Wednesday, April 09, 2014
What you see is so ten seconds ago
Fischer and Whitney's study.
Sunday, February 09, 2014
Retinex Effect to Allow the Blind to See
Friday, August 16, 2013
Finding Videos on Mobile Devices
IR and friends
Scalable mobile video retrieval with sparse projection learning and pseudo label mining
Lexing Xie (ANU)
Monday 19 August 2013
Retrieving relevant videos from a large corpus on mobile devices is a vital challenge. We address two key issues for mobile search on user-generated videos. The first is the lack of good relevance measurement, due to the unconstrained nature of online videos, for learning semantic-rich representations. The second is due to the limited resource on mobile devices, stringent bandwidth, and delay requirement between the device and the video server.
We propose a knowledge-embedded sparse projection learning approach. To alleviate the need for expensive annotation for hash learning, we investigate varying approaches for pseudo label mining, where explicit semantic analysis leverages Wikipedia and performs the best. In addition, we propose a novel sparse projection method to address the efficiency challenge. It learns a discriminative compact representation that drastically reduces transmission cost. With less than 10% non-zero element in the projection matrix, it also reduces computational and storage cost.
The experimental results on 100K videos show that our proposed algorithm is competitive in the performance to the prior state-of-the-art hashing methods which are not applicable for mobiles and solely rely on costly manual annotations. The average query time on 100K videos consumes only 0.592 seconds. This is joint work with Guan-Long Wu, Winston Hsu and others in National Taiwan University....
Thursday, June 20, 2013
Real-time surveillance with wireless cameras
Enabling real-time tracking in embedded camera networks using compressive sensing
Associate Professor Chun Tung Chou (School of Computer Science and Engineering, University of New South Wales, Sydney)
COMPUTER SCIENCE SEMINAR\
DATE: 2013-06-24 TIME: 11:00:00 - 12:00:00 LOCATION: CSIT Seminar Room, N101 ...
ABSTRACT:
An embedded camera network consists of distributed video sensors interconnected by wireless links. Due to the high data rate of video sensors, it is not feasible to stream the video data to a centralised location (sink) for processing. This means that a lot of the video processing must be performed on the embedded device. However, this presents a challenge because of limited computation power on embedded platforms. For example, conventional background subtraction methods can only process a few video frames per second on embedded platforms.
In this talk, we discuss how compressive sensing can be used to address this computational bottleneck. We present a compressive sensing based background subtraction method, whose accuracy is similar to conventional methods, but is five times faster. We show that this faster background subtraction method enables real-time tracking in an embedded camera network. If time allows, I will also discuss briefly on some of my other work on using compressive sensing in wireless sensor networks.
BIO:
Chun Tung Chou is an Associate Professor at the School of Computer Science and Engineering, University of New South Wales, Sydney, Australia. He received his Ph.D. from the University of Cambridge, UK. He has published over 100 articles on various topics, including, wireless mesh networks, wireless sensor networks and system identification. His current research interests are wireless sensor networks, compressive sensing and nano-communication.
Wednesday, May 30, 2012
Best of GovCamp Videos
I did not do any editing, as I found there was already an excellent three part video hosted by Senator Lundy, about GovHack. So interspersed this with screen-casts from Mashup Australia and videos from other such events around the world (all of which I found on YouTube):
- Teapots Animation
- Behind the scenes at GovHack - Part 1
- Fridgemate
- Behind the scenes at GovHack - Part 2
- NeoGopher Mashup Australia Screencast
- Behind the scenes at GovHack - Part 3
- GovCamp Singapore 2.0
Suggestions for additions would be welcome, as would an edited version.
Sunday, April 15, 2012
Surveillance Style Headset for Video Conferences
The surveillance headset has a soft ear plug which fits in one ear, connected to a clear plastic tube up over the ear and down the back of the neck with a curly tube. The tube ends in a clip attached to the collar. The neck clip has a wire which runs to a small box with clip to attach to your lapel, a push-to-talk (PTT) switch and microphone. This box has a second cable with a plug for the computer (some units have a separate lapel microphone).
That sounds complicated and it is a bit fiddly to put on. But once fitted, the ear-piece is very comfortable and not noticeable on camera. The unit I bought has a PTT switch. This is very handy for events with large numbers of participants where you only speak occasionally: you hold the switch to talk and so can't accidentally leave it on. Some units are designed for "VOX" so the microphone is on all the time and some have a VOX/PTT switch to select either mode. My unit has no VOX switch but I find a rubber band works fine to keep the microphone switch down for extended periods.
