Panoramic Imaging


Introduction

We have been working on wide angle imaging since 1994. The technology we have developed allows ultra-wide angle imaging with over 330 degrees in elevation, and 360 degrees in azimuth.

The images are captured from reflective surfaces, thus eliminating the requirement for motorized pan/tilt units for imaging an entire room without the blindspots normally associated with fixed security cameras.


The technology

The device in its simplest form consists of a normal off-the-shelf camera facing the apex of a convex reflective surface. The shape of the mirror is critical in controlling the quality and usefulness of the image.

The geometrical properties of the reflective surfaces we have developed allow rapid and accurate image processing, minimizing the computation cost associate with non-perspective projection images. The mirror surfaces are designed such that angles on the image plane, scale to angles in elevation, or put simply, the resolution (pixels/degree) in the radial direction in the images is constant. This feature effectively halves the time required to filter and process these images.


Limits of performance

The ratio of elevational angle to image radial angle is a design parameter of the surfaces. Large ratios allow very large fields of view, with an obvious compromise between field of view and angular resolution within the image.

We have constructed devices with fields of view from 180 degrees (ratio of 6 degrees elevation to 1 degree image plane, camera field of view of 30 degrees) to 330 degrees (ratio of 11, camera field of view of 30 degrees) for various applications.

Download MPEG of large field of view  (or Low Res version)


Example applications


1. Virtual Pan/Tilt/Zoom

Our processing techniques and the unique geometrical features of the surfaces combine to provide efficient real-time, remapping of the panoramic image into perspective projection images of varying fields of view.

A virtual pan/tilt/zoom camera using an image from a panoramic surveillance camera. High Res MPEG  Low Res MPEG


2. Motion detection

As the panoramic camera does not need to move to image the entire room, it is computationally simple to detect changes in the image. The demonstration shows a simple image subtraction routine operating on an unwrapped image (polar coordinate image). Note that changes in the image are persistent, so the location of events (for example a shelf or clothes rack) can be seen after the fact.

Online motion detection and digital unwrapping at video rates. High Res MPEG  Low Res MPEG


3. Ranging

Techniques have been developed to allow range-finding from two stationary panoramic cameras, or from a single moving panoramic camera. The range algorithm is fast, with processing approaching video rates on standard PC hardware.

Proximity images (white=close, black=distant) computed from an image sequence of panoramic images moving at a known speed.


4. Navigation

Using range images, and additional signal processing, it is possible to control the trajectory of a robot through a complex terrain. The demonstration shows a robotic positioning system picking a path through a cluttered terrain, with no knowledge of the terrain structure, knowing only its speed, and the relative location of the target way point.

Download MPEG of terrain following robot

This behavior has autonomous vehicle applications, where navigation in a cluttered terrain is required. We are at the early stages of implementing the technology on an autonomous helicopter.

Download MPEG captured from hovering helicopter  (or Low Res version)