The engine thrust and aerodynamic forces are highly coupled in vertical recovery rocket landing, and the rocket feasibility and safety depend on not only the trajectory terminal position, velocity and attitude deviation, but also the initial attitude deviation and the control accuracy of the landing energy. Aiming at the complex dynamic model trajectory planning with aerodynamic changes under the influence of the engine starting section and jet stream, the segmented trajectory online planning is conducted by adopting the polynomial guidance method in the starting section and the analytical initial value-sequential convex optimization method in the thrust throttling adjustment section. To eliminate the jumps of the switching attitude angle caused by the difference of segmented solving models, a quintuple polynomial form was used to obtain the analytical initial value reference profile with the constraints of the initial attitude angle. Then, through the dynamic adjustment of the thrust angle and trajectory tilt angle constraints as well as the introduction of a deviation relaxation term, the terminal mass constraint was added to the optimization performance index function, and the optimization problem with constraints on the terminal position, velocity, attitude, and specified mass was completed using the sequence-convex iterative method. The results show that the proposed method has advantages in accuracy index and number of constraints compared with the traditional polynomial or convex optimization methods. Capable of achieving accurate terminal position, velocity and terminal mass constraints under the combination of deviation conditions, with the terminal attitude angle deviation less than 1.5°, and the average time consumed for online planning less than 1 s, it has a good engineering application prospect and can provide reference for the development of China’s new generation of reusable manned launch vehicles.