A Clustering Approach to Object Estimation, Featuring Image Filtering Prototyping for DBSCAN in Virtual Sets

In this paper we propose an innovative approach to object shape and position estimation by stereometric, data mining and interpolating techniques. Our proposed system would be able to work with real-time performances. Unfortunately the most computational expensive part of our solution is clustering. To achieve our goal we also propose a new prototypes estimation technique based on the application of fast image filters to the point set before clustering, stressing and adapting the filter concept to a non Image context. These filters can be easily computed on Graphic Processing Unit (GPU), reducing the data amount in the clustering phase, achieving fast execution performances as for the other stages of our system.