The continuous advancement of imaging technology has spurred a growing demand for high-resolution, high-sensitivity, and compact image sensors. Traditional color filter-based image sensors, however, face critical limitations in signal-to-noise ratio (SNR) and efficiency, particularly at submicron pixel dimensions. The emergence of freeform metasurfaces offers a promising solution to overcome these limitations, enabling highly integrated and efficient optical components. In this work, we propose a metasurface-based color router that leverages multi-layer topological structures to split and focus visible light, rather than relying on conventional filtering approaches. This design achieves ultra-compact pixel dimensions while addressing efficiency constraints and wavelength crosstalk inherent in traditional architectures. To realize this, we employ a fully differentiable topology optimization framework, which allows for the design of multi-layer topological metasurfaces while accounting for manufacturing feasibility and robustness to fabrication errors. Through simulation and optimization, our color router demonstrates relative diffraction efficiencies of 69%, 82%, and 51% across three distinct wavelengths, operating at a pixel array size of 1.2μm×1.2μm. These results highlight significant improvements in energy utilization and reduced crosstalk between subpixel regions compared to traditional color filter-based image sensors. By advancing both energy efficiency and structural compactness, this design presents a novel solution for next-generation imaging technologies.