Driver entry into and exit from a car using smartphone sensors

Three subjects conducted the experiments, of (a) IN- getting on the car seat, (b) SITTING - sitting there for 2-3 seconds, then (c) OUT - coming out from the car and (d) STANDING - standing out of the car for 2-3 seconds, with the smartphones in their left pocket, with the screen facing the thigh. This activity was performed for 241 times by the 3 subjects combined with the smartphone in the left pocket. i.e. time-series data with 50 samples per second was collected for 241 iterations of each of the four activities listed above. The raw (without any feature engineering) dataset has 15 columns (features) and 121,989 records. This dataset has been made public in two forms which are described below: a) Raw labeled data i.e. without applying any data smoothing or feature extraction. b) Labeled signals which are smoothened and containing all the primary and secondary feature vectors. Also, a fixed width sliding window of 1 sec with 50% overlap is applied on the dataset. The primary features like acceleration, gravity, orientation, linear acceleration, and rotation were recorded over the three axis at a sampling rate of 50Hz. One of the important secondary (derived) data feature derived out of the primary data is the Root Mean Squares (RMS) over the 3 axes for each primary category. RMS is well known engineered feature, and is independent of axis direction, hence making the activity detection independent of the phone orientation in the subject’s pocket. Besides RMS, other aggregation features like mean value, standard deviation, median absolute value, maximum and minimum values are derived from the primary data. Thus, the total number of features including both the primary and secondary data is 120. i.e. 6 aggregation functions (mean, median, minimum, maximum, standard deviation and mean absolute deviation) applied over 5 primary features like acceleration, gravity, orientation, linear acceleration, and rotation with each of these features having 4 components viz. x, y, z axis values and their root mean square. For Raw Labeled Data, as in a) above the legends for the features are as follows: accm - Accelerometer grvm - Gravity Sensor gyrm - Gyrometer lacm - linear Acceleration rotm - Rotation Activity- Labels ; any one out of - {IN, OUT, SITTING, STANDING }