Research

I have worked with many areas of Human-Computer Interaction, including: accessible technology, wearable devices, gesture interaction, ultrasound haptics, and novel displays. This page gives an overview of my research in these areas.

Acoustic Levitation

I currently work on the Levitate project, which is investigating novel interfaces composed of objects levitating in mid-air. My research here is mostly focused on developing new interaction techniques for levitating object displays [1]; for example, Point-and-Shake, a selection and feedback technique presented at CHI 2018 [2].

Ultrasound Haptic Feedback

I have been working with ultrasound haptic feedback since 2012. My ICMI 2014 paper [3] used an early prototype of an Ultrahaptics device to give haptic feedback about mid-air gesture input for smartphones. My World Haptics Conference 2019 paper was about helping users position their hands for optimal ultrasound haptic output. I have also investigated the perception of this novel type of feedback, to better understand how it can be used effectively. I chaired sessions at the CHI 2016 and CHI 2018 workshops on mid-air haptics and displays, and have chaired paper sessions at CHI 2018 and CHI 2019 about touch and haptic interfaces.

Gesture Interaction

My PhD research focused on gesture interaction, including around-device interaction. Some of my early PhD work looked at above-device interaction with mobile phones [3, 4], which I discuss more here. Towards the end of my PhD I also studied gesture interaction with simple household devices, like thermostats and lights, which present interesting design problems due to their lack of screens or output capabilities [5].

I am particularly interested in how gesture interaction techniques can be improved with better feedback design. Whereas most gesture interfaces rely on visual feedback, I am more interested in non-visual modalities and how these can be used to help users interact more easily and effectively. I have looked at tactile feedback for gesture interfaces [3, 6]; this is a promising modality but requires novel hardware solutions to overcome the challenges of giving tactile feedback remotely. I have also looked at other types of output, including sound and interactive light, for giving feedback during gesture interaction. My PhD research in this area was partly funded by a studentship from Nokia Research in Finland.

Despite significant advances in gesture-sensing technology, there are some fundamental usability problems which we still need good solutions for. My PhD thesis focused on one of these problems in particular, the problem of addressing gesture systems. I write more about this here. My CHI 2016 paper [5] describes interaction techniques for addressing gesture systems.

Above-Device Gestures

Early in my PhD I looked at above-device gesture design. We asked users to create above-device gestures for some common mobile phone tasks. From the many gesture designs gathered in that study, we then created and evaluated two sets of gestures. We created design recommendations for good above-device interfaces based on the outcomes of these studies [4].

Tactile Feedback for Above-Device Interaction

Tactile Feedback for Gestures

Small devices, like mobile phones and wearables, have limited display capabilities. Gesture interaction, being very uncertain for users, requires feedback to help users gesture effectively, but giving feedback visually on small devices constraints other content. Instead, other modalities – like sound and touch – could be used to give feedback. However, an obvious limitation with touch feedback is that users don’t always touch devices that they gesture towards. We looked at how we could give tactile feedback during gesture interaction, using ultrasound haptics and distal feedback from wearables [3].

Interactive Light Feedback for Gestures

Another way of giving visual feedback on small devices without taking away limited screen space is to give visual cues in the space surrounding the device instead. We embedded LEDs in the edge of some devices so that they could illuminate surrounding table or wall surfaces, giving low fidelity – but effective – visual feedback about gestures. We call this interactive light feedback [7]. As well as keeping the screen free for interactive content, these interactive light cues were also noticeable from a short distance away. For more on this, see Interactive Light Feedback.

Photo of the HaptiGlow system. An Ultrahaptics UHEV1 device with a strip of LEDs around the front edge and left and right sides. The LEDs are green, indicating that the user has their hand in a good position.

Wearables for Visually Impaired Children

I worked on the ABBI (Audio Bracelet for Blind Interaction) project for a year. The ABBI project developed a bracelet for young visually impaired children; when the bracelet moved, it synthesised sound in response to that movement. The primary purpose of the bracelet was for sensory rehabilitation activities to improve spatial cognition; by hearing how other people and themselves moved, the children would be able to improve their understanding of movement and their spatial awareness.

Concept visualisation of the ABBI bracelet and Audible Beacons.

My research looked at how the capabilities of the ABBI bracelet could be used for other things. The bracelet had motion sensors, Bluetooth communication, on-board audio synthesis and limited processing power, so my research investigated how these might facilitate other interactions. Some of my work looked at how Bluetooth beacons could be used with a wearable device to present relevant audio cues about surroundings, to help visually impaired children understand what is happening nearby [8]. I also considered how the bracelet might be used to detect location and activity within the home, so that the lighting could be adapted to make it easier to see, or to draw attention to specific areas of the home [9].

Reminders: Tabletops and Digital Pens

Before starting my PhD I worked on two projects looking at home-care reminder systems for elderly people. Reminders can help people live independently by prompting them to do things, such as taking medication or making sure the heating is on, and helping them manage their lives, for example reminding them of upcoming appointments or tasks such as shopping.

