Feature-based visual simultaneous localization and mapping: a survey

Abstract

Visual simultaneous localization and mapping (SLAM) has attracted high attention over the past few years. In this paper, a comprehensive survey of the state-of-the-art feature-based visual SLAM approaches is presented. The reviewed approaches are classified based on the visual features observed in the environment. Visual features can be seen at different levels; low-level features like points and edges, middle-level features like planes and blobs, and high-level features like semantically labeled objects. One of the most critical research gaps regarding visual SLAM approaches concluded from this study is the lack of generality. Some approaches exhibit a very high level of maturity, in terms of accuracy and efficiency. Yet, they are tailored to very specific environments, like feature-rich and static environments. When operating in different environments, such approaches experience severe degradation in performance. In addition, due to software and hardware limitations, guaranteeing a robust visual SLAM approach is extremely challenging. Although semantics have been heavily exploited in visual SLAM, understanding of the scene by incorporating relationships between features is not yet fully explored. A detailed discussion of such research challenges is provided throughout the paper.

State of the Art in Vision-Based Localization Techniques for Autonomous Navigation Systems

“…Instead, they rely on the use of local sensory information for determining the robot's relative position and orientation with respect to its starting point. The main components of any SLAM technique are the map/trajectory initialization, data association, and loop closure [22]. Odometry algorithm is employed in SLAM system to localize the moving robot within the environment.…”

“…Therefore, SLAM techniques are able to reduce the accumulated pose error when the robot returns to a previously observed scene using the history of robot poses in the global map. In addition, odometry algorithms implement local map optimization methods, such as windowed bundle adjustment, to optimize the local map only over the last poses, leading to local map consistency [22], [23]. SLAM aims at maintaining a global map consistency and odometry method is used partially during the SLAM first process which is followed by other steps [24], i.e., local or global map optimization.…”

Pose-Graph Neural Network Classifier for Global Optimality Prediction in 2D SLAM

“…Moreover, the work in [13] proposed a technique to verify if a given solution to a pose-graph is globally optimal. To accomplish that, the pose-graph optimization problem is first reformulated where chordal distance [22] is used instead of angular distance to represent the magnitude of the difference between two rotation matrices in SO (2). Thereby, the optimization problem becomes amenable to implementing duality theory (e.g.…”

“…Robot poses along the trajectory and the observable landmarks in the environment under investigation are represented as vertices in the graph. In case landmarks are not inserted into the map and only robot poses are considered, the algorithm is referred to as pose-graph optimization [2], [3]. Spatial constraints, that are formulated based upon the sensory measurements collected by the robot, are encoded as edges that connect the graph vertices.…”

Role of Deep Learning in Loop Closure Detection for Visual and Lidar SLAM: A Survey

“…Hence, a thorough taxonomy is required to categorize the loop closure detection algorithms. In recent research [ 17 ], an extensive comparative analysis for feature-based visual SLAM algorithms was presented. The existing research is grouped into four categories based on visual features, i.e., low level, middle level, high level, and hybrid features, and highlighted their limitations.…”