Kalman Filter-Based Fusion of LiDAR and Camera Data in Bird’s Eye View for Multi-Object Tracking in Autonomous Vehicles

. 2024 Dec 3;24(23):7718. doi: 10.3390/s24237718

Algorithm 1 CLF-BEVSORT Track Manager
1:	Data:
2:	Track list T = {t₀, t₁, …, t_k}
3:	Unconfident Track list U = {u₀, u₁, …, u_g}
4:	Lost Track list L = {l₀, l₁, …, l_h}
5:	Input:
6:	3D LiDAR detections D_3d = {d₀, d₁, …, d_n}
7:	2D Camera detections D_2d = {d₀, d₁, …, d_m}
8:	Output: T
9:	Initialize:
10:	T ← φ
11:	U ← φ
12:	L ← φ
13:	for frame f in F do
14:	Associate T^f_3d, D^f_3d based on GIoU(D^f_bev, T^f_bev)
15:	Update T_3d given matched(T_3d,D_3d)
16:	Associate T^f_2d, D^f_2d based on GIoU(D^f_2d, T^f_2d)
17:	Fuse T_3d with D_2d given matched(T_3d,D_2d)
18:	D^f_3d ← unmatched D^f_3d
19:	D^f_2d ← unmatched D^f_2d
20:	Associate L^f_3d, D^f_2d based on GWAC_Affinity(D^f_2d, L^f_2d)
21:	Recover L_3d given matched(L_3d, D_2d): recovered ∈ U
22:	/ from lost track stage /
23:	D^f_2d ← unmatched D^f_2d
24:	Associate U^f_3d, D^f_3d based on GIoU(D^f_bev, U^f_bev)
25:	Update U_3d given matched(U_3d, D_3d)
26:	Associate U^f_2d, D^f_2d based on GIoU(D^f_2d, U^f_2d)
27:	Fuse U_3d with D_2d given matched(U_3d, D_2d)
28:	Upgrade Fused U_3d into new tracks: new track ∈ T
29:	∀u: u ∈ U delete t if age ≥ max age
30:	∀l: l ∈ L delete t if age ≥ max age
31:	end for
32:	Return: T