In this work, we propose a multi-constraint online planning framework designed for quadrotors to autonomously track randomly moving targets in complex environments. First, simple and effective target detection and motion prediction methods are used to obtain the target’s trajectory over the next period. Then, an intuitive enhanced perception path-finding method is proposed to provide a reliable initial solution for trajectory optimization. Furthermore, effective tracking constraints are deeply analyzed, and dynamic parameters are introduced. By comprehensively considering trajectory smoothness, obstacle avoidance requirements, and dynamic feasibility, a spatio-temporal optimization framework that can operate efficiently within milliseconds is established. Finally, simulations and practical experiments verify the effectiveness of our method.