Spatial grouping of 3D points from multiple stereovision sensors

Abstract

This paper will present a method for grouping 3 0 points into cuboids. The 3 0 points are extracted using multiple stereovision sensors, and the sensor &ion module performs thejioion of the data sets and the grouping of the points in a single algorithm. The fusioidgrouping algorithm is scalable, being able to work using any number of sensors, including a single one. The grouping method relies on a method of transforming the 3 0 space so that the density of the points is kept constant, and all the points belonging to a single object are adjacent, making the grouping of points into cuboids a simple labeling problem.Another approach for 3D points grouping is presented in [3]. They use spatial coherence to identify regions 6om the depth information. Firsf they are looking for connected components in the 8-neighborhood of the image dimensions. In the depth dimension a neighboring pixel is connected if the difference in depth is less than a threshold. In the grouping process an object could be split into different disjoint regions. If two regions belong to the same object, they must be close to each other in 3D space. For each pair of regions a probability measure gives the likelihood that the regions belong to the same object.

Uncertainty analysis for multi-stereo 3d shape estimation

“…This approach relies on the hypothesis that, if a 3D point belongs to the object surface, its projection into the different cameras which really see it will be closely correlated. In [3] a method for spatial grouping of 3D points viewed by multiple stereo systems is presented. The grouping algorithm comprises a 3D space compressing step in order to map the 3D points into a space of even density that allows an easier grouping through a neighborhood approach; a subsequent uncompressing step preserves the adjacencies of the compressed space and helps the fusion of grouped points seen by different cameras.…”