Project Overview
This research project was carried out in collaboration with Sony and ETH Zurich's Robotics Perception Group under Prof. Davide Scaramuzza. The focus was depth estimation using an event camera paired with a laser projector scanned at up to 350 Hz.
What made the work compelling was the sensing regime itself. Instead of relying on standard frame-based image pipelines, the project explored how event-driven measurements could be used to recover scene structure with extremely high temporal precision.
Technical Approach
Event-Based Vision
Worked with asynchronous visual data streams that respond to brightness changes rather than full image frames
Depth Estimation
Used active illumination and event sensing to reason about scene geometry at high speed
Laser Projection
Integrated a laser projector scanned at up to 350 Hz as part of the measurement setup
Research Workflow
Worked in a research-heavy environment where experimental design and interpretation mattered as much as implementation
Why It Was Technically Interesting
- Unconventional sensing: Event cameras require a different mental model than standard frame-based vision systems
- Temporal precision: The setup pushed toward sensing and reconstruction at a speed regime that standard pipelines struggle to match
- Active perception: The laser projector turned the problem into a tightly coupled sensing-and-inference task
- State-of-the-art context: The project sat close to the frontier of robotics perception research rather than routine product engineering
Research Context
This project expanded my understanding of how perception systems can be designed when timing and sensing characteristics fundamentally differ from conventional camera pipelines. It also strengthened my interest in problems where geometry, sensing hardware, and algorithm design have to be considered together.
The experience was valuable not just because of the topic, but because it sharpened how I think about experimental robotics work: choosing the right sensing setup, understanding failure modes, and evaluating results in a way that is technically honest.
Impact
This was one of the projects that pushed me deeper into perception research and reinforced my interest in state estimation and 3D vision under difficult sensing conditions. It also gave me a stronger base for later work on real-time tracking and navigation problems where conventional assumptions do not hold.