Visualizing the evolution of image features in time-series: supporting the exploration of sensor data

Abstract

Sensor image repositories are becoming the fastest growing archives of spatio-temporal information and they are only projected to grow through the twenty-first century. This continuous data flow leads to large time-series and accordingly, geoscientists are often confronted with the amount of data that need to be explored and the phenomena, present in time-series, to be understood and modeled. In this study, an approach to visual exploration of time-series of sensor data has been proposed. The approach describes exploration as a process where attention, memory, graphics and behaviour of the represented phenomena interrelate. It outlines the framework of how graphics can assist memory and attention by representing image features and their evolution. The main role of graphics in data exploration is to facilitate memorization and to guide visual search in the sensor data. Reasons why the current graphics fall short in supporting the exploration process are also outlined. These are related to insufficient support for re-presentation of image features. The review is further supported by case studies of remote sensing data exploration where the users are primarily interested in identifying, tracing and perceiving the evolution of two highly dynamic image features. The cases dealt with rip channels and convective clouds. Because geoscientists are primarily interested in image features, a workable solution is to focus on just those features -- that is to automatically extract and track them. For that purpose, the feature extraction and tracking algorithms used in computer vision, image processing and scientific visualization are reviewed and a post-processing tracking approach based on the overlap measure is adopted. Computational preprocessing is, amongst others, used to generate the quantitative attributes of objects. The attributes of image features are emphasized in the visualizations with rich graphical and interactive exploratory functionality. Further, a research prototype - a multiple-view exploratory environment based on Space-Time Cube metaphor was proposed. The four views of the prototype are linked and enable object brushing and view manipulation. Dynamic linking enables progressive knowledge construction because users can easily switch between spatial, attribute and temporal--oriented feature analysis. With the interactivity, the users are supported to search for features of interest, sieve them to further reduce the complexity, and focus attention on the selected objects. In particular, exploration of the essence of the object's evolution and history were supported. It was important, however, to verify the concepts developed by user testing. Series of users test conducted involving expert and novice participants. Two exploratory tools were tested: `typical' animation and the research prototype developed during this study. Two case studies described have been used with similar set of exploratory question. The test participants provided faster and more complete and accurate answers using the prototype than the animation. They were also more satisfied with the prototype than with the animation when answering these questions. Thus overall, the test results supported the initial hypothesis - that representing image features and their evolution assist users in exploring the time-series of sensor data.