On using regression for range data fusion

We consider an occupancy based approach for range data fusion, as it is used in mobile robotics. We identify two major problems of this approach. The first problem deals with the combination rule which in many cases assumes the independence of range data, contrary to the usual situation. The second problem concerns the redundancy of stored and processed data, which results from using the grid representation of the occupancy function and which is the main obstacle to building 3D occupancy world models. We propose a solution to these problems by proposing a new range data fusion technique based on regression. This technique uses the evidence theory in assigning occupancy values, which we argue is advantageous for fusion, and builds the occupancy function by fitting the sample data provided by a sensor with a piecewise linear function. Having developed a general framework for our approach, we apply it to building 3D occupancy models from visual range data, where the models are used for navigating a robot in an unknown environment.