Hough Transform and 3D SURF for Robust Three Dimensional Classification

Abstract

Most methods for the recognition of shape classes from 3D datasets focus on classifying clean, often manually generated models. However, 3D shapes obtained through acquisition techniques such as Structure-from-Motion or LIDAR scanning are noisy, clutter and holes. In that case global shape features-still dominating the 3D shape class recognition literature-are less appropriate. Inspired by 2D methods, recently researchers have started to work with local features. In keeping with this strand, we propose a new robust 3D shape classification method. It contains two main contributions. First, we extend a robust 2D feature descriptor, SURF, to be used in the context of 3D shapes. Second, we show how 3D shape class recognition can be improved by probabilistic Hough transform based methods, already popular in 2D.Through our experiments on partial shape retrieval, we show the power of the proposed 3D features. Their combination with the Hough transform yields superior results for class recognition on standard datasets. The potential for the applicability of such a method in classifying 3D obtained from Structure-from-Motion methods is promising, as we show in some initial experiments.

Robust Instance Recognition in Presence of Occlusion and Clutter

“…Recent methods [9,10,16,6] have begun to exploit the availability of cheap depth sensors to achieve success in instance recognition. Unlike earlier methods that relied on clean laser scanned data [14], these devices have encouraged research into the use of cheap depth data for real-time applications. However, these sensors are relatively noisy and contain missing depth values, thus making it difficult to extract reliable object shape information.…”

“…Popular approaches to occlusion based recognition mainly rely on a partsbased strategy where the object is divided into smaller parts which vote for the presence of an object [14]. However as demonstrated in Sec.…”

Scale-Invariant Vote-Based 3D Recognition and Registration from Point Clouds

“…In 3D, similar success stories have been reported with methods using local features [8,15,17,20,24,41]. These approaches can integrate information from a large number of object parts.…”

“…These arise in many vision tasks including 2D object detection [21,28,37], motion segmentation [39,40], and 3D shape registration and recognition [11,20,41]. These methods all share a common two stage framework: First they generate an empirical distribution of pose through the collation of a set of possible poses, or votes.…”

“…These methods all share a common two stage framework: First they generate an empirical distribution of pose through the collation of a set of possible poses, or votes. The votes are often computed by matching local features from a test object to those in a library with known pose [11,20,21,28,37,39,40,41], or by learning a function that maps features to votes [16,29]. The second step is then to find one or more "best" poses in the distribution (the maxima, in the case of ML/MAP estimation).…”