Tabletops in the Home

My final undergraduate project looked at how interactive tabletops could be used to deliver reminders. People often have coffee tables in a prominent location within the living room, making the tabletop an ideal display for ambient information and reminders. We wanted to see what challenges had to be overcome in order for tabletops to be an effective reminder display. One of the interesting challenges this project addressed was how to use the tabletop as a display and as a normal table. Clutter meant large parts of the display were often occluded so a solution was needed to allow reminders to be placed in a noticeable location. Part of this project was presented as an extended abstract at CHI 2013 [10].

Digital Pen and Paper Reminders

After graduating with my undergraduate degree, I worked on the MultiMemoHome project as a research assistant. My role in the project was to design and develop a paper-based diary system for digital pens which let users schedule reminders using pen and paper. Reminders were then delivered using a tablet placed in the living room. We were interested in using a paper-based approach because this was an approach already favoured by elderly people. We used a co-design approach to create a reminder system, Rememo, which we then deployed in peoples’ homes for two weeks at a time. This project was presented as an extended abstract at CHI 2013 [11] and as a workshop paper at Mobile HCI 2014 [12].

Predicting Visual Complexity

As an undergraduate I received two scholarships to fund research over my summer holidays. One of these scholarships funded research with Helen Purchase into visual complexity. We wanted to find out if we could predict how complex visual content was using image processing techniques to examine images. We gathered both rankings and ratings of visual complexity using an online survey and used this information to construct a model using linear regression with a collection of image metrics as predictors. This project was presented at Diagrammatic Representation and Inference 2012 [13] and Predicting Perceptions 2012 [14].

Aesthetic Properties of Graphs

An earlier research scholarship also funded research with Helen Purchase, this time looking at aesthetic properties of hand-drawn graphs using SketchNode, a tool which lets users draw graphs using a stylus. We devised a series of aesthetic properties describing graph appearance and created algorithms to measure these properties. Aesthetic properties included features such as node orthogonality (were nodes placed in a grid-like manner?), edge length consistency (were edges of similar length?) and edge orthogonality (were edges largely perpendicular and arranged in a grid-like manner?). I produced a tool to analyse a large corpus of user-drawn graphs from earlier research studies.

References

[1] Levitating Particle Displays with Interactive Voxels
E. Freeman, J. Williamson, P. Kourtelos, and S. Brewster.
In Proceedings of the 7th ACM International Symposium on Pervasive Displays – PerDis ’18, Article 15. 2018.

[2] Point-and-Shake: Selecting from Levitating Object Displays
E. Freeman, J. Williamson, S. Subramanian, and S. Brewster.
In Proceedings of the 36th Annual ACM Conference on Human Factors in Computing Systems – CHI ’18, Paper 18. 2018.

[3] Tactile Feedback for Above-Device Gesture Interfaces: Adding Touch to Touchless Interactions
E. Freeman, S. Brewster, and V. Lantz.
In Proceedings of the International Conference on Multimodal Interaction – ICMI ’14, 419-426. 2014.

[4] Towards Usable and Acceptable Above-Device Interactions
E. Freeman, S. Brewster, and V. Lantz.
In Mobile HCI ’14 Posters, 459-464. 2014.

[5] Do That, There: An Interaction Technique for Addressing In-Air Gesture Systems
E. Freeman, S. Brewster, and V. Lantz.
In Proceedings of the 34th Annual ACM Conference on Human Factors in Computing Systems – CHI ’16, 2319-2331. 2016.

[6] Towards Mid-Air Haptic Widgets
E. Freeman, D. Vo, G. Wilson, G. Shakeri, and S. Brewster.
In CHI 2016 Workshop on Mid-Air Haptics and Displays: Systems for Un-instrumented Mid-Air Interactions. 2016.

[7] Illuminating Gesture Interfaces with Interactive Light Feedback
E. Freeman, S. Brewster, and V. Lantz.
In Proceedings of NordiCHI ’14 Beyond the Switch Workshop. 2014.

[8] Audible Beacons and Wearables in Schools: Helping Young Visually Impaired Children Play and Move Independently
E. Freeman, G. Wilson, S. Brewster, G. Baud-Bovy, C. Magnusson, and H. Caltenco.
In Proceedings of the 35th Annual ACM Conference on Human Factors in Computing Systems – CHI ’17, 4146-4157. 2017.

[9] Towards a Multimodal Adaptive Lighting System for Visually Impaired Children
E. Freeman, G. Wilson, and S. Brewster.
In Proceedings of the 18th ACM International Conference on Multimodal Interaction – ICMI ’16, 398-399. 2016.

[10] Messy Tabletops: Clearing Up the Occlusion Problem
E. Freeman and S. Brewster.
In CHI ’13 Extended Abstracts on Human Factors in Computing Systems, 1515-1520. 2013.

