Radiation Detection and Localization Testbed

212,765pages on
this wiki
Add New Page
Add New Page Discuss this page0

We have built a scale-down version of the testbed, called work bench. One low-dose (0.25microcurie) and three RFTrax sensors are placed on a table. Each sensor is connected to a iServer, which maps the serial port to an ethernet IP address. So the SensorNet node are polling reading from these three sensors.

One problem is that the radiation source is quite weak, so that the sensor readings are quite low. Only few (<5) particles per second if placing the sensor at zero distance. This will definitely affect the accuracy of the detection and localization algorithms. Nagi is probably going to bring this work bench to some labs that have access to higher-level radiation source, and collect the readings.

We have not tested the detection and localization algorithms yet. We still do not know their performance. Several statisitical values are proposed to detect the radiation source. But they all need some kind of detection threshold, and we have not figured out how to properly set them.

What we can do with the testbed

  • Measure the detection probability and false alarm probability.
  • Evaluate the localization algorithm
  • Anything related to detection time? One assumption we can make is that the closer to the source, the shorter detection time. This could be true even if the signal propagation time is neglectible (by speed of light). The argument is that the closer to the source, the higher intensity, and the less samples we need to achieve some conclusion within a specified confidence interval.

Sensor Placement Problem

  • Assume disk sensing area, whose radius could depends on either the detection probability requirement [Guanghui He and Jennifer Hou] or the detection time requirement (needs enough samples before detection).
  • Futher, use a more realistic objective function. Before reach any sensor, the people in affected area should be less than some threshold.
  • Decision fusion from multiple sensors. It can be modeled in a way similar to k-coverage. The more number of sensors covering a point, the more likely detecting an event happenning there.

Also on Fandom

Random wikia