I received my Master of Engineering in Artificial Intelligence degree from Imperial College London in 2009 and currently study towards an Engineering Doctorate in the Doctoral Training Centre for Virtual Environments, Imaging and Visualisation at University College London. I also have a prior experience of working in the corporate environment of investment banks and financial software houses. My main interests cover real-time 3D graphics, professional photography and human-computer interaction.
Under the supervision of Prof. Anthony Steed and in cooperation with Arup, I am looking into large scale distribution of real-time 3D architectural geometry and network streaming while investigating the ad-hoc visibility and various culling methods trying to improve on the performance limiting the network bandwidth requirements. Furthermore, I will study rapid modelling of behaviour for interactive systems in terms of editing tools targeted at specific project requirements.
Abstract
We propose a non-linear concurrent revision control for centralised management of 3D assets and a novel approach to mesh differencing. Large models are decomposed into individual scene graph (SG) nodes through an asset import library and become versioned as collections of polymorphic documents in a NoSQL database (DB). Well-known operations such as 2- and 3-way diff and merging are supported via a custom DB front-end. By not relying on the knowledge of user edits, we make sure our system works with a range of editing software. We demonstrate the feasibility of our proposal on concurrent 3D editing and conflict resolution.
Abstract
Even though revision control has been successfully deployed for text-based files for many years, it does not efficiently map to 3D assets. In this research note we propose a non-linear concurrent revision control system for 3D assets. Scenes are represented as collections of polymorphic objects alongside their corresponding history in a database, which map existing revision paradigms to a scene graph manipulation. Our open source implementation leverages a NoSQL database and a 3D asset import library. Similarly to existing version control systems it works independently from editing software and does not require any knowledge of user edits. We demonstrate the feasibility of our framework by performing concurrent 3D model modifications.
Abstract
Time-of-Flight cameras provide high-frame-rate depth measurements within a limited range of distances. These readings can be extremely noisy and display unique errors, for instance, where scenes contain depth discontinuities or materials with low infrared reflectivity. Previous works have treated the amplitude of each Time-of-Flight sample as a measure of confidence. In this paper, we demonstrate the shortcomings of this common lone heuristic, and propose an improved per-pixel confidence measure using a Random Forest regressor trained with real-world data. Using an industrial laser scanner for ground truth acquisition, we evaluate our technique on data from two different Time-of-Flight cameras. We argue that an improved confidence measure leads to superior reconstructions in subsequent steps of traditional scan processing pipelines. At the same time, data with confidence reduces the need for point cloud smoothing and median filtering
Lego robot that follows the right wall until it reaches a white target in the National Institute of Informatics, Tokyo. More information about the trip and my participation can be found at: http://www-typo3.cs.ucl.ac.uk/cs_news/ucl_cs_completes_first_international_software_engineering_event/
The whole panorama can be seen at: http://web4.cs.ucl.ac.uk/staff/j.dobos/pano/6.22-15k-squared-compressed.html
Third year group project implemented in cooperation with Carmen Fan, Isaiah Fan, Charence Wong and Matthew Ko runs a delivery simulator based on a custom built maps. My contribution was the entire GUI design out of which the most interesting was the magnifying glass that bends Bezier curves (transport tracks) based on a simple inverse of a square root as a function of a distance from the mouse pointer. The bigger the magnifying circle, the greater the pushing force. This simple yet elegant solution provides fake 3D feel. Changing colours of the circles and their outlines represent station saturation by the trains and accumulated undelivered parcels that are being randomly generated. Trains (red numbered moving squares) behave according to selectable AI logic.
With Charence Wong we managed to get through several iterations of different prototypes of self-balanging Lego NXT robots while studying our 3rd year undergraduate at the Department of Computing, Imperial College London.
Here, you can see our initial approach firstly using two light sensors evaluating the differences of perceived values. The second improved prototype incorporated a gyro and values from the motor encoders. However, such solution accumulates drift over time, so we had to acquire additional sensor, an accelerometer, in order to fuse the readings, so that we knew, where the level was thanks to gravitation.
This demo was presented during interview days in DoC at Imperial College for several years. The corresponding slides can be found at http://docs.google.com/present/view?id=df652g29_194f5rnhsfg.
http://www.doc.ic.ac.uk/teaching/prizes/2007/ug/index.htm
