the threshold position. We
have three levels: difficult corresponds to 90% coverage, medium to 75% and easy to
50%.
We continue the description of the example given above: Once “currR” is smaller
than the threshold position, and while the direction of movement is going up, i.e., the
same as the starting direction of the exercise, the repetition count increases by one.
To obtain the ratio of the range of motion in each repetition, ExerciseCheck first
finds the local maximum of each trajectory for the practice exercise to determine the
maximum amplitude in each repetitions. It then takes the average and divides it by
the expected range of motion obtained from the reference exercise.