Algorithm 1 our hierarchical data association algorithm
Input:	Current tth frame and previous trajectories
	The detection bounding boxes computed from the detector described in Section 3.3
	The tag-boxes computed from the tracker described in Section 3.4
Output:	bounding boxes and trajectories for the tth frame
	Procedure
1	associate the detection bounding boxes $D{B}_t$ to the tag-boxes $T{B}_t$
2	If each $D{b}_t^j$ in $D{B}_t$ is assigned to the corresponding $T{b}_t^i$ in $T{B}_t$ with one-to-one manner
3	Return the $D{B}_t$ as tracked bounding boxes $D{B}_t{}^{•}$ and update the tracklets
4	else
5	Update the tracklets related to the $D{B}_t{}^{•}$ and return the assigned detection boxes ($D{B}_t{}^{•}$), unassigned detection boxes ($D{B}_t{}^{*}$) and unassigned tag-boxes ($T{B}_t{}^{*}$).
6	end if
7	for each tag-box $T{b}_t^i$ in the unassigned tag-boxes $T{B}_t{}^{*}$
8	set a default box ($D{b}_{t-l}^{•i}$) to the tag-box $T{b}_t^i$ according to the updated tracklets;
9	associate the unsigned detection boxes ($D{B}_t{}^{*}$) to the default box ($D{b}_{t-l}^{•i}$)
10	if find a best matched box in the $D{B}_t{}^{*}$
11	set the best matched box as the tracked bounding box and update the tracklet
12	else
13	set the default box as the tracked bounding box and update the tracklet,
14	end if
15	associate the unassigned tag-boxes ($T{B}_t{}^{*}$) to the default box ($D{b}_{t-l}^{•i}$)
16	if no matched tag-box is founded
17	set a counter array (age), age[$T{b}_t^i$] = age[$T{b}_t^i$] + 1
18	else
19	age[$T{b}_t^i$] = age[$T{b}_t^i$] − 1
20	end if
21	if age[$T{b}_t^i$] > threshold value (T)
22	Initialize the tag-box $T{b}_t^i$ and reset age[$T{b}_t^i$] = 0;
23	end if
24	associate the assigned detection boxes ($D{B}_t{}^{•}$) to the unassigned tag-box $T{b}_t^i$
25	if the assigned detection box has more than one tag-boxes
26	pend the tag-box $T{b}_t^i$
27	end if
28	end for