Argon Design
2019 proposal
The client is Imdad Sardharwalla imdad.sardharwalla@argondesign.com
suggestion
In a group call, Skype and Google Hangouts let you know who's talking by illuminating the speaker's name. But what if many of the participants are in the same room? Perhaps seated around a conference table? Your goal is to develop a system that is tailored to this scenario--it should be able to determine the speaker in a room of participants, and indicate them on a simple map of the room. Your client will provide a microphone array for you to work with.
feedback
(title - Room-Meet?)
I like the idea, and it’s nice for students to have a chance of working with a microphone array. I haven’t used these myself - what kinds of signal processing might students need to do, in order to achieve spatial localisation?
I would expect that one or two members of the team might focus on the localisation problem, in which case we’d need to think of other system aspects that the rest of the team might work on. For example, maybe it would be feasible to use a single wide-angle camera (or perhaps two or three, to cover a round table), and automatically zoom the video image in to the current speaker? Simple algorithms for face recognition might be sufficient to estimate the camera geometry in relation to audio localisation model.
2018 project
Contact: 'Steve Barlow' <steve.barlow@argondesign.com>
Client: Jack Haughton <jack.haughton@argondesign.com>
earlier ideas
I confirm we’d like to put forward a brief for a project for next year (Lent 2018).
My delay in replying was because I was chewing over ideas for possible projects. I’ve still not had an idea I’m really happy with so I thought I’d better reply now and then keep thinking.
An FPGA project is nice because it’s different technology. The downside is it takes more work to develop compared with software and so might not do something as obviously impressive. The reasons you would use an FPGA are if you ultimately want to implement something in hardwired logic for lower power or you need throughput or low-latency that is beyond a software solution. My current thought is a hardware corner detector that could provide input to a software SLAM algorithm.
I know Simon Moore from a number of years ago, when he was working on asynchronous processors. What FPGA boards do you have? I might alternatively provide a board if that is an easier way of connecting a camera and not having too many FPGA resource constraints.
Another possible non-FPGA project direction is “slow cameras”. If you only capture an image every hour, you can spend a long time doing image processing on it on a microcontroller. You don’t need a powerful system. So what could you do to create smart sensors using this?
I’ll continue to think over the next couple of months.
response
Simon Moore does still teach undergrad classes with FPGA boards, and we have quite often used these in the group design projects in the past. As you say, it is usually a little more difficult to bootstrap novel FPGA code to the level of getting an impressive application result, but our audience do appreciate the challenge, and give students credit for their achievements!
Information on their FPGA course is here: http://www.cl.cam.ac.uk/teaching/1718/ECAD+Arch/
They start with SystemVerilog, then FPGA synthesis using Altera Quartus and Qsys tools. The hardware is a DE1-SoC board from Terasic with our own custom I/O attached.
There are some further details of the hardware here. It looks as though it includes an LCD touch panel, but no camera input http://www.cl.cam.ac.uk/teaching/1718/ECAD+Arch/additional.html
Terasic seem to provide accessories including image capture devices: http://www.terasic.com.tw/cgi-bin/page/archive.pl?Language=English&CategoryNo=65#Category68
Hardware feature detectors for SLAM sound ambitious for undergrads! But would be exciting if this is feasible. If you want to pursue that, I should probably introduce you to one of Simon’s team, to discuss the level of skill that can be expected from students. I’ve had undergrads use SLAM libraries, but wouldn’t expect them to have much detailed understanding of the implementation in second year.
