[Univ of Cambridge] [Dept of Engineering]


Dr Tom Drummond

Senior University Lecturer in the Machine Intelligence Laboratory
in the Division of Information Engineering at
Cambridge University Engineering Department.
Also a fellow of St Catharine's College 		Small Photo
Teaching 3F5: Software Architecture
IB Computing exercise [exercises]
IIA Image Processing Project SF2 [Handout] [download Matlab files] Image for Competition Function to display images for competition.
IA Microprocessor Lab Assembler files
Software TooN (Tom's object oriented numerics library)
CVD (Cambridge Video Dynamics)
Research Group PostDocs, PhD Students and Former Students
Research Publications

High speed feature matching

High speed feature matching (Simon Taylor and Ed Rosten)

This work presents a novel local feature matching method designed with a focus on runtime speed. This enables frame-rate localisation of known targets on low-powered devices such as mobile phones. This work won the Best Demo prize at CVPR 2009 . [More]

[2009 CVPR Workshop on Feature Detectors and Descriptors Paper] [2009 BMVC Paper]
[Video showing operation] [Video showing target with few features] [Video showing multiple targets]
ProFORMA: Probabilistic Feature-based On-line Rapid Model Acquisition

ProFORMA: Probabilistic Feature-based On-line Rapid Model Acquisition (Qi Pan and Gerhard Reitmayr)

The generation of 3D models of real objects is very useful for many computer vision applications. This paper introduces ProFORMA, a system designed to enable on-line reconstruction of textured 3D objects rotated by a user's hand. Partial models are created very rapidly and displayed to the user to aid view planning, as well as used by the system to robustly track the object pose. The system works by calculating the Delaunay tetrahedralisation of a point cloud obtained from on-line structure from motion estimation which is then carved using a recursive probabilistic algorithm to rapidly obtain the surface mesh. This work won the Best Demo prize at ISMAR 2009. [More]

[2009 BMVC Paper] [Real-time system operation]
Reconstruction from Uncalibrated Affine Silhouettes

Reconstruction from Uncalibrated Affine Silhouettes (Paul McIlroy)

This work addresses the problem of model building from multiple affine silhouette views of an object in an uncontrolled environment such as an aircraft in flight. Each pair of silhouette views provides two outer epipolar tangency constraints on the relative motion between the cameras. For a scaled orthographic camera model with six degrees of freedom we show that it is possible to recover structure and motion from six or more silhouette views by solving the outer epipolar tangency constraints simultaneously. [More] This work won the Best Student Poster award at BMVC 2009

[2009 BMVC Paper]
Parametric Non-Gaussian Dynamical Filters

Parametric Non-Gaussian Dynamical Filters (James Loxam)

This work introduces the Student-t Mixture Filter as a robust method of filtering, capable of handling erroneous measurements and process models in a Bayesian manner without the need for ad-hoc outlier decisions and pre-filtering.

[2008 ECCV Paper] [video comparing our filter with others]
FAST Corner Detector

Natural feature tracking on mobile phones: (Gerhard Reitmayr and Daniel Wagner)

This paper presented two methods (Ferns and SIFT) for fast matching of feature points between live video and a stored planar scene. The methods ran at more than 10Hz on mobile phones with limited processing capability. This work won the best paper prize at ISMAR 2008.

[Prizewinning 2008 ISMAR paper]
SLAM as a graph of independent sets of observations

SLAM as a graph of independent sets of observations (Ethan Eade)

This work represents a new approach to SLAM, and has been specifically designed to address the consistency problem suffered by EKF and particle filter approaches. This technique also has the benefit that it permits much larger maps to be acquired and used.

[2007 ICCV Paper] [video]
SLAM-based Augmented Reality

SLAM-based Augmented Reality (Gerhard Reitmayr and Ethan Eade)

This work enables Augmented Reality applications to operate in unknown environments. The environment is mapped simultaneiously with localisation of the camera using a SLAM system which also assists the user in the placement of authored annotations. A key application domain for this technology is the provision of remote expertise. The system supports this use by communicating the video and annotations over a network link, thus providing an enriched communication channel between user and remote expert. [More]

[2007 ISMAR Paper] [A mac repair video] [Annotations by both local user and remote expert]
Edge landmarks in monocular SLAM

Edge landmarks in monocular SLAM (Ethan Eade)

This work allows edgelet landmarks to be included within SLAM systems by providing a method for detecting, parametrizing and tracking them.

