Radio frequency (RF) tomography for multi-target tracking

Radio frequency (RF) tomography is a recently developed method for multi-target localization and tracking. We deploy a network of radio frequency sensors on the periphery of the monitoring region. The region can have stationary non-metallic obstacles such as tables, chairs, etc. The sensors exchange signals among themselves and with a nearby monitoring station. An empty network measurement is recorded without any targets present. The signal strength is attenuated when targets are present within the network. This attenuation information is used to obtain position of the targets inside the network. Multiple sensor measurements allow us to make accurate location estimates. The measurements are modeled using the superpositional observation model. We are interested in tracking multiple targets with time-varying target number and in environments with detection challenges.

Real-time indoor tracking

Single target tracking

 

Two target tracking

Real-time estimation of target location (x markers) is displayed on the laptop screen for single target (left) and for two targets (right). The circles (o markers) represent locations of sensors which are mounted on verticle plastic stands with triangular base as seen in the video.

 


Related Publications :

  1. Nannuru, S.Y. LiY. ZengM. J. Coates, and B. Yang"Radio frequency tomography for passive indoor multi-target tracking", To appear, IEEE Trans. Mobile Computing, Dec 2013.
  2. Nannuru, S.Y. LiM. J. Coates, and B. Yang"Multi-target device-free tracking using radio frequency tomography", International Conference on Intelligent Sensors, Sensor Networks and Information Processing (ISSNIP), Dec. 2011.

 


Outdoor multi-target tracking

Experimental data is collected by deploying sensors nodes in a outdoor environment which has few obstructions. We demonstrate [2] simultaneous tracking of two and four targets. Different particle based multi-target tracking algorithms such as sequential importance sampling (SIR) filter, multiple particle filter (MPF) and Markov chain Monte Carlo (MCMC) filter are evaluated. An average accuracy of 0.22m for two targets and 0.63m for four targets is achieved for a sensor network of 24 nodes monitoring an area of approximately 50m2.

Real tracks and estimated tracks in an outdoor environment for two and four targets respectively.
Two target tracking     Four target tracks

 


Indoor multi-target tracking

We also work on multi-target tracking in an indoor environment [1]. Tracking is particularly challenging in indoor environment because of the multiple obstructions present in the signal path and the multi-path effects because of reflections from walls, ceilings, etc. We have collected data by deploying the sensors inside the Computer Networks Lab and Trottier building at McGill. The Lab is a considerably difficult environment for tracking because of the numerous desks and chairs present inside the monitored area and the Trottier building has a thick concrete pillar present within the network. We are able to track up to two moving targets inside an area of approximately 80m2 with an accuracy of 0.70m.

 

True target trajectories and estimated tracks in indoor environment for one target (Computer Networks Lab) and
two targets (Trottier building) respectively.
   

 


People