The figures in the following show the mass probability density of the localizzation error, when the radio tomography approach is used. This method is described in the paper Patwari & all; "See-Through Walls: Motion Traking Using Variance-Based Radio Tomography Networks"; IEEE Transaction On Mobile Computing, Vol. 10, Issue 5; May 2011.
==== Scenario C40 ====
In the following table, the VRTI dist. Error is the locallization error misured by the only Maximum Variance Image estimation,instead, Kalman dist. Error is the locallization error misured supporting the VRTI algorithm by the Kalman filter. Noise and No-Noise are the localization errors measured changing the parameters to calculate the weights of the regularization matrix.
^ Scenario ^ Parameters ^ Anchor positions ^ VRTI error dist. Noise/No-Noise . ^ KALMAN error dist. Noise/No-Noise ^ Metrics VRTI / KALMAN ^ Notes and video ^
^ C 40 | //area//: 6.4 X 4.5 [m]\\ //anchors//: 25\\ //speed//: 0.5[step/s]\\ //channels//: 1\\ //origin//: WnLab | {{c40_anchors.jpg?200}} | {{c40_error_mdf.jpg?200}} | {{c40_error_kalman_mdf.jpg?200}} | RMSE: 1.60 [m] / 1.59 [m]\\ 75th pc: 1.95 [m] / 1.93 [m] \\ 90th pc: 2.67 [m] / 2.62 [m] | Error with noise model in [0.5, 1.75] m.\\ Error without Noise model in [0.4, 2] m.\\ No improvement from tracking.\\ {{pat2_28sen_1p_1sec.avi|video}} |
^:::|:::|:::|{{c40_error_mdf_f1.jpg?200}}|{{c40_error_kalman_mdf_f1.jpg?200}}|RMSE: 2.55 [m] / 2.55 [m]\\ 75th pc: 3.28 [m] / 3.27 [m] \\ 90th pc: 3.78 [m] / 3.77 [m]|:::|
==== Scenario Pat_1 ====
In the following table, the VRTI dist. Error is the locallization error misured by the only Maximum Variance Image estimation,instead, Kalman dist. Error is the locallization error misured supporting the VRTI algorithm by the Kalman filter. Noise and No-Noise are the localization errors measured changing the parameters to calculate the weights of the regularization matrix.
^ Scenario ^ Parameters ^ Anchor positions ^ VRTI error dist. Noise/No-Noise . ^ KALMAN error dist. Noise/No-Noise ^ Metrics VRTI / KALMAN ^ Notes and video ^
^ Pat_1 | //area//: 8.4 X 6.5 [m]\\ //anchors//: 33\\ //speed//: 1[step/s]\\ //channels//: 5\\ //origin//: SPAN Lab. | {{pat_1_anchors.jpg?200}} | {{pat_1_error_mdf.jpg?200}} | {{pat_1_error_kalman_mdf.jpg?200}} | RMSE: 0.62 [m] / 0.62 [m]\\ 75th pc: 0.68 [m] / 0.67 [m] \\ 90th pc: 1.04 [m] / 1.02 [m] | Error with noise model in [0.2, 0.6] m. \\ Error without noise model in [1.1, 1.5] m. \\ No improvement from tracking.\\ {{pat2_28sen_1p_1sec.avi|video}} |
^:::|:::|:::|{{pat_1_error_mdf_f1.jpg?200|}}|{{pat_1_error_kalman_mdf_f1.jpg?200|}}|RMSE: 1.64 [m] / 1.62 [m]\\ 75th pc: 1.81 [m] / 1.80 [m] \\ 90th pc: 2.23 [m] / 2.23 [m]|:::|
==== Scenario Pat_2 ====
In the following table, the VRTI dist. Error is the locallization error misured by the only Maximum Variance Image estimation,instead, Kalman dist. Error is the locallization error misured supporting the VRTI algorithm by the Kalman filter. Noise and No-Noise are the localization errors measured changing the parameters to calculate the weights of the regularization matrix.