[11] Designing a Smartpen Reminder System for Older Adults
J. Williamson, M. McGee-Lennon, E. Freeman, and S. Brewster.
In CHI ’13 Extended Abstracts on Human Factors in Computing Systems, 73-78. 2013.

[12] Rememo: Designing a Multimodal Mobile Reminder App with and for Older Adults
M. Lennon, G. Hamilton, E. Freeman, and J. Williamson.
In Mobile HCI ’14 Workshop on Re-imagining Commonly Used Mobile Interfaces for Older Adults. 2014.

[13] An Exploration of Visual Complexity
H. C. Purchase, E. Freeman, and J. Hamer.
In Diagrammatic Representation and Inference, 200-213. 2012.

[14] Predicting Visual Complexity
H. C. Purchase, E. Freeman, and J. Hamer.
In Predicting Perceptions: The 3rd International Conference on Appearance., 62-65. 2012.

Occlusion Management

Overview


My final undergraduate project investigated the challenges involved when introducing interactive tabletops into the home. This project was motivated by home-care applications, specifically delivering home reminders to a tabletop in the living-room.

Interactive tables would likely play the role of coffee table in the living-room, and as such would likely be as cluttered as our normal coffee tables are. Our initial investigation into regular surface use confirmed that clutter is, indeed, a problem! This poses a problem for home-care applications – important information may not be visible because of items on the tabletop.

A key aspect of this project was therefore to try and deal with the issue of messy tabletops when delivering information and home-care reminers. I approached this by using the internal cameras in a diffused illumination tabletop to capture the footprint of items on the table surface (shown as reflected infrared light). This information was then used to find a suitable visible area of the display to show reminders in.

The algorithm used to find a visible area of space is published in the Extended Abstracts of CHI 2013 [1], along with a discussion of some of the issues we encountered when designing for the home. This is a first step into investigating how to design for tabletops in the home.

For a video demonstration, please see:

[1] Unknown bibtex entry with key [MessyTabletopsWIP]

Algorithm


To capture the 2D footprint of items on the table surface, we first look at an image from the internal infrared cameras in the Microsoft Surface 1.0 tabletop. An example of such an image is shown below. White regions in this image are reflected infrared light from items atop the table.

ir

Image processing techniques can then be used to discover these item footprints, either using a blob detection algorithm or simple colour thresholding. We create a binary matrix where a value of 1 represents an item and a value of 0 represents a visible part of the display. The image below demonstrates how such a matrix may appear.

matrix

Our algorithm then finds visible rectangular regions suitable for display, using a dynamic programming approach. This algorithm iterates over each cell in the matrix and looks to see how far it can extend a border to the top, left and right. These borders will represent the largest rectangular region which can be “grown” from this cell. Border positions for previous rows and cells are re-used where appropriate. This is more efficient than a naive brute-force approach.

The image below shows the largest region found in the matrix shown previous. Here, the largest area rectangle was chosen; more suitable heuristics may be appropriate.

maxArea1

Design issues


During the development and implementation of this solution we came across several design issues.

1. Content placement

As described above, we just used the largest free space on the table for displaying content. Future work could look at more appropriate content placement on the table surface. For example it may be better to place content so it is closest to the user or in a position which is considered easiest to notice.

2. Changing position of content

When first implementing this solution we just moved content to a new position on the screen immediately. Some users noted that they were not sure if this was old or new information being shown. To address this we used animation; content being moved was animated along a smooth path so that it appeared to slide into view, whereas new content was gradually faded in. There may be more appropriate ways to identify content being moved to a visible location.

3. Occlusion caused by hands

Our initial implementation just used the raw image from the tabletop cameras alone to detect items atop the table. There is an obvious problem with this approach – the hand would appear as an item and content would shift away from fingers trying to interact with it! To prevent this, we used the information about touch contact points provided by the Surface 1.0 SDK to remove fingers from the item matrix.

4. Very messy tables

Sometimes clutter just builds and builds and builds until the table surface is entirely covered! In such cases other modalities would have to be used to inform the user that content is appearing on the display. Ambient lights placed around the table surface, for example, could glow to inform the user that information is shown on screen. Alternatively, audio or music notifications could be used.

5. Intentional occlusion

Occlusion can often be meaningful. Digital and physical documents may be grouped together on the tabletop; in this case the occlusion is semantically meaningful. Future work could look at how occlusion management could be toggled to allow semantically meaningful occlusion, preventing these piles of related content from being rearranged on the surface.

6. 3D Occlusion

A brief photo-study performed during this project investigated how people use coffee tables in their homes. We found that 3D occlusion (where parts of the surface are hidden due to the height of items atop the table) was not a significant problem. Our algorithm could easily be extended to deal with 3D occlusion – areas of the display in the shadow of items atop the table could be included in the occlusion matrix.

Source code


A C# implementation of the algorithm to find visible regions of the display is available here. The Region and Matrix classes shown are basically a rectangle and 2D array.

Paper & Poster


This paper was presented as a poster at CHI 2013 in Paris.