The object is to explore different methods for testing if a number is prime, among them the famous so called AKS primality test. This algorithm runs in time that is a polynomial in the length of the number being tested.
In order to estimate the shape and measure a pair of feet, it is necessary to have an accurate estimation of the camera pose (including scale). In this project we are going to face challenging scenes with little texture and potentially rapid mo- tion. The goal is to improve the camera pose estimation and the robustness to meet the requirements in industry. This can be done by using more sensor data and/or using information about the scene that is going to be valid for our setting.
This project will focus on estimating the shape of the foot given a single image. From a top view perspective, can we accurately estimate the width and length of the foot? What can be inferred about height etc using machine learning techniques? In this exciting project we are going investigate what can be said about a foot from just a single image. The project will involve both statistics, optimization and computer vision.
Can modern machine learning techniques and computer vision be used to infer shoe fit from just a few images? Volumental dispose a large set of 3D scanned feet and customer data of bought shoes. Can we formulate this as a ML-problem and directly use imagery to find a perfect shoe fit? The goal of this project is to develop a neural network that can use multiple images as input and infer the correct shape parameters of a parametric foot model.
Recent breakthroughs in image analysis and text recognition, fuelled by a rapid progress in machine learning, means that software for the first time can compete with, and sometimes even surpass, humans in tasks such as classifying images and reading handwritten documents. In this project you will develop method for interpreting handwritten documents from the 16th century and evaluate how they work on 11 453 pages from Älvsborgs Ransom. The text is written by several scribes. The consistency of style and layout within the pages of one scribe is high, but there is significant differences between scribes. This is an excellent example of a machine learning problem, where the use of context (the scribe), could be used. This is a research area which is of theoretical interest, with applications in many areas. In the project one could also study how to extract the structure of the documents automatically using convolutional neural networks.
This master's thesis project concerns the development of methods to model and predict how harvest yield depends on data. In the project we will evaluate potential methods and algorithms to build a predicative model, that can be used for optimization of agriculture inputs and planning. In this project we work with Hushållningssällskapet in Skåne and with T-Kartor. We have harvest data (one point every fifth seconds) that can be used as evaluation data. We have georeferenced spatiotemporal and spectral data that can be used as potential indata both as points and imagery, but also data from soil samples, satellite sensor data, terrain model, rainfall etc. Data will be available from selected fields in Skåne from 2020.
In structure from motion the goal is to recover both the scene structure and
the movement of the camera. Traditionally this has been done using imagery
only with the well known scale-ambiguity. In practical applications, the scale is
of great importance in order to actually measure stuff. A modern smart-phone
have more sensors than just the camera, for example an accelerometer and a
gyroscope. This gives extra information about the distance traveled between
the cameras and how the camera was oriented. This project aims to combine
image information with IMU data in order to find both the scale and improve
the camera trajectory.
We’re looking for one or two master degree students, with an interest in exploring the capacity of an algorithm in an applied environment, able to segment one or multiple objects. When waste objects are distributed on the sorting robot’s conveyor belt, they can end up on top of each other. By identifying each object, both gripping the objects and sorting can be optimized.
The task for this thesis work is to successfully segment objects of various shapes that can only partially been seen by the camera.
Centre for Mathematical Sciences
Box 118, 221 00 LUND
Phone: 046-222 00 00 (växel)