An Evaluation of Local Shape Descriptors in Probabilistic Volumetric Scenes

Abstract

This paper presents the first performance evaluation of local shape descriptors in probabilistic volumetric models (PVM) that are learned from multi-view aerial imagery of large scale urban scenes. The PVM offers a dense solution to the multi-view stereo problem, handling in a probabilistic manner the ambiguities caused by highly reflective surfaces, varying illumination conditions, registration errors, and sensor noise. A GPUbased octree implementation guarantees scalability of the PVM to large urban models, encouraging its application to higher level 3-d computer vision tasks. Furthermore, local descriptors form the basis of many shape-based applications and their performance has been studied extensively for images-based applications. Local descriptors have also been popular for 3-d shape understanding, but most 3-d descriptors have been used within the context of point clouds data collected with range sensors or polygonal meshes of CAD models. This work presents an investigation of the performance of several local shape descriptors in the PVM, which is learned from image data and where surfaces ambiguities are common and explicitly modeled. Descriptors are evaluated on accuracy for object classification using Bag-of-Words models. This evaluation forms a step towards the characterization of a new 3-d probabilistic representation for 3-d scene understanding.

Evaluation of 3D Feature Descriptors for Multi-modal Data Registration

“…We consider four 3D descriptors which directly operate on 3D point clouds rather than meshes or 2D projection. They are used for unimodal data classification in Restrepo and Mundy's work [9], here we evaluate their performance for multi-modal data registration.…”

“…3D lines and circular feature matching were used for registration. Restrepo and Mundy evaluated performance of various local 3D descriptors for models reconstructed from multiple images [9]. They reconstructed urban scenes with a probabilistic volumetric modelling method and applied different descriptors for object classification to find the best descriptor.…”

Big Data Analysis for Media Production

“…There have been a few researches for 2D-3D data matching and registration [24]- [26], but they have focused only on registration for a single data modality. 2D-3D registration between pairs of modalities such as photos to LIDAR [27], [28], spherical images to LIDAR [29], and images to range sensor [30], [31] have also been investigated.…”

Multi-modal big-data management for film production

“…The datasets exist in different domains with different formats and characteristics such as sampling resolution, accuracy and colour. There has been some prior research into 2D/3D data matching and registration [1,2], but they assume only a single modality case. In our previous work, we tested the performance of existing 3D feature descriptors for multi-model data registration [3] and proposed a way to combine descriptors in different domains for more robust feature matching [4].…”