Robot Partners - Collaborative Perceptual Robotic Systems

The Institute for Robotics and Intelligent Systems (IRIS) 



 Panoramic Recognition

M. Brown and D. Lowe

We have developed a novel system for fully automatic full-view panorama stitching. Unlike previous approaches, which require a predefined ordering or user input to identify matching images, our system uses object recognition (SIFT features) for the matching step. Next we perform a joint optimisation (bundle adjustment) to solve for the internal (focal length) and external (rotation) parameters of each camera. Multi-band blending is used to create seamless panoramic images without loss of high frequency detail. Furthermore, the system is able to recognise multiple panoramas in an image set, whilst ignoring noise images which are not part of a larger scene. This suggests a useful application for photographers: the system takes as input the images on an entire flash card or film, recognises images that form part of a panorama, and stitches them with no user input whatsoever. 

Panorama of the LCI Lab, generated automatically from 150 images