[2006 BMVC Paper] [Real-time operation on a desktop sequence]
SLAM

Simultaneous Localisation and Mapping (SLAM) (Ethan Eade)

This work applies a Rao-blackwellised particle filter (the FastSLAM algorithm) to the monocular SLAM problem.

[2006 CVPR Paper] [Real-time operation on an indoor sequence]
Robust Model-based Tracking for Outdoor Augmented Reality

Going out: Robust Model-based Tracking for Outdoor Augmented Reality: (Gerhard Reitmayr)

This work performs edge-based visual tracking in outdoor environments. The system uses a textured model which is rendered using GL. Edges are extracted from this rendering and correspondences are then found in the live video, thus providing automatic detail culling. The system also exploits inertia information and recovery mechanisms based on stored key-frames to provide a robust tracking solution for outdoor augmented reality. [More]

[2006 ISMAR Paper]
PDAs as Tangible Interfaces

PDAs as Tangible Interfaces: (Jeremiah Neubert)

This method we created identifies handheld devices (e.g. smart phones and pocket PCs) to facilitate the use of these devices as tangible interfaces for desktop augmented reality systems. The proposed system leverages the ability of these handheld devices to programmatically control their backlight intensity to display a binary code. The codes produced are non-intrusive, require no specialized hardware, and can be generated with most handheld devices. This technique is shown to accurately and robustly identify up to 16 different devices in under 500 msec and is easily expandable to 256 or more devices.

[2006 ISMAR Paper] [Video]
FAST Corner Detector

FAST Corner Detector: (Ed Rosten)

This high speed interest point (corner) detector is based on considering pixels in a circular ring around the pixel of interest. A carefully chosen criterion allows the majority of non-corners to be rapidly rejected. The resulting algorithm can process VGA resolution images in a small number of milliseconds to produce very stable sets of interest points. [More]

[2006 ECCV paper]
Augmented Maps

Augmented Maps: (Gerhard Reitmayr and Ethan Eade)

Paper maps are much higher resolution than computerised maps and can be more readily manipulated. However, because they are physically printed they can only show static information. This work makes use of a camera-projector system to allow overlay of dynamic information on paper maps placed on a table surface. Tangible user interface tools are supported in a manner which allows multiple concurrent users to interact with the same map. [More]

[2005 ISMAR Paper]
A Single Frame Gyroscope

A Single Frame Gyroscope (Georg Klein)

Instead of using rate gyroscopes to predict camera motion in a frame, we attempt to estimate this information from the video input by analysing the motion blur present in a single video frame. By looking at the parts of the image which remain un-blurred, blur direction and magnitude can be determined very quickly (just over 2 milliseconds per frame on a 2.4GHz machine). This method only produces a 3-DOF answer and has a sign ambiguity; also the magnitude of blur is often inaccurate. However the system can usefully be combined with edge tracking. This work won the Industry Prize at BMVC 2005.

[Prizewinning 2005 BMVC Paper]
Hybrid Tracking for Man-machine Interfaces

Hybrid Tracking for Man-machine Interfaces: (Jeremiah Neubert)

This work focuses on developing hybrid tracking and spatial reference technologies which can be combined to deliver new human-machine interfaces. The project focuses on creating algorithms for tracking objects so that graphics can be overlaid on the image allowing a user to interact with it. The object localization system utilizes both image tracking and an inertial rate gyroscope unit to robustly track objects.

[140M Video] [4.2M Video]
Visually guided aerial robotics

Visually guided aerial robotics: (Chris Kemp)

This quad-rotor helicopter carries a miniature transmitting video camera. The video stream is received and fed into a computer which uses a visual tracking algorithm to compute the helicopter's position and generate control signals. These are sent to the remote control unit to stabilise the helicopter. The inset in the top right of the video shows the computer's view from the on-board camera.

