Table 1. The main characteristics of the most relevant State-of-the-art.

Published	Single/Multiple frame	Single/Multiple object	Static/Dynamic object	Input type	Methods
Kulikajevas (2019)	Single frame	Single object	Static object	RGB-D sensor	Hyper neural network
Kulikajevas (2019)	Single frame	Single object	Static object	3D Models	GANs neural network
Widya et al. (2019)	Multiple (2 image sequences)	Single object	Static object	Monocular endoscope	Structure from motion (SfM)
Wang (2018)	Single frame	Single object	Static object	RGB-D sensor	Monocular SLAM
Yang (2020)	Single frame	Multiple (full scene)	Static scene (remove dynamic objects)	Monocular RGB	Online incremental mesh generation
Shimada (2020)	Single frame	Single object	Dynamic object	Monocular RGB	Markless 3D human motion capture
Peng (2020)	Single frame	Single object	Dynamic object	Monocular RGB	GCN network
Ku (2019)	Single frame	Corp single object	Dynamic object	Monocular RGB	geometric priors, shape reconstruction, and depth prediction
Lu (2020)	Multiple (two consecutive point-cloud)	Multiple (full scene)	Dynamic objects	Outdoor LiDAR datasets	LSTM and GRU networks
Weng et al. (2020)	Single frame	Multiple (full scene)	Dynamic objects	Outdoor LiDAR datasets	Predict next scene using LSTM
Akhter (2010)	Single frame	Multiple objects	Dynamic objects	Monocular RGB	Structure from motion
Fragkiadaki et al. (2014)	Multiple frames	Single object	Dynamic object	Monocular RGB	Non-rigid structure-from-motion (NRSfM)
Ranftl (2016)	Multiple frames (two consecutive)	Multiple (full scene)	Dynamic object	Monocular RGB	Segments the optical flow field into a set of motion models
Kumar, Dai & Li (2019)	Multiple (2 frames)	Multiple (full scene)	Dynamic objects	Monocular RGB	Super pixel over segmentation
Proposed framework	Multiple (whole video frames sequence)	Multiple (full scene)	Dynamic objects	Monocular RGB	Unsupervised learning and point cloud fusion