Fracture structures in rock masses profoundly affect the stability of deep rock engineering, whereas conventional methods utilizing local surface fracture data on rock mass and disk models are limited in characterizing complex three-dimensional (3D) fractures. This raises two critical inquiries: How to enhance the comprehensiveness of fracture parameter collected on-site? And how to improve the accuracy of 3D fracture reconstruction? Hence, we proposed an approach for characterizing 3D fractures of rock mass by optimizing on-site acquisition techniques and discrete fracture network model. By integrating unmanned aerial vehicle (UAV) and borehole TV imaging techniques, we extracted comprehensive fracture parameters (e.g., trace length, orientation, and spacing) from the surface to the interior of the engineered rock mass. Leveraging the capability of the non-similar elliptical discrete fracture network model to more accurately represent fracture shapes, we introduced a dual-parameter correction method that integrates rotation angle and volume density adjustments into Monte Carlo simulation, resulting in an improved comprehensive parameter elliptical discrete fracture network (CPE-DFN) model. This model exhibited a substantial improvement in accuracy compared to previous disk models, achieving enhancements ranging from 11.2% to 24.9%. Finally, we validated the applicability and precision of this new method for 3D fracture characterization using comprehensive fracture data collected from a deep tunnel in western Sichuan, China, with an error of less than 8%. This work holds potential applications in investigating preferential flow paths as well as evaluating rock mass integrity and fractured zones in rock masses.