Visual SLAM algorithms: a survey from 2010 to 2016

Abstract

SLAM is an abbreviation for simultaneous localization and mapping, which is a technique for estimating sensor motion and reconstructing structure in an unknown environment. Especially, Simultaneous Localization and Mapping (SLAM) using cameras is referred to as visual SLAM (vSLAM) because it is based on visual information only. vSLAM can be used as a fundamental technology for various types of applications and has been discussed in the field of computer vision, augmented reality, and robotics in the literature. This paper aims to categorize and summarize recent vSLAM algorithms proposed in different research communities from both technical and historical points of views. Especially, we focus on vSLAM algorithms proposed mainly from 2010 to 2016 because major advance occurred in that period. The technical categories are summarized as follows: feature-based, direct, and RGB-D camera-based approaches.

How to Improve CNN-Based 6-DoF Camera Pose Estimation

“…The authors of PoseNet [9] leverage CNNs and transfer learning and propose a pure neural network based solution to 6-DoF camera pose estimation (i.e., 3D translation and 3D rotation) for a specific environment, addressing some limitations of traditional vSLAM algorithms [12]. However, the accuracy obtained with this architecture is still significantly below what can be obtained with vSLAM methods, especially if the latter are trained on full sequences.…”

Visual-Based Semantic SLAM with Landmarks for Large-Scale Outdoor Environment

“…Mur-Artal and Tards proposed ORB-SLAM [1], [2], which is one of visual SLAM systems with full sensor support and best performance, with applying ORB features in parallel tracking, mapping, and loop closure detection, and using pose graph optimization and bundle adjustment [13] based optimization. Another kind of visual SLAM systems, unlike feature-based methods mentioned above, directly uses images as input without any abstraction with descriptors or handcrafted feature detectors, called direct methods [14]. DTAM [15], in which tracking is implemented by associating the input image with synthetic view images generated from the reconstructed map, and LSD-SLAM [16], which follows the idea from semidense VO [17], are the leading strategies in direct methods.…”

Scale‐aware camera localization in 3D LiDAR maps with a monocular visual odometry

“…Several visual simultaneous localization and mapping (visual SLAM) systems can generate sparse or dense point cloud maps based on cameras. 12 The large-scale direct monocular SLAM (LSD-SLAM) 13 employed an efficient probabilistic direct approach to build semidense maps with an image alignment scheme. Instead of using feature points, LSD-SLAM uses the direct method by aligning pixels to pixels.…”