In the science of image processing, two-dimensional (2D) object becomes raw material that
can be processed into another form to get the desired information. Within its development, the image processing refers to three-dimensional (3D) images which have more angles than the 2D object, so that we could get more information from the object.
Various methods for reconstructing 3D objects have been done by various researchers. Reconstructing 3D objects from multiple images is done by combining some calibrated 2D images (Kolev, Brox, & Cremers, 2012). The use of the Voxel Hashing algorithm is to reconstruct 3D objects to achieve good quality results with scale (Matthias Nießner, 2013). The use of Voxel Hashing improves the effective 3D point-based method used to reconstruct thin objects (Ummenhofer & Brox, 2013).
Reconstructing 3D objects in real time becomes an option when being implemented into game-based models and survival activities that require 3D viewable objects. The use of depth vision cameras like Kinect Fusion for taking human objects in 3D has a limited area (R. Newcombe, 2011). The development of kinect camera in its utilization is within the field of
In the event of surveillance undertaken, underwater observations become different things because the objects in the sea have uniqueness in every inch of them. Therefore an observation in the sea is needed by showing the whole object in 3D. Problems arise because during the formation of objects into 3D form there are differences in terms of lighting conditions, the determination of the edge of the object, the distance between the object and camera surveillance, as well as the existence of irregular noise around the object to be reconstructed.
This research is conducted to determine the application of which method that is suitable for taking underwater objects and reconstructing objects in 3D. Reconstruction will be done by taking multiple images in 2D form, shooting will be done by ROV (Remote Operated Vehicle), and ROV will be equipped with sensors and actuators that set the accuracy of each distance.
In the reconstruction of 3D objects, several studies have been done. Initial methods such as
the use of laser light mixed with the use of the camera with relative position and scene measurement projected to an object to obtain the shape of objects to be reconstructed into 3D images (Dipanda, 2005). The use of low-resolution cameras to reconstruct a surface image of an object was performed without the use of a laser beam. The method was done by detecting the grid segment, then reconstructing the surface image of an object (Dvorak, 2010). On the other hand Kinect cameras manufactured by Microsoft have been developed. By using Kinect
Fusion, then the kinect camera would be able to quickly reconstruct objects captured in 3D, with the provision of objects being indoors (Izad, 2010).
Research 3D reconstructions
Various approaches used to maximize reconstruction are limited by the camera’s ability to capture images. The determination of point-based on reconstructing 3D objects into a fairly large selection, point-based can also be used to reconstruct thin objects (Ummenhofer & Brox, 2013), merging of multiple images to be reconstructed into 3D objects using Direct Linear Triangulation (DLT ) that allows reconstruction from simple cameras (Rachmawati,
2013). Differential scaling on multiple images is a problem for the DLT algorithm. Another way is to compress the scale using spatial hashing scheme or Voxel Hashing (Nießner, 2013). To create faster and real-time reconstruction, images are taken from the video by dividing the image into segments, then checking the image connectivity of each segment using the connectivity constraint coupled with spatio-temporal multi-view reconstruction which is useful to reconstruct 3D objects in real time (Oswald, 2014). The retrieval of objects by Martin and Oswald creates a more detailed picture of previous engineering techniques, but the focus on the object will be much disturbed by the ever-changing noise due to real time shots. Backgrounds such as floors, walls, and other objects will give a large noise to the object. Signed Distance Fields (SDFs) technique will separate the object from the floor as well as the surface of the wall, so that the objects will be easier to reconstruct (Dzitsiuk,
2017). Techniques used to obtain objects are by reducing effective noise in the room because the surface of the wall and the surface of the floor have the same segmentation. However, if the surface is dynamic and has a different structure such as the seabed or even rocks, the object reconstruction will be more difficult to do because the object will be biased with other objects.
Research 3D reconstructions
3D reconstruction is also a research area that is being developed in the world of robotics. 3D simultaneous localization and mapping (SLAM) technique is very actively used because the robot poses determination and geometry planning needs fast and accurate estimates. 3D reconstruction is a requirement because of the task of robotics in localization, navigation, and exploration (Bylow, 2013). In addition to use in the world of robotics, 3D reconstruction is also used for survival purposes.
Research 3D reconstructions
The research conducted by Xu Yongzhe in 2015 using multi-camera to reconstruct the ground surface into 3D coordinates. The system will detect floors, walls, or buildings. This can improve the tracking system of moving objects. This technique uses 3D reconstruction SFM (structure from motion) which reconstructs 3D geometric from a set of 2D images captured in real time (Xu Yongzhe, 2015).
Bylow, Erik, J. S. (2013). Real-Time Camera Tracking and 3D Reconstruction Using Signed
Distance Functions. Robotics: Science and Systems IX.
Dipanda, A. S. W. (2005). Towards a real-time 3D shape reconstruction using a structured light system. Pattern Recognition.
Dzitsiuk,Maksym, J. S. (2017). De-noising, Stabilizing and Completing 3D Reconstructions
On-the-go using Plane Priors. Arxiv.
Dvorak,Radim, M. D. (2010). Object Surface Reconstruction from One Camera System . Izad, Shahram e. a. (2010). KinectFusion: Real-time 3D Reconstruction and Interaction
Using a Moving Depth Camera.
Kolev, K., Brox, T., & Cremers, a. D. (2012). Fast Joint Estimation of Silhouettes and Dense
3D Geometry from Multiple Images. EEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INT, 1-13.
Oswald, Martin R. J. S. (2014). Generalized Connectivity Constraints for Spatio-temporal 3D Reconstruction. Germany.
Newcombe,R. A. D. (2011). KinectFusion:Real-time dense surface mapping and tracking.
Nießner,Matthias, M. Z. (2013). Real-time 3D Reconstruction at Scale using Voxel Hashing.
ACM Transactions Graphics (TOG), vol. 32, no. 6.
Rachmawati. (2013). 3D Object Recontruction from Multiple Images. JNTETI. Ummenhofer, B., & Brox, T. (2013). Point-Based 3D Reconstruction of Thin Objects. Yongzhe,Xu B. L. (2015). Multi Camera for Surveillance System Ground Detection and 3D
Reconstruction. International Journal of Smart Home, 103-110.