Deep Metric Learning-Assisted 3D Audio-Visual Speaker Tracking via Two-Layer Particle Filter

Table 1

MAE in 3D (m) and on the image plane (pixels) on seq08, seq11, and seq12 over cameras 1, 2, and 3.

Sequence		3D MAE (m)			2D MAE (pixels)
Seq	Camera	[21]	[12]	Ours	[5]	[12]	Ours
08	1	0.15	0.12	0.08	10.75	4.31	3.24
	2	0.24	0.11	0.07	7.33	4.66	3.11
	3	0.20	0.09	0.07	9.85	5.34	3.28
11	1	0.31	0.33	0.23	14.66	8.15	6.04
	2	0.29	0.14	0.07	14.01	7.48	5.13
	3	0.26	0.12	0.08	13.96	6.64	4.06
12	1	0.41	0.26	0.18	12.49	6.86	4.15
	2	0.51	0.17	0.10	10.81	10.67	5.19
	3	0.47	0.20	0.13	11.86	9.71	5.58
Average		0.32	0.17	0.11	11.75	7.09	4.42