| | Methods | System name | Algorithm and requirements | Accuracy of RM | Testbed area |
| | Crowdsourcing | RedPIN [25] | Label position by user, indoor map | Room level (90%) | 26 rooms | | Molé [22] | Kernel, accelerometer | Room level (91%) | 3-floor building | | FreeLoc [21] | Relative RSS comparison | <3 m | A laboratory, a corridor |
| | SLAM | WiFi-SLAM [8] | GP-LVM, initial ISO map model | 3.97 m (ME) | 250–500 m (traces) | | SignalSLAM [26] | Least square, PDR, GraphSLAM, landmarks, accelerometer, gyroscope, magnetometer | <16.5 m (MD) | | | Graph-SLAM [18, 19] | Sparse graph, constraint optimization, least square, linearization, approximation, EKF SLAM, accelerometer, GPS | <10 m | Urban area
|
| | Inertial sensors | Zee [10] | DR, augmented particle filter, indoor map, accelerometer, gyroscope, magnetometer | 1.2 m (50%), 1.8 m (80%) | | | LiFS [24] | DR, feature extraction, indoor map, accelerometer | 5.88 m (ME) | | | WILL [27] | PDR, -means, accelerometer | Room level (86%) | |
| | Semisupervised learning | Manifold learning [28] | Manifold alignment, inherent spatial correlation of RSS, path loss model, partial RPs, APs’ locations, indoor map | 3.8 m (ME); 2.4 m (ME) | , 5 APs; , 4 APs | | Coforest [29] | Implicit crowdsourcing sampling, random forest ensemble classifier, partial RPs, RSS | 3.65 m (ME) | 800 m2, 30 APs |
| | Unsupervised learning | WRM [30, 31] | HMM, EM, memetic algorithm, path loss model, indoor map, APs’ locations | Around 3 m (ME) | , 30 APs |
| | Path loss model | ARIADNE [32] | Ray tracing, path loss model, simulated annealing algorithm, APs’ location, partial RPs, indoor map | 3 m (ME), 2.5 m (STD) | , 5 APs | | Multiwall Path Loss Model (MWM) [33] | MWM, APs’ location, indoor map, parameters setting for Gaussian distribution, Euclidean distance error, kNN | 1.2 m (ME) | 480 m2, 3 APs |
| | Interpolation | Inverse Distance Weighting (IDW) [34] | RSS, interpolation and extrapolation methods, estimation error statistics, uniform grid, IDW, probabilistic positioning | 5~20 m (ME) | , 316 2.4 GHz APs, 106 5 GHz APs | | Kriging [35] | Kriging algorithm, spatial interpolation, semivariogram model fitting, unbiased estimation, RSS, -weighted nearest neighbours, | 1.12 m (ME) | 9.5 m × 2.5 m, 9 APs | | Forward Interpolation [36] | cubic spline, boundary condition (fixed, zero-slope natural, nonnode), RMS, RSS | 2.82 m (best) | 5 rooms, 4 APs |
|
|
