Download Project (9.3 MB)
We present a 3D reconstruction technique designed to support an autonomously navigated unmanned aerial system (UAS). The algorithm presented focuses on the 3D reconstruction of a scene using images from a single moving camera and can be used to construct a point cloud model of unknown areas. The reconstruction process, resulting in a point cloud model is computed using a feature point matching process and depth triangulation analysis, is a six step process. The first step is feature extraction from each frame of video; a neighborhood-magnitude-direction dependent matching procedure is applied to track feature points through subsequent frames. The distance a feature point travels, in pixels, becomes the feature disparity which can be translated into depth. The Cartesian depth coordinate, in the z direction, is determined using the disparity values, while the x and y coordinates are determined using the focal length information of the camera. The process consists of determining the size of the image at a particular depth and computing the width and height, x and y directions, for each feature point. The final output is a point cloud, a collection of points accurately positioned within a model. With enough points, surfaces and textures can be added to create a realistic model. An autonomous navigation control system utilizes the resulting visually reconstructed scene, centered at the current camera location, to either register its position within a known 3D model, or for obstacle avoidance and area exploration while mapping an unknown environment. The presented reconstruction algorithm forms a foundation for computer vision self-positioning techniques within a known environment without the use of GPS or any other sensor. The suitability of the reconstruction for mapping tasks is to be evaluated using ground-truth measurements of actual objects.
Vijayan K. Asari
Primary Advisor's Department
Electrical and Computer Engineering
Stander Symposium poster
"Automatic Scene Rendering for Unmanned Aerial Systems" (2013). Stander Symposium Projects. 248.