=>

COLLECTING DATA

KASS: the sample data can be recorded in two ways: wearable sensors and
smartphone-based sensors. Furthermore wearable sensors can be divided into two categories:
multi-accelerometer sensors and single accelerometer sensor (33 pages doc)
1)METHOD : 2)NAMES : 3)TYPES : 4)MORE INFO : 5)PROBLEM& Results
1)ON USER DIRECTLY ;  :  biaxial accelerometers
worn on the 4 limb positions and right hip simultaneously to collect user’s activity data.
2) Bao and Intille
3) ACCELEROMETR
4) data were sent to a mobile computing device to perform classification
using Machine Learning Algorithms: decision table, nearest neighbour and naive Bayes
5) More sensors benefit the accuracy rate, it requires heavy
computation and power consumption
Khan et al.
[9]
implemented the physical activity recognition using one single tri-axial
accelerometer with sampling frequency of 20 Hz and a data-window of 3.2 seconds without
overlap. They also proposed a novel hierarchical recognition scheme which is capable of
recognizing 15 physical activities of daily life. However the single accelerometer sensor is a
special tailored design mainly for laboratory test. The wearable sensors are not practical
for real-world application.
=> Reconize  15 Physical activities.
The fact that smartphones are ubiquitous and embedded with build-in accelerometers make
it an ideal device for monitoring
Kwapisz et al. used smartphone-based accelerometers to detect 6 activities using 43 features with high accuracy. The data were collected in a custom-build Android application WISDM (Wireless Sensor Data  Mining) with sampling frequency of 20 Hz and data-window at 10 seconds with phone in pocket. It is proposed that either minimize the intelligence needed on the phone or completing the activity recognition model directly on the phones to save computational work and power consumption
However the features are computed over the long data-window which reduce the chance of
capturing the transition movements