Abstract: To address the problems of moving state modeling and path optimization for the folk activity “Bench Dragon”, this study proposes a recursive modeling method based on planar geometric analysis and kinematic constraints. A recursive model for handle positions and velocities is first constructed, and then a turning path length optimization model incorporating collision constraints is developed. Based on the recursive model, the motion parameters of handles at each time step are obtained. A collision time estimation model is constructed through geometric relationship analysis and coordinate system transformation. The constraints of the model are determined based on the conditions that paths do not intersect, no collisions occur, and the dragon head does not move backward. A multi-search algorithm is designed to achieve precise determination of collision times. Under the given helix pitch and turning space constraints, the approximate shortest turning path for the “bench dragon” is computed as 11.7915 m using a genetic algorithm. The results demonstrate that when the velocity of each handle is required not to exceed 2 m/s under the optimized path, the dragon head velocity must be constrained to no more than 0.9670 m/s, providing a quantitative basis for safety control.