Real-time monocular object SLAM

Abstract

We present a real-time object-based SLAM system that leverages the largest object database to date. Our approach comprises two main components: 1) a monocular SLAM algorithm that exploits object rigidity constraints to improve the map and find its real scale, and 2) a novel object recognition algorithm based on bags of binary words, which provides live detections with a database of 500 3D objects. The two components work together and benefit each other: the SLAM algorithm accumulates information from the observations of the objects, anchors object features to especial map landmarks and sets constrains on the optimization. At the same time, objects partially or fully located within the map are used as a prior to guide the recognition algorithm, achieving higher recall. We evaluate our proposal on five real environments showing improvements on the accuracy of the map and efficiency with respect to other state-of-the-art techniques.

On-patient see-through augmented reality based on visual SLAM

“…The practitioner selects the 2D anchor points over the live video stream by tapping on the tactile screen of the tablet-PC. The 3D coordinates of the 2D anchor points are computed and appended to the map following the procedure described in Algorithm 1 (a variant of [26]). The P3DM is then translated, rotated and scaled to align the landmarks in the model with the anchors in the map.…”

Semi-semantic Line-Cluster Assisted Monocular SLAM for Indoor Environments

“…Some attempts have been made to integrate object recognition results into the SLAM system [6]- [9], so the scale information can be obtained from known objects. These methods require training a detector or classifier ahead, and assume that particular objects can be encountered by the robot, which does not usually occur in a real application scenario.…”

Object-aware bundle adjustment for correcting monocular scale drift

“…Two further works that make use of objects to improve map accuracy are [30] and [31]. Bao et al [30] jointly estimate camera, point and rectangular object positions in a bundle adjustment, however the focus of their work is on enhancing 3D reconstruction and object recognition within offline structure from motion.…”

“…Bao et al [30] jointly estimate camera, point and rectangular object positions in a bundle adjustment, however the focus of their work is on enhancing 3D reconstruction and object recognition within offline structure from motion. Gálvez-López et al [31] recently proposed a method that adds objects modelled from point clouds to the map in which known distances between object-points are used for adding additional geometrical constraints to a bundle adjustment and enforcing scale in the map. A database of 500 objects is used, although the method is unable to extend to general object classes.…”