^ Scenario ^ Parameters ^ Anchor positions ^ Raw error dist. ^ Tracking error dist. ^ Metrics raw / trk ^ Notes and video ^
^ Pat_2 | //area//: 6.5x4.5[m]\\ //anchors//: 28\\ //speed//: 1[step/s]\\ //channels//: 1\\ //origin//: SPAN Lab. | {{pat_2_anchors.jpg?150}} | {{pat_2_error_mdf.jpg?150}} | {{pat_2_error_kalman_mdf.jpg?150}} | RMSE: 1.26 [m]/ 1.26[m]\\ 75th pc: 1.55 [m]/ 1.55 [m]\\ 90th pc: 2.0 [m] / 1.97 [m] | Error with noise model in [0.35, 1.7]m. \\ Error without Noise model in [0.7, 1.7] m. \\ No improvement from tracking.\\ {{pat2_28sen_1p_1sec.avi|video}} |
^:::|:::|:::|{{pat_2_error_mdf_f1.jpg?200}}|{{pat_2_error_kalman_mdf_f1.jpg?200}}|RMSE: 1.67 [m] / 1.66 [m]\\ 75th pc: 1.85 [m] / 1.83 [m] \\ 90th pc: 2.53[m] / 2.53 [m]|:::|
==== Scenario Runner ====
In the following table, the VRTI dist. Error is the locallization error misured by the only Maximum Variance Image estimation,instead, Kalman dist. Error is the locallization error misured supporting the VRTI algorithm by the Kalman filter. Noise and No-Noise are the localization errors measured changing the parameters to calculate the weights of the regularization matrix.
^ Scenario ^ Parameters ^ Anchor positions ^ Raw error dist. ^ Tracking error dist. ^ Metrics raw / trk ^ Notes and video ^
^ Runner | //area//: 6.5x4.5[m]\\ //anchors//: 28\\ //speed//: 0.5[step/s]\\ //channels//: 1\\ //origin//: SPAN Lab. | {{pat_2_anchors.jpg?150}} | {{runner.png?200|}} | {{runner_k.png?200|}}| RMSE: 3.056 [m]/ 3.051 [m]\\ 75th pc: 4.24 [m]/ 4.23 [m]\\ 90th pc: 5.60 [m] / 5.59 [m] | Error with noise model in [0.4, 1.8]m. \\ Error without Noise model in [1.2, 2] m. \\ No improvement from tracking.\\ {{runner.avi|video}} |
^:::|:::|:::|{{runner_phi_1.png?200|}}|{{runner_phi_1_k.png?200|}}|RMSE: 3.52 [m] / 3.51 [m]\\ 75th pc: 5.07 [m] / 5.09 [m] \\ 90th pc: 5.78[m] / 5.78 [m]|:::|
In [[1306_by_movement]], the VRTI dist. Error is the localization error evaluated using the Maximum Variance Image estimation. The metrics are evaluated with (first value) and without (second value) a WiFi AP. The same holds for the videos. Channels 17 and 24 were only slightly disturbed, while 14 and 21 were significantly disturbed by the WiFi AP.
====Localization Performance Vs Link Quality====
The following three figures show the RMSE, the 75-th and 90-th Percentile's resilience toward the noise that affects the cannel, respectivelly. Moreover, the figures show as the multi-channels improve the algorithm's performance rather than in the case of single-channel.
In brief, the three figures show the comparison about the localization performance in term of number of channels used to acquire the data, link quality and scenario taken into account.
^ RMSE ^ 75° Percentile ^ 90° Percentile ^
|{{rmse_vs_ap_int.png?200}}|{{75prc_vs_ap_int.png?200}}|{{90prc_vs_ap_int.png?200}}|
^ Channel ZigBee ^ S/N (dBm) ^
|12 | -180 |
|14 | -160 |
|16 | -140 |
|17 | -100 |
|20 | -160 |
|21 | -140 |
|24 | -60 |
|25 | -60 |