Resampling using Small-World Networks

  • image.avi: The figure shows the 30 noisy images fed to the filters. The noise standard deviation at each pixel of the 20x20 frame is 4.

  • imagetrack.avi: This figure shows the tracking estimate of various filters for the non-linear image sensing model. The target is present all the time. The blue line is the ground truth. The red solid line is the filter estimate. The magenta dots are the particles. The red solid line is the filter estimate, which is the mean of all the particles.

  • clutter.avi: This figure show the simulation of various filters for the 1D linear Gaussian model in a cluttered environment. The probability of missing a target is 0.2. The red solid line is the filter estimate which is the mean of the particles. The cyan circles are the observations (target detections as well as false alarms). The actual target position (is not shown in the video) moves in a straight line fashion from 15m to 82m in 80 seconds.

  • sparse_KS3.pdf: This figure analyses the number of particles required by our small-world networks based resamplers to achieve the same posterior approximation as achieved by the systematic resampler (which is the best so far). It can be observed our parallel-friendly and minimally-interacting small-world random network based resampler can achieve the same performance as the resource-hungry and parallel-unfriendly systematic resampler by using a few extra particles. For example, the performance obtained by the systematic resampler at 32 and 64 particles can be achieved by using our proposed resamplers at 64 and 128 particles respectively.