Vision-Based Detection of Low-Emission Sources in Suburban Areas Using Unmanned Aerial Vehicles

. 2023 Feb 16;23(4):2235. doi: 10.3390/s23042235

Algorithm 1 Motion objects contour processing
Input: $M_{m}$ —binary mask of objects in motion, $h, w$ —height and width of the image
Output: $C_{m}$ —list of final objects in motion

1:	functionfilterMotionMasks ( $M_{m}$ )
2:	$m i n \leftarrow h * w * 0.0001$ ▹ min. contour area
3:	$m a x \leftarrow h * w * 0.2$ ▹ max. contour area
4:	$m a x d i s t \leftarrow h * 0.05$ ▹ min. distance for contour merging
5:	`$c n t \leftarrow$ findContours( $M_{m}$ )` ▹ motion mask contours and object labeling
6:	`$c n t \leftarrow$ filterContoursByArea( $c n t$ , $m i n$ , $m a x$ )` ▹ filtering objects of extreme sizes
7:	`$c n t \leftarrow$ mergeCloseContours( $c n t$ , $m a x d i s t$ )` ▹ merging close objects using
	Euclidean distance
8:	$C_{m} \leftarrow []$
9:	for $c \in c n t$ do
10:	$c \leftarrow$ `convexHull(c)` ▹ convex hulls of objects
11:	`$C_{m}$ .append(c)`
12:	end for
13:	return $C_{m}$ ▹ filtered moving objects
14:	end function