Linear Model Hashing and Batch RANSAC for Rapid and Accurate Object Recognition

Abstract

This paper proposes a joint feature-based model indexing and geometric constraint based alignment pipeline for efficient and accurate recognition of 3D objects from a large model database. Traditional approaches either first prune the model database using indexing without geometric alignment or directly perform recognition based alignment. The indexing based pruning methods without geometric constraints can miss the correct models under imperfections such as noise, clutter and obscurations. Alignment based verification methods have to linearly verify each model in the database and hence do not scale up. complicated problems, i.e., (i) the unknown pose between the query and the model, (ii) the query-model discrepancy due to occlusion and clutter in the scene, and (iii) the computational cost of comparing each individual model from the database to match against the query. Alignment based verification techniques (often based on RANSAC) have been used to address the first two problems. However, existing alignment based techniques apply RANSAC sequentially to each individual model from the database, and hence do not address the computational issues related to the third problem when matching to a large model database.