Activity Analysis

Our research on activity analysis from video has focussed largely on human behaviour. We are interested in all aspects of this problem, including fundamental research on categorisation, tracking, segmentation and motion modelling, through to applied work addressing social and commercial priorities. Some of our recent work, in collaboration with the School's KRR group, is exploring the integration of vision within a broader cognitive framework, including audition, inductive reasoning, and intentionality.

[Detecting Carried Objects]

Relational Description of Video Scenes

Learn about the activities within a scene, and then about the objects involved in these activities

[Bicycle Theft Detection]

Bicycle theft detection

Recognising drop and pick events and linking related events applied to bicycles. This method could assist raising a warning when the person picking the bicycle does not 'look like' the person who dropped it

[Detecting Carried Objects]

Detecting carried objects

Detect large objects (e.g. bags) carried by pedestrians depicted in short video sequences

[Detecting Carried Objects]

Motion segmentation by consensus

Segment moving objects from tracked features

[Image of person playing against computer]

Learning to play table-top games

Learn about the objects and patterns of moves used in simple table-top games, and then apply these to play the game

[Image of intentional map]

Modelling the intentions of pedestrians

Detect atypical pedestrian trajectories, assuming a simple model of goal-directed navigational behaviour

[Image of WaterMark Extraction]

Watermark extraction

Localise paper-based watermarks using image processing


Earlier work - but still of interest...

[Image of cogvis game playing]


A generic object tracker, developed initially for tracking vehicles
[Image of Music via Motion]

Music via Motion

Create an augmented and interactive audio-visual space
[Image of soccer player tracking]

Soccer Tracking

Tracking the movements of individual players in team games
[Image of temporal continuity]

Temporal Continuity

Enforcing global spatio-temporal consistency to enhance reliability of moving object tracking and classification
[Image of cogvis game playing]

Traffic Interactions

Modelling traffic interaction using learnt qualitative spatio-temporal relations and variable length Markov models
[Image of synthetic head]

Synthetic Interaction

Synthesise an interactive agent by learning from the interactive behaviour of people

Electronic stockman's eye

Tracking cows and looking for abnormalities in their movements
[Image of old stuff]

VLMMs of behaviour

Learning variable length markov models of human behaviour


Detecting unusual events by modelling simple interactions between people and vehicles


Optical music recognition (with dept. of music)

Behaviour modelling

Modelling of `object behaviours' using detailed, learnt statistical models

Hand tracking

Robust, model-based methods for tracking deformable objects

Pedestrian tracking

Contour tracking using Active Shape Models.
Source code for the Baumberg tracker.