The tricky part is to find a surveillance headset with the plug for your computer. Most of the headsets have just one 3.5mm stereo style male plug, some have two mono 3.5mm male plugs, some a 3.5mm and a 2.5 mm plug. Most computers have two 3.5mm stereo sockets: one for the microphone and one for the earphone.
Also I found the headset works fine with my mobile phone, using a 3.5 mm to 2.5mm stereo adapter.
Saturday, March 31, 2012
Twitter-like Video Conferences
One thought was to accept that communication in a group of people, such as a class, is not like a one-to-one conversation. If we assumed that only one person can talk at a time (half-duplex) and only can say a little and then pause for a reply, that will make the technology easier and allow for more computer mediation.
A time limit of about 30 seconds, being the speaking equivalent of an SMS message, would seem reasonable. Breaking a conversation up into 30 second packets might seem very unnatural. But Twitter has been very successful, breaking text conversations up this way. While that makes the conversation less free-flowing, it adds many advantages.
But is 30 seconds a good length for an audio message? I started searching to see if there was any research on how long people talk for continuously in a conversation. After a couple of web searches I found there was a whole area of research devoted to the "length of utterance" and this is used to measure language proficiency.
The "turn length" for Small-group Discussions (Li & Nesi, 2004) discussions has been reported to be from 23.1 to 71.5 seconds. This suggests that the maximum length for an audio/video message needs to be longer than 30 seconds, perhaps 120 seconds.
Also interesting is that discussions between students finished after 5 minutes 49 seconds to 6 minutes 31 seconds. This is much shorter than the typical 30 or 60 minute business meeting or education class. Perhaps if we had a way to organize meetings and classes more easily, they could be of this more natural short length.
In addition the researchers noted differences in the structure of discussion between Chinese and English speakers. They also noted the frustration when the groups mixed and while they spoke the same language, they used different discussion techniques. Perhaps explicit cues could be built into the Computer Medicated Communications system, where something like emote-icons would indicate if the speaker was Reformulating a point, Endorsing, or so on. Also it may be useful to teach students about different discussion styles.
Reference
Li, M., & Nesi, H. (2004). Exchange Patterns in Small-group Discussions: a comparative study of Chinese and English discourse in peer group, divergent discussion tasks. The East Asian Learner, 1(2). Retrieved from: http://www.brookes.ac.uk/schools/education/eal/eal-1-2/vol1-no2-li-nesi.html
Saturday, March 17, 2012
Captioning Digital Video Using Web Standards
Thursday, March 15, 2012
Bandwidth and Data Usage of On-line Education
- Just listening, not video: 4kBps, 9.4MB.
- Just listing and watching video: 20kBps, 46.88MB.
- Just listing to a prerecorded session, with no video: 2.5kBps, 5.9MB.
- Just listing and watching prerecorded video: 16kBps, 37.5MB.
An interesting aspect of this is the lower bandwidth and total data for prerecorded sessions. This is because better compression can be used. Just as students can assimilate information better if given more time, so can computers. ;-)
Wednesday, March 14, 2012
Web Video Standard to Revolutionize e-Learning
VIDEO SYNCHRONIZATION
First speaker was Christopher Giffard on problems with "video", in terms of interaction and accessibility. Assistive aids for the disability, such as closed captions, are considered excessively time consuming. He pointed out that HTML5 could be used poorly to support accessibility.
Christopher advocated the use of "timed data". This is interesting as I have been considered the issue of synchronous and asynchronous online learning. Many of the issues are the same: how do you allow for events which happen at a particular point in time, and others which do not.
There is an emerging standard for the use of <track>, which is similar to <source>. However, there have been few implementations. Track can be sued in <video> or <audio> to provide subtitles and captions. The captions are stored externally and then specifies an algorithm for synchronising this with the video or audio. Track can be manipulated with Javascript. Track works by specificity a source file for the caption content, a language and a data type (such as text/vtt) and a kind (such as "captions").
Christopher built a tool Captionator.js to support track, using the Media Text Tracks JavaScript API.
Christopher recommended WebVTT (Web Video Timed Text), a new caption specification language. This is a flat text file format based on SRT. It has lines of text starting with start and stop time codes, followed by text captions. There can be limited HTML type mark-up in the text, for bold, italics and the like. There are also "Chapters" to tag segments of video (with sub-chapters).
These features could be very useful for educational videos, for all students, not just those with accessibility requirements. Videos can be given a table of contents like a book.
In 200 I experimented with using synchronized audio with slides for education. See "Tips and Traps With Electronic Presentation Tools":
- Text,
- Slides (in OpenOffice format),
- Audio Slideshow (in Real Media Slideshow).
However, this material took considerable effort to prepare and depended on proprietary formats for playback.
