Deep Learning-Based Real-Time Multiple-Object Detection and Tracking from Aerial Imagery via a Flying Robot with GPU-Based Embedded Devices

. 2019 Jul 31;19(15):3371. doi: 10.3390/s19153371

Algorithm 2. Following target.
Input: Class for person $L = 1$ , area of the bounding box $A = h e i g h t \times w e i g h t$ , maximum area $A_{m a x}$ = constant, center of the whole image $C_{i m a g e} = constast$
Output: $E r r o r_{C e n t e r} = C_{i m a g e} - C_{b o u n d i n g}$ , $E r r o r_{A r e a} = A_{m a x} - A$
1:	Look for target $N$
2:	if $N = 1$ then
3:	if $L = 1$ then
4:	Calculate $A$ and $C_{b o u n d i n g}$
5:	$E r r o r_{C e n t e r} = C_{i m a g e} - C_{b o u n d i n g}$
6:	$E r r o r_{A r e a} = A_{m a x} - A$
7:	Repeat Step 5 until $E r r o r_{C e n t e r} = 0$ , and use $E r r o r_{C e n t e r}$ for yaw angle
8:	Repeat Step 6 until $A_{m a x} \leq A$ , and use $E r r o r_{A r e a}$ for forwarding velocity
9:	end if
10:	else
11:	Do nothing
12:	end if