A Qualitative Comparison of the State-of-the-Art Next-Best-View Planners for 3D Scanning

This is a survey paper in which we review the state-of-the-art Next-Best-View planners with the focus on their application in solving an autonomous 3D scanning task.According to market reports, the 3D scanning market will continue to grow in response to the increasing demand for augmented and virtual reality solutions.Taking into account that the number of skilled 3D artists is limited and their labor is highly paid, an alternative way of creating high quality 3D models is 3D scanning existing objects.In many cases, 3D scanning is the only way to get photorealistic textures and high-definition models.Automated 3D scanning can be used as a way to preserve art, document changes in the environment, create detailed models of consumer products.Six next-best-view planners were compared using ROS in the Gazebo simulation environment.The MA-SCVP machine learning method achieved on average 93.1% coverage, that is 5.9% higher than ScanRL, 36% higher than SEE, and 1% higher than volumetric information gain methods.Maximum coverage with the MA-SCVP method was achieved after 12.2 views on average, versus 20 views for the volumetric information gain methods.