[ Home : Programs | libCVD | Hardware hacks | Publications | Teaching | TooN | Research ]

Academic work

I work in the field of computer vision in the machine intelligence lab in the engineering department of the University of Cambridge.

Research Topics



3B Microscopy

Sample 3B reconstruction of podosomes

Tracking vehicles from the air

Aggregate tracks

Object Detection and Grammar Guided Feature Extraction

Illustration of Grammar Guided Feature Extraction Object detection is still a difficult problem, especially when the objects are very small and ambiguous in appearance. The problem is tackled using the new Grammar-Guided features and optimization of all layers the system to maximize the object detection score.



Convert-XY was developed to enable writing programs in a natural mix of C++ and Python. It allows the mising of dynamic types with statically instantiated C++ templates, and maintains a natural type system in both languages.

FAST corner detection

Illustration of FAST corner detection

This work presenst the FAST (Features from Accelerated Segment Test) and the FAST-ER (FAST-Enhanced Repeatability) corner detectors. Full source code for both using existing detectors and creating new detectors is available.

Radial distortion autocalibration

Barrel distortion Undistorted image

Radial distortion can be deduced from a single image by finding parameters which make straight lines in the world straight in the image.

Surveillance and distributed sensor networks

Surveillance image


A diffraction pattern

Corner detector testing

Illustration of repeatability testing

The repeatability of a corner detector is an important factor in determining its performance. This work makes available a dataset for easil ymeasuring the repeatability of a corner detector.

Automatic label placement

Label placement in video

Label placement is an important topic in augmented reality. This work presents an efficient method for unobtrusice placement of labels in live video without the need for a model.

3D Model Based Tracking

Tracking a lab space

3D tracking can be made to be very robust by fusing corner point and edge based information. Additionally, curved 3D object can be tracked from a model with edge based tracking and a hidden line renderer.

Updated February 26th 2014, 05:49