Adventures in archiving and using three years of webcam images

Abstract

Recent descriptions of algorithms applied to images archived from webcams tend to underplay the challenges in working with large data sets acquired from uncontrolled webcams in real environments. In building a database of images captured from 1000 webcams, every 30 minutes for the last 3 years, we observe that these cameras have a wide variety of failure modes. This paper details steps we have taken to make this dataset more easily useful to the research community, including (a) tools for finding stable temporal segments, and stabilizing images when the camera is nearly stable, (b) visualization tools to quickly summarize a years worth of image data from one camera and to give a set of exemplars that highlight anomalies within the scene, and (c) integration with LabelMe, allowing labels of static features in one image of a scene to propagate to the thousands of other images of that scene. We also present proof-of-concept algorithms showing how this data conditioning supports several problems in inferring properties of the scene from image data.

Exploratory analysis of time-lapse imagery with fast subset PCA

“…The dominant modes of variation from the resulting image sequence will come from camera motion, rather than changes within the scene itself, as noted by [7]. Although the top few of these coefficients only account for global changes in the scene, they may provide a first step in selecting new images upon which to run FSPCA.…”

“…In our experiments, we set the value of pixel i to be the projection of the first three normalized PCA coefficients onto the RGB cube. Using PCA coefficients in yearlong summary images, as in [7], tends to highlight daily and seasonal changes through time; an example is provided in Figure 1(a).…”

“…In [7], Jacobs et al document their efforts to make large webcam data sets more accessible to the research community. In the paper, the authors make use of PCA to generate year-long summary images, which describe how a webcam's image changes due to high-level daily and seasonal variations.…”

“…To find the top k exemplar images, we follow [7], and employ a greedy algorithm to approximate a maximally separated set. We first find the image whose PCA coefficients are farthest from the origin and add it to the exemplar set.…”

“…PCA is an often-used tool for exploratory data analysis for large volumes of data, because very high-dimensional data can be visualized in a low-dimensional space, such as 2D plots or as a single pixel in an RGB summary image. Previously, PCA has been used on large sets of webcam imagery to find the dominant ways in which a particular webcam image changes over one or several years [8,7]. Figure 1 shows one example visualization that captures the variation of a scene over the course of a year in a false color plot; the summary image captures variation over times of day and times of year on different axes and color codes each location by the top 3 PCA coefficients computed for the image taken at each time (see Section 5.1 for more details).…”