Surface reconstruction using high resolution stereo vision in a micro air vehicle

The autonomous operation of an unmanned ground vehicle requires map information to be able to plan its route efficiently. It is possible to obtain information from the vicinity of the machine but ground based machinery is not ideal for quickly mapping large areas of unknown terrain. The aim of this thesis is to make a proof of concept implementation of an aerial system that provides point cloud data to be used for mapping areas. The focus is in areas which do not have many distinctive features, such as dirt fields or forest. Stereo vision is not a new concept but many of the existing methods cannot process high resolution data or are utilizing a small part of the available information. Stereo cameras are constructed from two consumer grade digital cameras which are mounted below the hexacopter. Cameras are placed close to each other because of the payload and size constrains, this makes the stereo camera more compact but reduces the accuracy of the stereo vision. The algorithm developed in this thesis uses correlation based block matching to determine the corresponding features from the stereo pair. The algorithm is able to find details from low feature surfaces when the altitude is below 5m. Useful data can be obtained from altitudes up to 30 − 40m. The largest errors are caused by the consumer grade cameras having inaccurate triggering. This thesis serves as a good starting point for developing more effcient hardware and software for small aerial vehicles. Many of the problems encountered in this thesis can be mitigated with different hardware