The following is a list of final year project ideas for undergraduate students. A separate list of MSc projects are published separately.
Room tracker via wifi with machine learning There is a whole industry and research area around activity tracking of people using sensors - e.g verifying that elderly people are safe and active in their home. This often involves the use of obtrusive sensors and privacy concerns. Interestingly, human reflect, or at least interfere with, wireless signals. So in theory, by examining the impact on wifi signals of a person or people in a room, you can develop a model to determine interesting things about what is going on in that space - such as number of people, whether there is movement and so forth. This would ideally be done using machine learning, training a model based on labelled data. This project involves building a proof of concept application that demonstrates how wifi can be used to predict/ determine behaviour in a space. This is a reasonably complex project, with plenty of scope to make it bigger if you wish - so it's a good one to pick if you're up for both research and implementation work.
A suite of useful mobile apps The idea here is to create two useful mobile apps: (1) Have you ever noticed how you end up with so many redundant photos on your phones, of images you only wanted for a few minutes or days - e.g. snapshotting a sticker to remember a code. They are not real images for long term keeping. This app starts with the simple premise of creating a camera app that deliberately wipes the camera image after a set time e.g. defaulting to a (changeable) duration such as day - but changeable per picture too. This app is too simple to be a project on its own, but might be the seeding of a larger project. (2) The second app is for those people who need to keep all either their SMS messages or Whatsapp (or just one) messages for an individual or group into some sort of record. This could be a formatted PDF "story book" style - with customisation options. The concept is based on people who want, for personal reasons, to save down their communcations with a person or group, for practical logging reasons - for personal memento reasons.
Automatic Cycle lane inspector
Develop an application that can automatically stream images and other information about a cycle lane. Physically, the application will include at least one sensor that attached to the bike - and which can take images (either static or video) and track GPS. As the cyclist moves along the cycle lane, images/ location are tracked building a data store that can be mined to generate useful knowledge about the cycling route - e.g. quality report about the cycle lane route - there are a variety of things that can be done here - driver proximity to the cyclist, analysis of the cyclist journey. But going with the basic idea, image processing will be needed to determine problems in the cycle lane (potholes), debris, even parked cars). A predictive model, trained on previous images could work well here.
Image topics
The idea of this project is to develop and application that can automatically group images into topics. E.g. if you have 100 images, 10 of them are kids, 12 of them are buildings, 10 are of ... etc .. the application would be able to assess the images and group them into related "topics". The likely method of working these would be to get the "objects of interest" out of the image (there are image processing models already built for this), and then derive a set of text words for each image. Standard topic modelling could then be used to group the images.
Image Filter This application would detect whether an image contains a risky image or not. "Risk" can be contained for the scope of this project e.g. you could have a taxonomy of "weapons", breaking into sub categories - guns, knives etc - and if an image contains any of these, it is classified as problematic (or not). By extending the taxonomy, and risk categories associated with it (like a rules base for determining risk), the app will work as a flexible image assessor, determining whether an image is okay or not. The scope would have to be limited to some specific domain - you couldn't build this for all possible image topics - for practical, technical and ethical reasons.
Automatic Topic Discovery in Text
There is a concept in computer science of automatic topic modelling
- whereby algorithms can analyse texts to automatically group them into topics.
The idea of this project is to apply, and possibly tune/optimise topic modelling
for a particular set of texts.
The student can pick a set of texts that they wish to use as their dataset
or we can use the first part of the project to research this.
The idea of this project is to apply, and possibly tune/optimise topic modelling
for a particular set of texts.
The student can pick a set of texts that they wish to use as their dataset
or we can use the first part of the project to research this.
Video analyser
There is an enormous volume of video posted online.
It is a difficult problem for videos to be automatically classified or analysed.
The idea of this project is to supply a tool that automatically extract information
about a video, with a view to helping identify the type of video without having to watch it.
The information extracted has to be decided but might include things such as
whether there is a lot of noise in the video, sound of people laughing,
cheering, silence, movement, talking, etc - and that these characteristics would in turn
allow the video to be automatically classified into some sort of category such as sports,
comedy, training video, news report etc. There is plenty of scope for
the student to decide how, and to what extent this problem can be solved.
This could be done using a rule based approach or a machine learning
approach (if labelled video data available).
It is a difficult problem for videos to be automatically classified or analysed.
The idea of this project is to supply a tool that automatically extract information
about a video, with a view to helping identify the type of video without having to watch it.
The information extracted has to be decided but might include things such as
whether there is a lot of noise in the video, sound of people laughing,
cheering, silence, movement, talking, etc - and that these characteristics would in turn
allow the video to be automatically classified into some sort of category such as sports,
comedy, training video, news report etc. There is plenty of scope for
the student to decide how, and to what extent this problem can be solved.
This could be done using a rule based approach or a machine learning
approach (if labelled video data available).
Identify a person in a crowd
This project is based on solving facial recognition problems.
In face recognition, a common problem is to compare two images of a face and decide
if they are the same person, or similar. A variation on this problem is to identify a person
in a crowd based on an image of their face.
Once in a crowd, their facial image is smaller, with less definition so it becomes
of task of producing some level of certainty as to whether they are the image or not.
A wider problem here is identifying whether someone appears in a video,
based on a static image of them.
based on a static image of them.
What can the government data tell us?
Right now, there are hundreds of datasets published by the government
at: https://data.gov.ie/data.
There are stacks of opportunities to analyse/ combine datasets in here to make
interesting discoveries. The first challenge is to pick an area of interest to you,
and then we can hone in on the project definition.
at: https://data.gov.ie/data.
There are stacks of opportunities to analyse/ combine datasets in here to make
interesting discoveries. The first challenge is to pick an area of interest to you,
and then we can hone in on the project definition.
Internet of Things: What's the data saying?
The internet of things is about ordinary objects being able to generate
data that it is measuring e.g. a mobile phone generating location over time data,
motion sensors, heart rate monitors etc.
The idea of this project is put meaning on data that is being generated
from sensors or devices, using machine learning. This project is related to getting data
from sensors and using a machine learning algorithm to interpret what the sensor
data means. The student can either gather their own dataset (and label it)
e.g. using their mobile phone data/ smart watch or other data.
Or they can find a labelled dataset of sensor data and use that one.
The project is deliberately not specific to any domain.
The student can choose that.
data that it is measuring e.g. a mobile phone generating location over time data,
motion sensors, heart rate monitors etc.
The idea of this project is put meaning on data that is being generated
from sensors or devices, using machine learning. This project is related to getting data
from sensors and using a machine learning algorithm to interpret what the sensor
data means. The student can either gather their own dataset (and label it)
e.g. using their mobile phone data/ smart watch or other data.
Or they can find a labelled dataset of sensor data and use that one.
The project is deliberately not specific to any domain.
The student can choose that.