As it stands, the second definition is a concrete description of the spin group in dimension four. It defines an action of the simply connected group $SU(2)\times SU(2)$ on a vector space of real dimension four, which preserves a positive definite inner product, and this identifies $SU(2)\times SU(2)$ with the universal covering of the special orthogonal group of this four-dimensional Euclidean space.
To view it as a definition of a spin-structure on a manifold, one has to do this in each point of the manifold. This means that the spin strucuture in this sense is given by two auxiliary complex rank two bundles $S^+$ and $S^-$ which are endowed with Hermitian bundle metrics and compatible complex symplectic forms, togehter with an isomorphism between the tangent bundle and the bundle $Hom_J(S^+,S^-)$ which respects the inner products on the two spaces (the given Riemannian metric and the inner product constructed point-wise as in definition 2).
The equivalence between the two definitions is obtained as follows: To go from definition 1 to definition 2, you form the associated bundles corresponding to the two basic (complex) spin-representations of $Spin(n)$. In the opposite direction, you form the frame bundle of $S^+\oplus S^-$ (with structure group $SU(2)\times SU(2)$, which is isomorphic to $Spin(n)$) and the identification of $Hom_J(S^+,S^-)$ with $TM$ preserving bundle metrics shows that this bundle is a two fold covering of the $SO(n)$-frame bundle associated to the Riemannian metric.