Generalized Hough transform for 3D object recognition and visualization in integral imaging

Abstract

Object recognition is an automated image processing application of great interest in areas ranging from defect inspection to robot vision. In this regard, the generalized Hough transform is a well-established technique for the recognition of geometrical features even when they are partially occluded or corrupted by noise. To extend the original algorithm—aimed at detecting 2D geometrical features out of single images—we propose the robust integral generalized Hough transform, which corresponds to transformation under the generalized Hough transform of an elemental image array obtained from a 3D scene under integral imaging capture. The proposed algorithm constitutes a robust approach to pattern recognition in 3D scenes that takes into account information obtained not only from the individual processing of each image of the array but also from the spatial restrictions arising from perspective shifts between images. The problem of global detection of a 3D object of given size, position, and orientation is then exchanged under the robust integral generalized Hough transform for a more easily solved maximum detection in an accumulation (Hough) space dual to the elemental image array of the scene. Detected objects can then be visualized following refocusing schemes of integral imaging. Validation experiments for the detection and visualization of partially occluded 3D objects are presented. To the best of our knowledge, this is the first implementation of the generalized Hough transform for 3D object detection in integral imaging.

An Innovative n-Shifted Sigmoid Channel and Spatial Attention Module for Efficient 3D Scene Object Detection

“…This paper introduces a n-shifted sigmoid channel and spatial attention module to deduce 3D bounding boxes of the objects in the scene and suggest object proposals from a point cloud focused 3D object detection. This model is based on recent advances in 3D deep learning models for point clouds and is inspired by both an innovative channel and spatial attention module and the generalized Hough voting process [13]. Table 1 describes the advantages of the proposed model when compared to the existing ones on feature extraction, attention benefits, predictions, and accuracy.…”