3D Object reconstruction (in this scope) is the process of representing an object’s appearance using a 2D image. This is achieved by analysing the object’s shape to form a coordinate system representing all points within the object’s frame. Reconstruction information can be applied to various fields of work, such as medical imaging, graphics, virtual reality, etc.

Robot devices are often used in conjunction with reconstruction to develop automated and seamless methods of defining an object. Manipulators are an example device which is commonly applied to reconstruction problems, as their arm-like-manoeuvres along with information-gathering-sensors form an efficient method of image acquisition (the gathering of images to be used for reconstruction). The nature of image acquisition algorithms depends on the task at hand. A common example is multi-view, which takes multiple images (with different views) of an object and uses corresponding pixels to infer the object’s reconstruction. When using robots to perform this reconstruction, it is widespread practice to take images of the object from set waypoints, that will be referred to as picture-points. Note that these picture-points are structured to lie on a hemispherical (or spherical) set of coordinates, so that each waypoint is the same radius (or depth) from the object, creating the foundation to optimised methods (such as next-best-view). Whilst these methods are efficient, it means the manipulator must move between picture-points, and during this movement, the manipulator pays no attention to imaging the object, as its sole goal is to reach the next picture-point.

The aim of this project is to develop a method of using this between picture-point movement to gain information about the object. Implementation and testing for the project will be completed using the ROS2 middleware, with packages and modules being written in languages Python and C++. Accuracy will be measured using object reconstruction percentage, and will be compared to accuracy from other methods of video reconstruction (such as straight-line videos). This project will mostly consist of testing in simulation, but can branch to real-world scenarios if fit within the time constraints.