A brute force way to do this is to use rotation matrices. You can rewrite the Hadamard matrix as
\begin{align}
H &= \frac{1}{\sqrt{2}}\begin{pmatrix} 1 & 1 \\ 1 & \text{-}1 \end{pmatrix} \\
&= \frac{1}{\sqrt{2}}\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix} \begin{pmatrix} 1 & \text{-}1 \\ 1 & 1 \end{pmatrix} \\
&= R_x(\pi)R_y\left(\frac{\pi}{2}\right)
\end{align}
So its just the composition of two rotation matrices. If you write out the $SO(3)$ matrix representations for these rotations you get
\begin{align}
R_x(\pi)R_y\left(\frac{\pi}{2}\right) &\rightarrow \begin{pmatrix}
1 & 0 & 0 \\
0 & \cos \pi & \text{-}\sin \pi \\
0 & \sin \pi & \cos \pi
\end{pmatrix} \begin{pmatrix} \cos \frac{\pi}{2} & 0 & \sin \frac{\pi}{2}\\
0 & 1 & 0 \\
\text{-}\sin \frac{\pi}{2} & 0 & \cos \frac{\pi}{2}\end{pmatrix} \\
&= \begin{pmatrix}
0 & 0 & 1\\
0 & \text{-}1 & 0 \\
1 & 0 & 0\end{pmatrix}
\end{align}
You can compare this to the rotation matrix for an arbitrary 3D rotation matrix $R_{\vec{n}}(\theta)$ around a unit vector $\vec{n}$ (see Wikipedia for example) to get an overconstrained system of equations. However in this case its easy to confirm that $\vec{n} = (1, 0, 1)/\sqrt{2}$ and $\theta=\pi$ is the correct choice.