EDIT: We’re continuing to add more videos of and from the flier as we make them

Newer, Still Rather Cheap 5.8 GHz 720 CMOS Camera

Fighting winds in the park in Corona, Jan 25 (no onboard footage) – click here
Flight in the park, 17 Jan – click here
Nighttime flight with camera a 45degree view angle – click here
Evening flight over the building – click here
Nighttime flight with camera looking nadir – click here

Old, Super-Cheap 2.4 GHz Camera Test Camera

Long flight video – click here
Picture-in-picture from live camera during flight – click here
High-level flight over the building – click here
Nighttime flight – click here
Night roof flight 1 – click here
Night roof flight 2 – click here
Sunset onboard flight – click here
Park Flight 1 – click here
Park Flight 2 – click here
Park Flight 3 – click here
Parking Lot – click here

The purpose of the demo is to provide a mock-up of a customer-end product.  The goal of the Ashima Devices system is to combine reasonably well-known components of 3D reconstruction from 2D imagery into a seamless system that ends with an intuitive and easy-to-use web interface for exploration of the resultant 3D models.  While the component pieces are not new, an affordable, high-quality end-to-end system is.  Ashima Devices envisions potential markets in venue, hotel, and realty marketing, as well as in field of augmented reality and custom video game design.

The processes of creating the explorable 3D model starts with photography of the target building or object.  Multiple views of the same target from different locations and angles, combined with “ephemeris” data for the camera allows for the mapping 2D image points to 3D world points.  From there, objects can be reconstructed in a number of different ways.  With sparse matching, planes can be extracted and associated with model polygons for rendering.  With dense matching, individual “voxels” can be extracted and more complex shapes reconstructed.

The current 3D demo results from a limited subset of this full processing chain.  Projection matrices have been derived from a limited number of known 2D-3D correspondences.   Taking advantage of the mostly planar structure of our building, these projection matrices have then been used to find the image texture data for a small number of polygons that form the building shape.  The polygons define the model shape and the images are reprojected and sampled to yield the textures.  You can see artifacts resulting from this simplified approach in the squashed light posts in the car park, for example.