Derek McAuley


Stuff everyone should know


Implementing algorithms to conduct event segmentation on image and sensor data collected by wearable cameras

With the arrival of ‘wearable’ technologies for capturing images and videos (such as the Sensecam derivative Autographer, Memoto, SpyGlasses, and the upcoming Google Glass), it is becoming possible to study behaviour remotely over longer periods of time, from a ‘first-person perspective’. Images collected on wearable cameras can provide great potential for assisting research studies on behaviour. However, such methods will produce large amounts of image and associated sensor data, and it is not yet clear what the best approach to utilize this data to study human behaviour are. One of the first steps in analysing first person perspective images is to be able to break them down into events. The project will focus on implementing published event segmentation algorithms and evaluating those utilising previously collected data sets from a wearable camera. Java on a standard computing platdform (Windows / Linux / MacOS) is the preferred language environment as the final target implementation is in an Android mobile phone (but that is beyond the scope of this project!).

Sanity for wireless laptops

Accessing public WiFi networking is more difficult than necessary as the comon method of "intercept the first hhtp request from the device  and get the user to login" presumes the first http request comes from a browser when in fact it is usually some application like email or dropbox that jumps on the network as soon as it is available - as the login webpage is not what the applicaiton expected it throws it on the ground, often with "Bad SSL Certificate" or some such nonsense. It also relies (at least EastMidlands trains does) on you running a browser with no proper cookie / adware / malware blockers in place. This project will provide a mechanism that blocks all processes from using a NIC until a specific web browser process enables it - having performed the WifFi login. This could be implmented using firewall rules or tun/tap (google it!) - the mechanism dependant on the OS (Linux, WIndows, MacOS) and student preference! Hacking the funcionality is not sufficient, it requires a usable by mortals installation, setup and configuration process.

Who's been eating my bandwidth?

Wireshark and tcpdump are powerful tools based on the widely available libpcap, to monitor traffic in and out of your computer / home / network - powerful, for those who know how to use them. However, crawling around in individual packets is for those skilled in the ways of networking. What the average user like to know is where is all the traffic is coming from / going to - "who's download World Of Warcraft update?", "which device is streaming HD content?", "what is talking to". The project is to use the widely available pcap library (Windows, Linux, MacOS - admin privileges required, so either your own computer or one can be supplied...) to develop a range of visualization tools that provide useful information on network behaviour to average home users, and indeed flag up things that might be consider bad news - like talking to bot net controllers. For bonus points, a student might like to consider the ethics of releasing such a tool to the general public and how might mitigate the risks in the design.