Scene analysis using fusion of range and color data

In this paper, we present a segmentation algorithm based on 3D-color (3DC) images acquired with a calibrated heterogeneous sensor composed of a video camera and a laser range finder. The first part of the paper is dedicated to the presentation of the multi-sensor system and the specific calibration process which allows it to register color and range data in a unique frame. The next part describes the region growing algorithm used for the 3DC image segmentation. The color criterion used to select homogeneous pixels is defined in a perceptual representation space. A 3D criterion is used to select voxels which belong to the same planar face in the observed scene. We show how these two criterion may be combined to segment the 3DC image into planar faces which are characterized by homogeneous colors. Results obtained with real dense 3DC images illustrate the performance of this method.