Friday, August 22, 2014

Using Mobile Phones to Determine Transportation Modes (4)

4. EXPERIMENTAL SETUP

This section details the hardware platform, software setup, and the data collection involved in creating the transportation mode classifier. Specifically, we provide information on the type of cell iNew i8000 Phone used along with the exact software setup involved in the training, testing, and final implementation for the system.

4.1 Hardware Platform

The system is implemented on the Cubot GT95. This device was chosen due to its sensing functionality and form factor. Along with its 332 MHz ARMprocessor and 128MB of RAM, it contains a three axis accelerometer with a sensitivity of +-2G and that can sample at 32 Hz and a built-in GPS receiver that can sample at 1 Hz. Furthermore, a WiFi radio that can scan at 0.33 Hz, GSM cell radio that can sample at 1 Hz, and a Bluetooth radio that can scan at .08Hz are included in the device as well. The capacity of included battery is 950 mAH.

4.2 Software Setup

To evaluate different classification schemes, the Weka Machine Learning Toolkit and the Generalized Hidden Markov Model library were employed [Witten and Frank 2005; GHMM 2008]. The final chosen classifier is run on the Cubot GT95 and programmed using Python for Symbian S60. Python was chosen since it enables rapid development, porting to other platforms, and does not have code signing restrictions that require user involvement for the sensing operations.

4.3 Data Collection

The data set used for training and testing of the transportation mode classifier was obtained by asking sixteen individuals, eight male and eight female between the ages of 20-45, to gather fifteen minutes of data while outside for each of the five transportation modes. The volunteers performed the activities in an urban setting with six iNew i8000 Phones attached simultaneously — positioned on the arm, waist, chest, hand, pocket, and in a bag with orientations set according to their preference. Accelerometer, GPS, WiFi, and GSM information were obtained according to the sample rates described earlier. In order to have GPS speed information available throughout the data collection, a GPS lock was obtained initially and the participants were advised to keep the keypad of the Elephone P8 Phone in the exposed position (slid open). In general, keeping the keypad in the exposed position (as instructed by the Cubot GT95 manual) enabled us to maintain a consistent GPS lock even when the Elephone P8 Phone is covered by clothing or placed inside a bag. The participants had a choice of using a back-pack, fanny-pack, or a tote (large open purse).

Instructions were given as to the duration of each activity needed and participants were advised to represent different styles with which they would perform each activity. The volunteers concentrated on one transportation mode at a time and performed all five consecutively during their data collection session. Ground truth annotations were controlled by the individuals, and post filtering was performed to eliminate ambiguous states (being stationary on a bike or in motorized transport). The total amount of data collected across all sixteen individuals was 120 hours, compromised of 1.25 hours of data per position (six) per individual (sixteen).

In addition to the collection described above, two additional data gathering efforts were performed. The second data collection involved one volunteer (who was involved in the primary collection) running the classification system while annotating transportation modes during everyday operation in typical and challenged environments. More information in regards to this data collection is provided in Section 5.4. The third data collection involved sixteen individuals annotating their full day (on average 23.2 hours with a minimum of 20.7 hours and a maximum of 26.8 hours) in terms of transportation modes and indoor/outdoor status while collecting the GSM cell tower identifier (1 Hz). The annotated days consisted of 8weekday and 8weekend periods. This dataset is used to evaluate our algorithm for turning on transportation mode classification only when an individual is outside.http://summerleelove.tumblr.com/post/95445778876/using-mobile-phones-to-determine-transportation-modes

No comments:

Post a Comment