Make Simulatable 3D Cable Model from Single RGB Image

Abstract

With the development of the times, use the electrical devices is essential for our daily live. For using these devices, we need cables to charge or connect them. So, for people, the cables can be found almost everywhere. The cables also bring new problems, like the cables always appear in a messy form with crosses and knots. We have to tidy up the cables before using them and this is a time-consuming and tedious task. And in colleges and companies where have a large number of cables, when we clean the room or laboratory, we can always find these cables are annoying. For this reason we think that it would make our daily live more convenient to use robots to manipulate and untie the cables. Therefore, for manipulating and untying the cables, this paper proposes a method which can convert the 2D cable data from image into 3D cable data in Unity3D, where the 3D cable model is movable and can simulate the real cable, we call this 3D cable model the “Simulatable Cable Model”. In our approach, we use 2D and 3D neural networks to recover the 3D position information of the cable from the input image, then adjust this 3D position information to increase it's accuracy, and finally create the “simulatable cable model” in Unity3D. The “Simulatable Cable Model” provides a new way to manipulate the cables, that is, to simulate the actions in the virtual environment and then apply it in the real world. Such a method can be used not only to support people daily life with robots, but also can be used to arrange cables in the workplace like factories and so on. We believe that our research is applicable and helpful to the recognition and manipulation of all cord-like objects, and will also be useful in the field of recognizing and manipulating soft objects.