Motion Segmentation and Geometric Motion Modelling
Designed and developed a pipeline to distinguish small moving objects from camera-induced background motion. Developed a geometric motion model with flow representations and training losses.
About
I am a computer vision and machine learning researcher with a background spanning software engineering, applied research, and real-world problem solving. My work has progressively moved from applying established engineering tools toward formulating problems from first principles and developing methods around the mathematical and physical structure of the data.
A strong classical foundation in visual and multimedia data has shaped how I approach perception problems. I pay close attention to how data is acquired, represented, compressed, transformed, and degraded; to differences between sensing modalities; and to what information is actually preserved before a model ever sees the input. This perspective has repeatedly proved as important as model selection itself.
Over time, working on increasingly complex and application-driven research problems changed the way I solve problems. Rather than beginning with an existing algorithm and asking whether it can be adapted, I increasingly start from the measurement process: what has physically been observed, what mathematical relationships are present in the data, which information is reliable, and what computational formulation follows from those properties.
I see research as a disciplined process of formulation, analysis, experimentation, validation, and refinement, not as the application of increasingly complex tools. My current direction is toward mathematics-driven computer vision and machine perception: strengthening and connecting the mathematical areas most useful to perception, and using them to derive representations, models, and algorithms for problems where standard pipelines are insufficient.
My long-term goal is to work at the intersection of mathematics, computer vision, machine learning, and intelligent perception, particularly on technically demanding problems where a deeper understanding of the measurement and the structure of the problem can lead to simpler, more reliable, and more deployable solutions.
My work connects mathematical reasoning and an understanding of a problem’s physical and dynamic behaviour with practical challenges in visual perception, signal analysis, and event-based vision. I begin by examining what the sensor measures and identifying the relationships within the data. Those relationships guide problem formulation, representation design, and the choice of model. I am particularly interested in probabilistic modelling, causal reasoning, and integrating geometric structure with uncertainty. I evaluate these choices using classical methods and machine learning, with experiments that test the underlying assumptions and reveal each approach’s limitations. Based on this analysis, I design models suited to the structure of the problem.
Work experience
-
Jan 2024 to Dec 2025
Visiting Scholar, full-time research
University of Waterloo
Vision and Image Processing (VIP) Research Group, Waterloo, ON
Self-Supervised Camera Pose Estimation
Developed a self-supervised camera-pose model using geometric constraints, with tools to analyse estimation errors.
Welding Signal Analysis
Investigated welding anomalies through temporal and statistical analysis of sensor signals.
Infrared Object Detection and Tracking
Built an infrared detection and tracking pipeline using temporal information, with training and evaluation workflows.
-
2018 to Jul 2022
Student Researcher / Project Researcher
University of Salzburg
Multimedia Signal Processing and Security Lab (WaveLab), Salzburg, Austria
Co-authored seven peer-reviewed papers on finger-vein biometrics, covering sensor identification using PRNU analysis and CNNs, presentation-attack detection, and image compression. Also worked on satellite-image analysis for sealed-surface detection. Student research from 2018; project-researcher contracts from May 2020.
-
2008 to 2015
Earlier Career
Software Engineering, Product Management, and Teaching
Product Manager for a commercial broadcast newsroom system; IT Expert and System Analyst at Knauf; Adjunct Lecturer at the University of Applied Science and Technology (UAST), teaching Introduction to Programming (C), Object-Oriented Programming (C#), Software Modelling, and Telecommunication and Mobile Networks.
Research
Counter-UAV thermal imaging
Detecting small drones in thermal video when the camera itself is moving.
Read the pageMotion from a monocular camera
Detecting and 3-D-tracking rigid and non-rigid objects from a single moving camera.
Read the pageBiometrics
Five papers on two forensic questions about a finger-vein image: which sensor took it, and was the finger real?
Read the pageMedia security and data formats
Two papers on how far a finger-vein image can be compressed before recognition starts to fail.
Read the pageSatellite imagery
Segmenting sealed surfaces from satellite imagery for urbanisation tracking and land-use monitoring.
Read the pageSkills
-
Computer Vision and Geometry
Camera models, geometric constraints, motion analysis, and depth and pose estimation.
-
Learning and Statistical Methods
Self-supervised learning, probabilistic clustering, and analysis of temporal and statistical dependencies.
-
Image, Signal and Event Data
Temporal and frequency-domain analysis, infrared imagery, event-data representations, and frame-based event simulation.
-
Research Implementation
Python, PyTorch, NumPy, SciPy and OpenCV; model implementation, experimental pipelines, and diagnostic tools.
Publications
- 2025
- 2021
- 2019
- 2018
-
In preparation
Multi-Method Analysis of Voltage-Current Correlation for Real-Time Porosity Detection in Gas Metal Arc Welding
Education
-
Oct 2022 to Dec 2025
Doctoral Studies, Digital & Analytical Sciences (Informatics)
University of Salzburg, Austria; programme discontinued in 2025.
-
Oct 2015 to Jun 2022
Graduate Studies, Applied Image & Signal Processing
University of Salzburg & Salzburg University of Applied Sciences, Austria
111/120 credits completed; master’s thesis not submitted.
-
2005 to 2008
M.Sc., Computer Science
University of Pune (Fergusson College), India
Certifications
University of Alberta / AMII (2024)
Coursera Project Network (2024)
- Classify Radio Signals with PyTorch (guided project)
Imperial College London (2020 to 2021)