It should be noted that the technology is not limited to linear video. Christopher pointed out that not only can closed captions be displayed, but they can automatically pause the video to allow time to be read.
FORMAL LANGUAGE TO SUPPORT LEARNING SYNCHRONIZATION?
The use of a formal language to synchronize text and video suggests that perhaps something similar for learning. At present an educational designer will describe what the student needs to know and have done at the beginning and end of a course, but not in detail in between. In effect there is only synchronisation at two points: the start end of the course. In contrast a computer programmer will specify precise details as to what a process has to have done which it reaches a point to exchange data with another. This would get around the problem of students not knowing what they need to do or when they need to do it.
NEW WORLD OF VIDEO A FEW MONTHS AWAY
Christopher pointed out that support for video synchronization will be available in web browsers within a few months. Mozilla are "making progress" with Firefox. IE10 will support WebVTT, TTML and the JavaScript API.
Christopher demonstrated using a video of "Minister Garrett Introduces the School Funding Review". With this the transcript introduces each speaker and allows the viewer to click on a line of the text and play the relevant clip. He also showed a media management system to keep track of how accessible videos are.
Thursday, February 16, 2012
HTML5 Video, Captioning, and Timed Metadata
Tuesday, February 14, 2012
Editing Educational Video
One of the first issues which came up is the very large size of the video files. This makes it difficult to copy them from computer to computer so that several people can work on the project at the same time. Another issue is which software to use. Simple applications like Apple iMovie may not have fine enough details.
Tuesday, February 07, 2012
Reliable Low Bandwidth Video for eLearning
Thursday, January 26, 2012
Web Conference for Teaching
Recently I took part in on-line training in the use of the "Adobe Connect", video conference software for education. The training was fun: Like a group of kids talking into tin cans connected by string.
Adobe Connect worked much better than when I tried Wimba Classroom and should be adequate for some teaching and research supervision.
I was able to install Adobe Connect for Linux and run the compatibility tests successfully. But it did not work for the live conference (I suspect there is just some little security setting to change for it to work). It worked well enough under Ms-Windows on a borrowed laptop (but like many in Computer Science I don't normally use MS-Windows).
While adequate, I did not think Adobe Connect as good as "Blackboard Collaborate" (previously called Elluminate Live!") and no better than the free open source alternatives available. In particular Connect is heavily dependent on Flash, giving it poor accessibility and limited support on non-Windows/Mac platforms. I was not able to increase the size of the text in the interface and could not move windows around on the screen.
I was able to change the bandwidth setting to "modem" and change to lower bandwidth audio and video settings. So this should work adequately for students on slow links.
A major lack is an automated on-line end-to-end test. What is needed is a place where you can verify the microphone, speakers and software are working, by sending audio and video over the network and having it played back.
One of the participants expressed surprise that video conference software has not improved much in the last ten years. What has improved is the training provided. Having the training on how to conduct a lesson on-line makes a bigger difference than the features of the software. But there are some extra features in the web classroom systems.
While I now know enough to use Adobe Connect, I am not sure what I might use it for, or how we use it with other teaching software, particularly Moodle. As an example, Adobe Connect has provision for dividing a class into groups.
In theory I could have a virtual TEAL room. That is 100 to 200 students could be on-line watching an instructor perform some demonstration or min-lecture Then the class world divide into groups of about 6 to 9 students and each try the experiment themselves or discuss the topic. Then the class would come back together and the groups report. But that would take more support than Adobe Connect provides.
Wednesday, January 25, 2012
Reducing the bandwidth requirements of web conferences with Adobe Connect
- Set the maximum bandwidth to dial-up speed: Select Meetings > Preferences > Optimize Room Bandwidth and click on "Modem".
- Set the video to low resolution: Meetings > Preferences > Video and select Video quality low. This is sufficient to give the sense of someone being there. Particularly if you are not the main speaker, this should do.
- Set the audio to phone quality: By default the audio for conferences will be set higher than standard telephone quality audio (8,000 Hz). Unless you are conducting music lessons, setting this to telephone quality should be fine. This setting might improve the quality of the audio in some settings, as it will reject noise not in the range of the human voice. Select Meetings > Preferences > Audio and select the Speex Codec.
- Set push to talk, with half duplex sound: Select Meetings > Preferences > Audio and set the Echo cancellation mode to "Half Duplex". By default the audio should be set so that you have to push a button to be heard. This prevents stray room sounds disturbing the participants and saves on bandwidth. Also setting half duplex helps with clarity: that is when your microphone is turned on, your speakers are turned off: you cannot talk and hear at the same time. People have to take turns to speak, which works well for large groups.
