In this paper, the real-time trajectory planning for autonomous agricultural vehicles is investigated, addressing the challenges posed by a shrinking labor force in agriculture. A real-time dynamic trajectory planning framework based on the Frenet coordinate system is proposed, aimed at enhancing efficiency and safety in obstacle avoidance within farm road scenarios. Initially, the algorithm utilizes the Frenet coordinate system to decouple vehicle motion into unconstrained lateral and longitudinal movements. Subsequently, trajectory costs are assessed using a stability cost function centered on Jerk and a safety cost function based on adaptive weights for global path and historical trajectory consistency. Together with feasibility checks, this method enables the selection of the optimal trajectory. Finally, the efficacy and innovation of the approach are demonstrated through both simulation and practical field tests.