This paper proposes a new method to enhance the impact resistance of earthquake shelters. Traditional methods for protecting individuals during earthquakes focus on improving or reinforcing the building structures, which is time-consuming and inflexible. The SHELTER project developed an earthquake shelter using a conventional cubic frame. This study builds upon the SHELTER project by optimizing its frame structure to improve the shelter's impact resistance against falling debris during an earthquake. Firstly, a right frustum frame is selected as the basic geometric configuration for optimization (with the cuboid being a particular case of the frustum). Six critical design parameters serve as variables for optimization, establishing a mathematical model. Secondly, to maximize the permissible impact loads in both vertical and horizontal directions (F1 and F2), a multi-objective optimization using the NSGA-II algorithm is performed on the selected parameters, resulting in a Pareto front. Furthermore, to choose a solution from the Pareto front that meets the needs of this study, the relationship between the forces F1 and F2 applied to the shelter frame by falling debris during an earthquake is examined, converting the multi-objective optimization problem into a single-objective optimization problem. Finally, based on the relationship between F1 and F2, a single-objective optimization function is established and solved using the Genetic Algorithm (GA) to obtain the optimal parameters. The shelter frame manufactured with these optimal parameters exhibits better overall impact resistance than the cubic frame. This improvement is attributed not only to the geometric properties of the frustum, which results in a lower centre of gravity, but also to the optimization method, which enhances the frame's vertical impact resistance while maintaining good horizontal impact resistance.