[Dataset] umich/rss (v. 2011-08-10)

top

the initial version

@MISC{umich-rss-2011-08-10,
  author = {Neal Patwari and Alfred O. Hero III and Kumar Sricharan},
  title = {{CRAWDAD} data set umich/rss (v. 2011-08-10)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/umich/rss},
  month = aug,  
  year = 2011
}
					

This is a dataset of RSS measurements collected by Mica2 sensor nodes deployed 
inside and outside a lab room, with anomaly patterns occurring when students
walked into and out of the lab. A web camera recorded the activity that could 
be matched with detected anomalies.

The experiment was set up on a Mica2 platform, which consisted of 14 sensor 
nodes randomly deployed inside and outside a lab room. During the measuring 
period, students walked into and out of lab at random times, which caused 
anomaly patterns in the RSSI measurements. Finally, a web camera was employed 
to record activity for ground truth.

Wireless sensors communicate with each other by broadcasting and the received 
signal strength (RSS), defined as the voltage measured by a receiver’s 
received signal strength indicator circuit (RSSI), was recorded for each pair 
of transmitting and receiving nodes.

The experiment was conducted by Prof. Neal Patwari of the University of Utah
in the Winter of 2006, when he was a post-doctoral student at the University of
Michigan under the guidance of Prof. Alfred O. Hero III. The site of the 
experiment was the 4th floor of the EECS building, University of Michigan, Ann 
Arbor.

There were 14 × 13 = 182 pairs of RSSI measurements over a 30 minute period, 
and each sample was acquired every 0.5 sec.

[Traceset] umich/rss/sensor (v. 2011-08-10)

top

the initial version.

@MISC{umich-rss-sensor-2011-08-10,
  author = {Neal Patwari and Alfred O. Hero III and Kumar Sricharan},
  title = {{CRAWDAD} trace set umich/rss/sensor (v. 2011-08-10)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/umich/rss/sensor},
  month = aug,  
  year = 2011
}
					

This is a traceset of RSS measurements collected by Mica2 sensor nodes deployed 
inside and outside a lab room, with anomaly patterns occurring when students
walked into and out of the lab. A web camera recorded the activity that could 
be matched with detected anomalies. The mission of this experiment was to use 
the RSS sequences to detect any intruders (anomalies). The ground truth 
indicator (camera recordings) can be used for evaluating the detection 
performance.

Wireless sensors communicated by broadcasting and recorded RSS for each pair of
transmitting and receiving nodes every 0.5 sec. A web camera recorded activity
for ground truth.

The original raw data is not synchronous. This is corrected for by using
interpolations. Data was also pre-processed to remove temperature drifts.

[Trace] umich/rss/sensor/anomaly (v. 2011-08-10)

top

the initial version

@MISC{umich-rss-sensor-anomaly-2011-08-10,
  author = {Neal Patwari and Alfred O. Hero III and Kumar Sricharan},
  title = {{CRAWDAD} trace umich/rss/sensor/anomaly (v. 2011-08-10)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/umich/rss/sensor/anomaly},
  month = aug,  
  year = 2011
}
					

This is a trace of RSS measurements collected by sensor nodes, with anomalies
detected when students entered or left the room. The mission of this experiment
was to use the RSS sequences to detect any intruders (anomalies). The ground 
truth indicator can be used for evaluating the detection performance.

This experiment was set up on a Mica2 platform, which consisted of 14 sensor 
nodes randomly deployed inside and outside a lab room. Sensors communicated by
broadcasting every 0.5 sec.

rssdata.zip contains the original Matlab files and the preprocessing code
preprocess.m . 

rssdata.tar.gz contains the same data converted to css and the preprocessing
code preprocess.m.

The original raw data is stored in the matrix ‘dataLinear’ (of size 182 x 3191) in the file ‘dataLinear.mat’. The ground truth is recorded in the vector ‘motionCode’ (of size 1 x 3191) in the file ‘motionCode.mat’. A value of 1 in motionCode indicates the presence of an intruder.

Pre-processing

This original raw data is not synchronous. This is corrected for by using interpolations to give the modified data in the matrix ‘Y’ (of size 182 x 3127) in the file ‘Y.mat’. The corresponding ground truth is stored in the vector ‘motion’ (of size 1 x 3127) in the file ‘motion.mat’.

To remove the temperature drifts of receivers we pre-process the data by removing their local mean values. Let y_i[n] be the n-th sample of the i-th signal and denote

y[n] = (y_1[n], …, y_{182}[n])'.

Due to temperature drifts, certain trends exist in y[n]. We de-trend the data by

z[n] = y[n] – y_m[n]

yielding z[n] for anomaly detection, where

y_m[n] = (2m+1)^{-1} \sum_{i=n-m}^{n+m} y[n]

is the local mean value. We set m=50 in this experiment. The modified data ‘Z’ is stored in the matrix ‘Z’ (of size 182 x 3127) in the file ‘Z.mat’. This pre-processing was done in Matlab and the code used for doing this pre-processing is given in ‘preprocess.m’.

The original raw data is not synchronous. This is corrected for by using
interpolations. Data was also pre-processed to remove temperature drifts.

[Author] Neal Patwari

top

[Author] Alfred O. Hero III

top

[Author] Kumar Sricharan

top

[Paper] chen-robust

top