[2005 ICCV paper (detailing the tracking)] [Video in corridor] [Video in lab] [Video showing response to disturbance] [Chris Kemp's PhD Thesis]
Sensor Fusion and Occlusion Refinement for Tablet-based AR

Sensor Fusion and Occlusion Refinement for Tablet-based AR (Georg Klein)

Tablet PCs offer an alternative to head-mounted displays for delivering augmented reality. This work shows a tablet-based AR application which combines inside-out edge tracking and outside-in LED tracking for robust registration: the inside-out system provides a high level of registration accuracy while the outside-in system provides robustness and recovery from shake and camera occlusions. To provide a high level of rendering quality, we look at the specific case of virtual graphics occluded by real objects for which we have a model. We show that instead of just clipping virtual graphics using the real geometry projected into the z-buffer, an individual treatment and refinement of occluding edges produces a far more convincing integration of real and virtual objects.

[2004 ISMAR Paper]
Multi-modal RANSAC

Multi-Modal RANSAC (Chris Kemp)

This work uses RANSAC as the proposal distribution to generate samples drawn from the posterior distribution in a filtering framework. This technique allows the pose estimation problem to be partitioned such that robust estimates can be obtained with low computational overhead. A multi-modal texture changepoint detector is also introduced. This work won the Science Prize at BMVC 2004.

[Prizewinning 2004 BMVC Paper] [2008 IVC Paper]
ProFORMA: Probabilistic Feature-based On-line Rapid Model Acquisition

HMD-based AR (Georg Klein)

Edge-based visual tracking was applied to a system with a head-mounted camera and intertia unit. The system is used to calibrate a user's view of the world through a Head Mounted Display (HMD) with computer graphics displayed on the HMD. A protoype application was then developed which displayed instructions to the user over real world objects.

[2003 ISMAR Paper] [video of prototype application]
Integrating Visual and Inertial Tracking

Integrating Visual and Inertial Tracking: (Georg Klein)

This work combines visual edge-based tracking with inertial sensors. The inertial sensors give a prediction of both the motion of the camera and the blur that should be present in the image. The edge detectors in the visual tracker are then be adapted to search for edges with the appropriate amount of blur for each match. The results of the visual tracker are also then used to calibrate the bias and drift in the intertial sensors. This work won the Industry Prize at BMVC 2002.

[Prizewinning 2002 BMVC Paper] [2004 IVC Paper]
Markerless Human Tracking

Markerless Human Tracking:

Articulated tracking can be applied to tracking of humans. For this work we used three synchronised cameras (fed into an SGI O2 as the red, green and blue planes of a digital 4:2:2 image. Tracking then imposes the additional constraint that the motion observed by all three cameras must be the same. The green square was for anonymity, but I have no means of removing it now!

This is interesting because it worked at all, not because it worked well.

[2001 ICCV Paper] [Video from first camera] [Video from second camera] [Video from third camera]
Articulated Tracking

Articulated Tracking:

Articulated tracking is achieved by tracking the components individually and subsequently imposing the articulation constraints on the component motions. The most probable pose that satisfies the constraints is then found. [More]

[2000 ECCV Paper] [Video of simple hinge] [Video of filing cabinet]
Edge Based Tracking

Edge Based Tracking:

This was our first research into edge-based tracking based on the Harris' RAPiD tracker. The videos below were made in 1999 on a 225 MHz SGI. This work won the Industry prize at BMVC 1999 [More]

[Prizewinning 1999 BMVC Paper] [Video1] [Video2 (with occlusion)]
Contact Information
E-mail: twd20@eng.cam.ac.uk
Address: Department of Engineering. University of Cambridge
Trumpington Street, Cambridge CB2 1PZ
Telephone: +44 1223 7 65960
Fax: +44 1223 3 32662
[ Cambridge University | CUED | MIL ]

T.W. Drummond - twd20@cam.ac.uk