TIDUE71D March   2018  – April 2020

 

  1.   Revision History

0.3.2.2.1 Test Results

See Table 11 for testing results.

Table 11. Test Results

Test Number Number of People GFR (%) GFR1 (%) GFR2 (%) MDR (%) FDR (%) Postional Error (m)
3 1 99.22 NA NA 1.24 0 0.31
4 1 99.84 NA NA 0.21 0 0.18
5 2 61.56 100 NA 24.97 0 0.18
6 2 97.50 100 NA 0.46 1.35 0.20
7 2 97.19 100 NA 1.10 0.70 0.23
8 2 99.84 100 NA 0.11 0.00 0.21
9 1 98.59 NA NA 2.33 0.00 0.24
10 5 29.06 52.81 100 31.85 0.19 0.29
11 1 99.06 NA NA 1.19 0.00 0.35
12 1 79.38 NA NA 23.57 0.00 0.35
13 3 40.16 62.19 78.28 47.31 0.82 0.34
14 4 33.44 78.28 99.53 24.68 0.00 0.25
15 5 79.69 87.19 100 8.02 0.00 0.22
16 6 41.41 60.63 69.06 25.16 0.00 0.28

From these results, we can see that the software performs well when the targets are moving. The static clutter removal algorithm discussed previously removes any noisy points generated from non-human objects that don't interest us. This leaves each person as a singular, isolated cluster of points. As we add more people to the scene, the tracker accuracy slowly decreases, due to complexities introduced in the point cloud when multiple people interact with eachother. The tracker performance is also poor when many people in the room are sitting, as they become static and are no longer detected by the point cloud. This can be rectified either through point cloud improvements, tracker improvements, or both. Otherwise, the people counting and tracking algorithm shows that the mmWave IWR6843 device is more than capable of tracking people in complex, indoor environments.