Multi-Camera Panorama Generation with Uncalibrated Camera

点击翻译按钮获取中文摘要

In this project, we deal with a scenario where there is a vehicle in which cameras are installed to record**panoramic video/images as the vehicle moves; i.e., images from the cameras are stitched together into a**wide-view (possibly 360 degree) panorama. Applications include civilian topography, e.g. Google Map cars, or**military reconnaissance. In such systems, when a camera fails, it must be replaced by a spare camera. The**problem is that the calibration process for the spare camera is slow and must be done by an expert operator in**controlled condition. This requires the vehicle to stop operations and return to base, reducing the capabilities of**the operator until the vehicle returns to duty. We propose to solve this problem by having a system whereby the**driver of the vehicle can quickly replace the failed camera by a spare camera, and our system would**automatically create the panorama from the uncalibrated spare camera, accurately enough for the operation to**continue. We propose to achieve this using machine learning. When the vehicle returns to base normally, a new**camera is then installed and calibrated by an expert.**