Visual and Visual-Inertial SLAM: State of the Art, Classification, and Experimental Benchmarking

Table 3

Indications on the robustness to various scenarios of the most famous vSLAM methods.

Algorithm	Recommended usages					Objectives
	Lifelong exp.	Large envir.	Low textured	Outdoor (light, outliers)	Robust to mov.
MonoSLAM [21]	−	−	−	−	−	Pose estimation in robotics
PTAM [27]	−	−	−	−	∼	A.R. in small workspaces
ORB-SLAM2 [76]	+	+	∼	+	+	Robust large path tracking
Edge-SLAM [81]	∼	+	+	+	+	Low-textured environments
DTAM [34]	−?	−	∼	∼	+	Robustness to motion blur
MobileFusion [66]	+	−	−	−	−	3D object modeling on phone
LSD-SLAM [35]	∼	+	−	+	∼	Semidense trajectory estimation
SVO [67]	+	∼	+	+	+	Fast, consistent, semidirect method
DSO [33]	+	∼	+	+	+	Direct and sparse VO method
KinectFusion [68]	+	−	+	+	∼	3D modeling with the Kinect
ElasticFusion [70]	−	−	+	∼	∼	Map-centric vSLAM
S-MSCKF [17]	+	∼	∼	∼	+	Rapid and consistent Kalman filter
ROVIO [26]	+	∼	−	∼	∼	Robust VIO for UAVs
OKVIS [73]	+	+	∼	+	+	Robust stereo VIO for UAVs
Vins-Mono [74]	+	+	∼	+	+	Full viSLAM method
Kimera [60]	+	∼	+	∼	+	VIO+3D semantic-metric mesh
VIORB [75]	+	+	∼	+	+	VI method based on ORB-SLAM

For each difficulty, we consider the method to be either robust (+), to have potential difficulties (∼), or to not be recommended at all (−). This does not reflect the overall accuracy of the method or the robustness of the initialization procedure.