|
Methods | Dataset | Advantage | Disadvantage |
|
Zhang et al. [118] | MAV | Correct ratio >99.9%. | — |
Lipschutz et al. [119], probability distribution edge detection Hough transform | — | Visible-light images: MHD 3.69 pixels; AD 0.28 degrees. Infrared images: MHD 1.49 pixels; AD 0.14 degrees | — |
Gershikov et al. [114], H-REM | — | MHD 2.28 pixels; AD 0.19 degree; mean run time 0.14 s | — |
Prasad et al. [110], weighted edge Radon multiscale consistence | SMD MAR-DCT Buoy dataset | MPE 2 pixels; MAE 0.4 degrees | Does not work well in certain scenarios |
Jeong et al. [116], multiscale approach region of interest (RoI) | SMD Buoy dataset | 15 fps; MPE <2 pixels; MAE 0.15 degrees | Performance reduction (edges related to horizon) |
Sun et al. [120], coarse-fine-stitched hybrid filtering Random sample | SMD Marine Obstacle | MHD 0.89 pixels; MAD 0.19 degrees | — |
Liang et al. [117], probability distribution physical characteristics | SMD Buoy dataset | MPE 7.6 pixels; MAE 0.4 degrees | Ineffective (large area occlusion) |
Jeong et al. [109], multiscale approach NN | SMD | MPE <1.7 pixels; MAE 0.1 degrees | Line features absent |
Yang et al. [121], probabilistic graphical expectation-max Gaussian models | Marine obstacle | — | Reflection and illumination |
|