The purpose of this project was to develope an algorithm for creating a 3D model of a vehicle given a set of still images (received from UVEye). The vehicle was photographed by an arc array of 12 cameras, located around the vehicle as the vehicle travels through the arc. We chose to solve the problem with structure from Motion with SIFT and Deepmatches features.
Images taken at different times show different scenes as the vehicle moves across a static background, so we built binary masks that separate the vehicle from the background. The challenging part was finding points of interest on the vehicle because it is mostly smooth and not characterized by texture. We built several models for different frames and combined them with the Coherent Point Drift algorithm after filtering out non-vehicle points.
The final model we received contains the front half of the vehicle. Building a complete model of the vehicle was not possible due to lack of strong features on the vehicle. We believe there are things that can be improved on the system, as well as other problem-solving methods that can be more effective.