Simultaneous optimization of structural topology and material orientation is effective for maximizing the performance of anisotropic materials. However, parameterizing 3D orientation fields for such optimizationremains a significant challenge. Kubalak et al. [1] pointed out that Euler angle-based methods suffer from numerical singularities and poor convergence, and employed quaternions to resolve these singularities. Nevertheless, similar to Euler angles, even quaternion-based approaches fail to address the incompatibility between material $\pi$-periodicity and angular $2\pi$-periodicity, which prevents the direct application of spatial filtering. Although Jantos et al. [2] achieved spatially continuous fields via elasticity tensor filtering, Mokhtarzadeh et al. [3] suggested that this mismatch could potentially compromise the stability of numerical analysis methods, such as isogeometric analysis. Tensor field parameterization is considered an effective approach to fundamentally resolve these issues. Despite this potential, existing tensor-based methods [4] often yield optimization results that violate the invariant constraints required for physical validity. To comprehensively address these issues, this study proposes a novel optimization framework utilizing a tensor field as the design variable. By incorporating eigenvalue decomposition after standard Helmholtz filtering, our method can ensure strict satisfaction of invariant constraints. Furthermore, we introduce eigenvalue constraints to suppress eigenvalue degeneracy and control the excessive degrees of freedom of the design variables, alongside a design space relaxation scheme to prevent stagnation in local minima characteristic of orientation optimization. We present 3D numerical examples to demonstrate the effectiveness of the proposed method.