A perception-aware Nonlinear Model Predictive Control (NMPC) strategy aimed at performing vision-based target tracking and collision avoidance with a multi-rotor aerial vehicle is presented in this paper. The proposed control strategy considers both realistic actuation limits at the torque level and visual perception constraints to enforce the visibility coverage of a target while complying with the mission objectives. Furthermore, the approach allows to safely navigate in a workspace area populated by dynamic obstacles with a ballistic motion. The formulation is meant to be generic and set upon a large class of multi-rotor vehicles that covers both coplanar designs like quadrotors as well as fully-actuated platforms with tilted propellers. The feasibility and effectiveness of the control strategy are demonstrated via closed-loop simulations achieved in MATLAB.