Crackington Haven 2016, taken with IPhone 5S

I am a Reader in Computer Vision at the University of Bristol in the Department of Computer Science and a member of the Visual Information Laboratory (VIL) and the Bristol Robotics Laboratory (BRL). My research covers computer vision and its applications - robotics, wearable computing and augmented reality - and I have done a lot of work on 3-D tracking and scene reconstruction, mainly in simultaneous localisation and mapping (SLAM). Working with industry and on interdisciplinary projects is a high priority for me - please get in touch if you are interested in working with me. More details can be found below and in my publications.

Contact details: Department of Computer Science, University of Bristol, Merchant Venturers Building, Woodland Road, Bristol BS8 1UB, UK; T: +44 117 9545149; E: andrew at cs dot bris dot ac dot uk.


News


Research Assistants and Students

I enjoy working with people who want to discover, innovate and work with others to make things happen. If you are interested in working with me then please get in touch. If you want to do a PhD then I'd be happy to hear from you but please take a look at what I do and how we might work together before you contact me. If you are looking for funding, then any vacancies I have will be advertised on this page; otherwise, you may like to consider the various scholarships offered by the University.


Projects

AUTOMATED MAP READING USING SEMANTIC FEATURES
New research on localising images in 2-D cartographic maps by linking semantic information, akin to human map reading. Based on minimal binary route patterns indicating presence or absence of semantic features and trained networks to detect such features in images. Leads to highly scalable map representations.
[IROS 2018 paper][Project page]

OUT-OF-VIEW MODEL-BASED TRACKING
3-D model-based tracking which uses a trained network to predict out-of-view feature points, allowing tracking when only partial views of an object are available. Designed to deal with tracking scenarios involving large objects and close view camera motion.
[IROS 2018 paper]

HDRFUSION: RGB-D SLAM WITH AUTO EXPOSURE
RGB-D SLAM system which is robust to appearance changes caused by RGB auto exposure and is able to fuse multiple exposure frames to build HDR scene reconstructions. Results demonstrate high tracking reliability and reconstructions with far greater dynamic range of luminosity.
[3DV 2016 paper][Project page]

LDD PLACE RECOGNITION
Place recognition using landmark distribution descriptors (LDD) which encode the spatial organisation of salient landmarks detected using edge boxes and represented using CNN features. Results demonstrate high accuracy for highly disparate views in urban environments.
[ACCV 2016 paper][Project page]

MULTI-CORRESPONDENCE 3-D POSE ESTIMATION
Novel algorithm for estimating the 3-D pose of an RGB-D sensor which uses multiple forms of correspondence - 2-D, 3-D and surface normals - to gain improved performance in terms of accuracy and robustness. Results demonstrate significant improvement over existing algorithms.
[ICRA 2016 paper][Project page]

RGB-D RELOCALISATION USING PAIRWISE GEOMETRY
fast and robust relocalisation in an RGB-D SLAM system based on pairwise 3-D geometry of key points encoded within a graph type structure combined with efficient key point representation based on octree representation. results demonstrate that the relocalisation out performs that of other approaches.
[ICRA 2015 paper][Project page]