0:00:07 | this purpose i |
---|
0:00:13 | we work on automated algorithms for image understanding and were specifically looking at trying to |
---|
0:00:18 | understand images of outdoor scenes |
---|
0:00:21 | we collect images from lots of different types of cameras be they web cams that |
---|
0:00:26 | we can download images from or images that are uploaded to social network sites and |
---|
0:00:31 | we write algorithms to try to interpret the images |
---|
0:00:34 | we solve a number of different kinds of problems so one of the problems might |
---|
0:00:38 | be given the set of images what is the three d geometry in the scene |
---|
0:00:42 | or what are the material properties of the things in the scene or if we |
---|
0:00:47 | have images captured over a longer period of time it might be something like howard |
---|
0:00:52 | the plants growing in this scene or how is the weather changing from data one |
---|
0:00:57 | of the things we're focused on right now is trying to use what is typically |
---|
0:01:01 | been considered a nuisance |
---|
0:01:02 | for outdoor scene understanding and try to use it to our advantage clouds moved to |
---|
0:01:06 | outdoor scenes they cause various lighting changes and on partly cloudy days you have lighting |
---|
0:01:12 | that gets brighter or darker brighter or darker that causes lots of trouble for algorithms |
---|
0:01:16 | that are trying to say to recognize if something is changed so what we're trying |
---|
0:01:20 | to do is |
---|
0:01:22 | flip that on its head and say how can we actually use the fact that |
---|
0:01:25 | it's a cloudy day to infer things about the scene and the camera so trying |
---|
0:01:30 | to use clouds moving to the scene to estimate three d geometry |
---|
0:01:34 | of what we're looking at trying to estimate what direction the cameras looking trying to |
---|
0:01:40 | characterize the types of clouds that are passing through the scene we take large sets |
---|
0:01:45 | of images and we try to extract patterns and they can be patterns that we |
---|
0:01:50 | used understand the patterns themselves or patterns that we used to understand |
---|
0:01:55 | things about the camera for the location that were in so we're really interested in |
---|
0:02:00 | trying to take |
---|
0:02:02 | video of the outdoor scenes and understand how people are moving through then how the |
---|
0:02:07 | way that people move through an outdoor scene changes based on various other conditions |
---|
0:02:13 | so for example there are more people walking around the u k campus on a |
---|
0:02:19 | monday then there are on a sunday in general because people are walking classes and |
---|
0:02:23 | are not on the weekends |
---|
0:02:24 | and those are very simple patterns but they're also more complex patterns we want to |
---|
0:02:28 | characterize how normal |
---|
0:02:31 | the motion of people is on a particular day and how normal what people look |
---|
0:02:35 | like use of a particular day as the search starts from you moving around the |
---|
0:02:39 | world and seeing things to collecting some data and then trying to get things to |
---|
0:02:43 | work |
---|
0:02:43 | on that data to solve a problem and so we write code to try to |
---|
0:02:48 | create this pipeline from source data to this target we're trying to get to |
---|
0:02:54 | and we see how it works and then we iterate so we really have a |
---|
0:02:56 | nice collection of different people working in this or different stages of their education we |
---|
0:03:02 | can have people to make significant contributions at |
---|
0:03:06 | all of these different levels |
---|
0:03:07 | we have undergrads we're working on different ways to collected and different ways to visualise |
---|
0:03:12 | it and to interact with it |
---|
0:03:13 | and then we have phd students were doing things like let's take this sort of |
---|
0:03:17 | images and let's build some machine learning for computer vision based approach to extract information |
---|
0:03:24 | there's a world out there that's changing in sometimes predictable ways sometimes unpredictable way so |
---|
0:03:30 | we want somehow |
---|
0:03:31 | combine all of this together and build algorithms that can somehow take advantage of these |
---|
0:03:37 | patterns are in the world |
---|
0:03:40 | i |
---|
0:03:49 | modeling |
---|