(1) Vision-based methods: such as “multi-camera video surveillance system” [3].
First of all, they have great limitations and cannot be popularized in every
household, which can only be applied to specific occasions. Secondly, the visual
product judges whether the fall is theoretically based on the appearance of the
human body falling, and the detection accuracy is difficult to be improved in
essence.
(2) Sensor-based methods have fewer features to judge first, which are generally
based on acceleration or angular velocity, or both. Such as the “wearable sensors
for reliable fall detection,” it just judges the falling by the threshold of acceleration
[4]. And “elderly fall monitoring method and device” [5] pointed out that using
acceleration and angle to monitor falling, and “fall detection analysis with
wearable MEMS-based sensors” [6] has consistent ideology with our paper to
some degree. However, it just proposed a method and the execution of method is
unknown.
Based on the survey of the above products or papers: If we want to put such
products into wide application, there are two main problems to be solved: The first is
the high accuracy with the balance between sensitivity and specificity, and the second
is household, which means simplicity and convenience. Considering that XGBoost has
higher accuracy and lower false-positive rate than other algorithm, it is a kind of
gradient boosting which has proven many times to be an effective prediction algorithm
for both classification and regression task, such as crude oil price forecasting [7], DDoS
attack detection [8], and so on. Thus, we decide to use XGBoost as the basis of
detection.
2 Proposed Falling Detection Based on XGBoost
2.1 The Process of Establishing the Model
The falling process includes:
(1) The horizontal acceleration relative to the heaven and earth coordinate system
increases: This phenomenon occurs because during the movement of the human
body, the horizontal acceleration increases and the speed decreases due to being
tripped.
(2) The X-axis angular velocity and Y-axis angular velocity relative to the heaven and
earth coordinate system increase.
(3) The vertical acceleration relative to the heaven and earth coordinate system
increases: This is caused by the interaction between human body and the ground.
(4) The static process after falling, during which the human body’s pitch angle will
reduce (that is, the included angle between the human body and the horizontal
ground becomes smaller).
Generally, unless the person faints after falling, it is difficult to detect this very
weak stationary process. Process (2) is a necessary condition for the fall to occur; that
is, it is considered to have fallen when the body is dumped. The occurrence of the
process (3) substantively causes harm to the human body, and the alarming of damage
Design of Elderly Fall Detection Based on XGBoost
47
First of all, they have great limitations and cannot be popularized in every
household, which can only be applied to specific occasions. Secondly, the visual
product judges whether the fall is theoretically based on the appearance of the
human body falling, and the detection accuracy is difficult to be improved in
essence.
(2) Sensor-based methods have fewer features to judge first, which are generally
based on acceleration or angular velocity, or both. Such as the “wearable sensors
for reliable fall detection,” it just judges the falling by the threshold of acceleration
[4]. And “elderly fall monitoring method and device” [5] pointed out that using
acceleration and angle to monitor falling, and “fall detection analysis with
wearable MEMS-based sensors” [6] has consistent ideology with our paper to
some degree. However, it just proposed a method and the execution of method is
unknown.
Based on the survey of the above products or papers: If we want to put such
products into wide application, there are two main problems to be solved: The first is
the high accuracy with the balance between sensitivity and specificity, and the second
is household, which means simplicity and convenience. Considering that XGBoost has
higher accuracy and lower false-positive rate than other algorithm, it is a kind of
gradient boosting which has proven many times to be an effective prediction algorithm
for both classification and regression task, such as crude oil price forecasting [7], DDoS
attack detection [8], and so on. Thus, we decide to use XGBoost as the basis of
detection.
2 Proposed Falling Detection Based on XGBoost
2.1 The Process of Establishing the Model
The falling process includes:
(1) The horizontal acceleration relative to the heaven and earth coordinate system
increases: This phenomenon occurs because during the movement of the human
body, the horizontal acceleration increases and the speed decreases due to being
tripped.
(2) The X-axis angular velocity and Y-axis angular velocity relative to the heaven and
earth coordinate system increase.
(3) The vertical acceleration relative to the heaven and earth coordinate system
increases: This is caused by the interaction between human body and the ground.
(4) The static process after falling, during which the human body’s pitch angle will
reduce (that is, the included angle between the human body and the horizontal
ground becomes smaller).
Generally, unless the person faints after falling, it is difficult to detect this very
weak stationary process. Process (2) is a necessary condition for the fall to occur; that
is, it is considered to have fallen when the body is dumped. The occurrence of the
process (3) substantively causes harm to the human body, and the alarming of damage
Design of Elderly Fall Detection Based on XGBoost
47