$\phi* = \phi_{ref} + \frac
	 {\sum_{x,y} angDiff (\phi(x,y), \phi_{ref}) * (\tau, \delta, tMHI_{\delta}(x,y))}
	 {\sum_{x,y}norm(\tau, \delta, tMHI_{\delta}(x,y)) }$

where $\phi*$ is the global motion orientation, $\phi_{ref}$is the maximum orientation of the pixels in the segment
, $f(x,y)$ is the motion orientation map found
from gradient convolutions with the Sobel kernels described above, $norm(\tau, \delta,tMHI_{\delta (x, y))}
is a normalized $tMHI$ value (linearly normalizing the $tMHI$ from 0-1 using the current time-stamp $\tau$ and duration $\delta$ ), and $angDiff (f(x,y),\phi_{ref})$ is the minimum, signed angular difference of an orientation from the reference angle. A histogram-based reference angle ($\phi_{ref}$) is required due to problems associated with
averaging circular distance measurements