Wednesday, November 09, 2011
Video enhancement software for poor eyesight
Video enhancement software for poor eyesight
Project Code: CECS_686
This project is available at the following levels:
CS single semester, Engn4200, Engn R&D, Honours, Masters, PhDKeywords:
TV, Video, accessibility, disability, vision, software development
Supervisor:
Mr Tom Worthington
Outline:
This projects investigates algorithms for
enhancing video images for people with poor eyesight, in particular macular degeneration. The software would enhance the import parts of images, such as people, particularly facial features and objects in the foreground. The result would be a cartoon-like effect with a solid line
around geometric shapes, such as faces, with and high contrast eyes, nose and mouth.
There has been extensive research on modifying video for people with limited eyesight, including by modification of open source MPEG decoding algorithms (1). Techniques for measuring the effectiveness of the modifications have been investigated (2).
However, the techniques used have assumed that the resulting image should still be acceptable to a person with normal vision. This limits the usefulness of the techniques for people with very limited eyesight.
Also it is assumed that there is limited processing capacity to run the enhancement algorithm and so only relatively simple techniques are used and that what is sued has to be suitable for mass deployment hard coded
into the TV set.
It should be feasible to produce cartoon-like images which would be easy to see with very limited eyesight. This should be feasible using the higher processing capacity now available in modern digital TV sets. Also it should be feasible to download the software into exiting TV sets via
an Internet connection. The same technique should be usable as a downloaded PC, smart phone and tablet "App". It should also be usable as an add-on for YouTube and simile web based video services.
The software would be developed at ANU CECS udner the supervision of Tom Worthington and tested
at the London Eye Hospital under the supervision of Lyndon Da Cruz.
The software would be made available for free use under an open source licence and the research findings published free online with a Creative Commons licence.Goals of this project
Understanding of the effects of image manipulation on vision.
Development of software for consumer devices.Background Literature
References:
1. Post Transmission Digital Video Enhancement for People with Visual Impairments Matthew Fullerton and Eli Peli J Soc Inf Disp. 2006 January; 14(1): 15–24. doi: 10.1889/1.2166829.
http://ukpmc.ac.uk/articles/PMC1413511.
2. Measuring perceived video quality of MPEG enhancement by people with impaired vision Matthew Fullerton, Russell L. Woods, Fuensanta A. Vera-Diaz, and Eli Peli J Opt Soc Am A Opt Image Sci Vis. Author manuscript; available in PMC 2008 December 1. PMCID: PMC2131737 Published in final edited form as: J Opt Soc Am A Opt Image Sci Vis.
2007 December; 24(12): B174–B187.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2131737/Links
Tom Worthington
Lyndon Da Cruz
Post Transmission Digital Video Enhancement for People with Visual Impairments
Measuring perceived video quality of MPEG enhancement by people with impaired vision
Contact:
Monday, November 07, 2011
Documentary in HTML5
Tuesday, November 01, 2011
National TelePresence System
The NTS was used for organizing the recent CHOGM meeting in Perth. It was claimed to have saved $12M and 2,330 tonnes of carbon emissions.
The NTS is also used for video conferences with state government officials for organizing routine meetings and during emergencies, such as floods. The system uses Telstra Next IPTM network and Cisco hardware and has been in operation since October 2009. The system cost $13.8 M over four years, with state government contributing $4.6M. There have been 1031 meetings totaling 1660 hours held.
Thursday, August 18, 2011
Giving an on-line conferecne presentation with web slides
Friday, July 15, 2011
Reducing the bandwidth requirements of web conferences
- Set the maximum bandwidth to dial-up speed: Collaborate has a minimum setting of 28.8 kbps. While slow by broadband speeds, this is still fast enough for two way audio, a small video window and slides. Even if your link has a higher theoretical speed, it may be worth setting this lower to allow a slow PC to keep up.
- Set the video to a few frames per second: Collaborate has a minimum setting of 1 frame per second. This is sufficient to give the sense of someone being there. Particularly if you are not the main speaker, this should do. You can also make the image lower resolution. It also helps to set the system to automatically highlight the video of the current speaker, so you know who is talking.
- Set the audio to 8,000 Hz: By default the audio for conferences will be set higher than standard telephone quality audio (8,000 Hz). Unless you are conducting music lessons, setting this to telephone quality should be fine. This setting might improve the quality of the audio in some settings, as it will reject noise not in the range of the human voice.
- Set push to talk, with half duplex sound: By default the audio should be set so that you have to push a button to be heard. This prevents stray room sounds disturbing the participants and saves on bandwidth. Also setting half duplex helps with clarity: that is when your microphone is turned on, your speakers are turned off: you cannot talk and hear at the same time. People have to take turns to speak, which works well for large groups.