背景
Non-filter based (i.e., akin to SfM solutions), which are more efficient, are becoming the de facto methodology for building a Visual SLAM system.
介绍
Localization solutions using a single camera have been gaining considerable popularity in the past fifteen years.
Putting aside localization solutions relying on tracking of markers or objects, camera-based localization can be broadly categorized into two approaches.
The remainder of the paper is structured as follows. Section2 reviews the historical evolution of Visual SLAM systems,from the time of Mono SLAM to this date. Section 3 describes the fundamental building blocks of a Visual SLAM system and critically evaluates the differences in the proposed open-source solutions; namely in the initialization, measurement and data association, pose estimation, map generation, map maintenance, failure recovery, and loop closure. Section 4 summarizes closed source non-filter based Visual SLAM systems and finally Section 5 concludes the paper.
概述
In 2007, Parallel Tracking and Mapping was released, and since then many variations and modifications of it have been propose.
PTAM was the first algorithm to successfully separate tracking and mapping into two parallel computation threads that run simultaneously and share information whenever necessary.
In 2014, SVO was published as an open-source implementation of a hybrid system that employs both direct and indirect methods in its proposed solution for solving the Visual SLAM task.
in 2014, Large Scale Direct monocular SLAM(LSD SLAM) employs an efficient probabilistic direct approach to estimate semi-dense maps to be used with an image alignment scheme to solve the SLAM task.
In 2015, ORB SLAM (Mur-Artal et al, 2015) was released as an indirect Visual SLAM system.
———————A survey on non-filter-based monocular Visual SLAM systems———————————-
系统设计
A generic non-filter Visual SLAM system is concerned with eight main components; namely (1) input data type, (2) data association, (3) initialization, (4) pose estimation, (5) map generation, (6) map maintenance, (7) failure recovery, and (8) loop closure.
1. Input data type
- Direct methods
- Indirect methods
- Hybrid methods
方法:
2. Data association
- 2D-2D
- 3D-2D
- 3D-3D
3. Initialization
流程:
方法:
4. Pose estimation
5. Map generation
流程:
方法:
6. Map maintenance
流程:
方法:
7. Failure recovery
方法:
8. Loop closure
方法:
闭源系统
This section aims to provide a quick overview of these systems, which include many interesting ideas for the reader.
总结
Although extensive research has been dedicated to this field, it is our opinion that each of the building blocks discussed above could benefit from improvements.
We are currently working on creating a set of experiments,which cater to the requirements of each of the above open-source systems–for initialization, camera frame rate,and depth homogeneity.
参考
[1] A survey on non-filter-based monocular Visual SLAM systems, Georges Younes