University of Twente Student Theses

Login

Accelerated Implementation of 3D Visual EKF SLAM For Handheld Perfusion Imaging

Wolters, S.H.G. (2021) Accelerated Implementation of 3D Visual EKF SLAM For Handheld Perfusion Imaging.

[img] PDF
5MB
Abstract:This thesis presents an accelerated implementation of the Visual Error State EKF SLAM algorithm for the handheld perfusion imaging device. The algorithm can be used to increase the accuracy of the handheld perfusion imaging, while providing a 3D map of the scanned surface. Unfortunately, the ES-EKF SLAM algorithm has a high order computational complexity. The computational bottlenecks of the algorithm are mapped and a set of system requirements is composed. Through a design space exploration, the GPU with Tensor cores and CUDA cores is then selected to accelerate the computational bottlenecks of the ES-EKF SLAM algorithm. The computation actions are allocated on a system consisting of two computers: a single board computer with a GPU that performs the landmark measurement acquisition and a desktop computer with a high-end GPU that performs the ES-EKF SLAM computations. Through runtime evaluation of the accelerated implementation and an existing non-accelerated implementation, it can be concluded that the GPU accelerated implementation has a higher runtime performance than a non-GPU accelerated implementation. However, in order to satisfy the runtime requirements of the system, the input size must be limited. In the future implementations opportunities for further acceleration of the algorithm must be taken in consideration.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:https://purl.utwente.nl/essays/86173
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page