RESULTS: REAL-TIME ACQUISITION OF PERSPECTIVELY CORRECT IMAGES

Testing of the Homography-Based Module
Tests for the Panoramic Reprojection Module
Summary of Results

The goal of this activity is to provide existing machine vision techniques with transformed visual input, so that further image analysis is facilitated and system cost is reduced. Two such transformations have been developed.

The first operates on video provided by classical cameras and the goal is to transform any view to a top-down view. By achieving this transformation the problem of further image analysis for incident detection becomes a two-dimensional one, as opposed to the original three-dimensional problem. This task simplification is expected to reduce the work required for installing, calibrating and updating complex machine-vision based equipment and software for incident detection.

The second transformation is applied on video data provided by a 360 degrees panoramic camera. The goal of the transformation is to correct the geometric distortions that are due to the process of panoramic image acquisition, and transform the panoramic video to a form that can be used by existing machine vision techniques. Use of the panoramic camera can reduce the cost of equipment needed for machine vision based incident detection installation at sites.

Each of the developed transformations is implemented as a separate module that enables real time video transformation through a highly configurable user interface. In order to test the developed transformations a series of tests have been performed. For all tests, on-line data are used from the project sites and analysis of images is performed through off-line testing. This chapter describes the testing methodology, tests performed, results and performance obtained in these tests, for each transformation module.