Micro Aerial Vehicles and especially multi rotors are gaining more and more attention for accomplishing complex tasks, considering their simple mechanical design and their versatile movement. MAVs are ideal candidates to perform tasks autonomously, to work safely in close proximity and in collaboration with humans, and to operate safely and effectively in natural human environments, like infrastructure inspection-maintenance, underground mine operations and surveillance missions. Adopting this vision, this thesis contributes in the aerial platform ecosystem that can be summarized by the term Aerial Robotic Worker (ARW). An ARW is characterized, among others, by its advanced capabilities on environmental perception and 3D reconstruction and active aerial manipulation.Using cameras for localization, mapping of an ARW as well as guidance on aerial manipulation is appealing mainly because of the small size and cost of such sensors. Nevertheless, visualinformation provided from the cameras is enormous, posing significant challenges in real-time data processing, while meeting the constraints of these platforms. Additionally, another challenge on visual perception considers the usage of multiple agents that collaboratively perceive their surroundings forming an aerial sensor. This thesis also investigates the applicability of visual SLAM algorithms in uncontrolled and cluttered environments. Furthermore, work will be presented on visual guidance for an aerial manipulator, which is challenging regarding the object detection, tracking and the platform approaching strategies. The first contribution will be the establishment of a flexible virtual stereo rig consisted of MPC controlled MAVs. The advantage of this approach is the varying baseline sensor that is composed from independently moving cameras, adjusting the depth perception accordingly. This method is able to provide the 3D reconstruction of the environment in a sparse pointcloud. The second contribution of this this thesis will examine the single agents in two different scenarios. Initially, experimental trials of commonly used visual sensors in hard and challenging environments will be presented in real scale underground ore mine to evaluate the localization and mapping performance of such technology for potential usage in UAVs. Secondly, theoretical work will be performed regarding attitude regulation of a hexacopter for stable hovering based on visual localization. In this work the time delays induced from the processing should be compensated with a switching control scheme which is able to maintain the stability of the platform. Finally, a third contribution of this thesis will be vision for aerial manipulation. The developed system includes a stereo camera that is attached on the end-effector of the aerial manipulator and is used to provide robust target detection and tracking. The visual feedback is processed to co-localize the aerial agent with the target and generate a waypoint that allows to approach the target.