t is well known that accurate aerial position and attitude control is beneficial for image orientation in airborne photogrammetry. The aerial control is traditionally obtained by Kalman filtering/smoothing inertial and GNSS observations prior to the bundle-adjustment. However, in Micro Aerial Vehicles this process may result in poor attitude determination due to the limited quality of the inertial sensors, large alignment uncertainty and residual correlations between sensor biases and initial attitude. We propose to include the raw inertial observations directly into the bundle-adjustment instead of as position and attitude weighted observations from a separate inertial/GNSS fusion step. The necessary observation models are derived in detail within the context of the so called “Dynamic Networks”. We examine different real world cases and we show that the proposed approach is superior to the established processing pipeline in challenging scenarios such as mapping in corridors and in areas where the reception of GNSS signals is denied.