Algorithm 1. long trajectory calibration scheme
1:	Input: online trajectory t and training trajectory $T=\left\{t_i,i=1,2,\cdots ,n\right\}$
2:	Output: user position $(x_i,y_i)$
3:	Get the direction $\theta$ and length $l$ of online trajectory t
4:	If the last step of t is turn then
5:	Select similar trajectory according to $\theta$ and $l$
6:	For each training trajectory T do
7:	Compute trajectory similarity degree $d_i$ between t and $t_i$ using DTW algorithm
8:	Result validation
9:	End for
10:	If have perfect result then
11:	Calibrate the result of single-step tracking algorithm
12:	End if
13:	End if
14:	If the last step of t is not turn, then
15:	Select similar trajectory according to $\theta$
16:	For each training trajectory T do
17:	Compute trajectory similarity $d_i$ between t and $t_i$ using subsequence DTW algorithm
18:	Result validation
19:	End for
20:	If have perfect match result then
21:	Calibrate the result of single-step tracking algorithm
22:	End if
23:	End if
24:	Return $(x_i,y_i)$