Mebby a few interested so thought I'd mention it. Not sure if in right forum area but is a political and social issue so....
As some may know I been doing my PhD in Norway on Image processing and pattern recognition, am in my final year now and have submitted one paper and will shortly submit another.
@Winty and HugeArse if you're still around...thx for the advise to take it a couple years ago, you weren't wrong I'll never regret it
My main goal has been to develop software that can analyse real time video images, find objects and classify them based on their colour and shape.
Regions of interest (objects of any colour/shape) are found by modified Gaussian filtering and Canny edge detection colour and light intensity invariant calculations compensate for different light conditions. Mathematicians among you may know that n-th degree moments calculations can be used to describe a polygon of n-sides. (1st and 2nd degree soloutions are centre of area/gravity and principal axis). Anyway by this method I can describe any shape and a full analysis of the RGB colour space provides even more data. This colour and shape data is normalised and then analysed using canonical discriminant analysis, fuzzy logic and a neural network, then a majority vote gives a decision on what the object is. I can't put details here until I've been published but whenm I am I could post the papers if anyone interested
Really exciting now as is all working well, I've trained the software to find and recognise over 200 different types of objects - people, faces, balls, fish (21 different species), cars etc etc. Looks like something off Terminator when the system is running - objects are found, separated, analysed and classified on the screen real time as the video cameras runs. Have trained up some of the peeps at work and they seem tickled pink when they walk past the camera and are recognised by name and species by the computer - "species: human, name: Cato, height: 1.9m" etc
In August the system will be installed on a robotic submarine and used to help test self piloting functionality - search for new species of fish, look for underwater damage to oil pipes/platforms and control it's own movement using my software and GPS/ultrasound data. Futuristic stuff
I manage to max out the fx-57 processor no problem Looking into parallel processing now to improve power. Monster heavy calculations being done in real time (XP embedded custom OS), mean that until recently this sort of calculations and system would not have been possible. (@ programmers: heavily nested threads are a nightmare lol)
The company that pays my scholarship have asked me to stay with them for duration of my PhD - yipee another year of no rent or bills and have a job after. They will allow me to continue my research all funded after I finish too which is ace. My thesis is also to have a confidentiality note applied as the work is of commercial value now. We are getting heavy funding for the project now and have over £1.5m just for last year - not bad considering only me and one other guy do the work!
I'm doing a presentation at an international science conference in Maastricht in Sept if anyone there wants to come they are welcome :
As some may know I been doing my PhD in Norway on Image processing and pattern recognition, am in my final year now and have submitted one paper and will shortly submit another.
@Winty and HugeArse if you're still around...thx for the advise to take it a couple years ago, you weren't wrong I'll never regret it
My main goal has been to develop software that can analyse real time video images, find objects and classify them based on their colour and shape.
Regions of interest (objects of any colour/shape) are found by modified Gaussian filtering and Canny edge detection colour and light intensity invariant calculations compensate for different light conditions. Mathematicians among you may know that n-th degree moments calculations can be used to describe a polygon of n-sides. (1st and 2nd degree soloutions are centre of area/gravity and principal axis). Anyway by this method I can describe any shape and a full analysis of the RGB colour space provides even more data. This colour and shape data is normalised and then analysed using canonical discriminant analysis, fuzzy logic and a neural network, then a majority vote gives a decision on what the object is. I can't put details here until I've been published but whenm I am I could post the papers if anyone interested
Really exciting now as is all working well, I've trained the software to find and recognise over 200 different types of objects - people, faces, balls, fish (21 different species), cars etc etc. Looks like something off Terminator when the system is running - objects are found, separated, analysed and classified on the screen real time as the video cameras runs. Have trained up some of the peeps at work and they seem tickled pink when they walk past the camera and are recognised by name and species by the computer - "species: human, name: Cato, height: 1.9m" etc
In August the system will be installed on a robotic submarine and used to help test self piloting functionality - search for new species of fish, look for underwater damage to oil pipes/platforms and control it's own movement using my software and GPS/ultrasound data. Futuristic stuff
I manage to max out the fx-57 processor no problem
Looking into parallel processing now to improve power. Monster heavy calculations being done in real time (XP embedded custom OS), mean that until recently this sort of calculations and system would not have been possible. (@ programmers: heavily nested threads are a nightmare lol)The company that pays my scholarship have asked me to stay with them for duration of my PhD - yipee another year of no rent or bills and have a job after. They will allow me to continue my research all funded after I finish too which is ace. My thesis is also to have a confidentiality note applied as the work is of commercial value now. We are getting heavy funding for the project now and have over £1.5m just for last year - not bad considering only me and one other guy do the work!
I'm doing a presentation at an international science conference in Maastricht in Sept if anyone there wants to come they are welcome :
You're an intelligent bastard. The closest I ever came to pattern recognition was recognising the pattern made by my flies always coming undone while riding on public transport.
