Alexandre Bourgoin

ETH Zurich - Computer Vision Engineer - Full-Stack Web Dev

Obstacle Detection

Real-time terrain hazard perception that fuses stereo depth into a live drivable-or-blocked occupancy grid — running on embedded hardware.

Real-time terrain hazard perception for autonomous rovers. Stereo depth is fused into a live occupancy grid that classifies every cell as drivable or blocked, letting the planner slalom craters and boulders at speed — all running on embedded hardware.

PyTorch
TensorRT
Occupancy Grid
Stereo Depth
Jetson

Paper→Demo→

Open Banana

Dense metric depth and camera ego-motion recovered from a single moving camera — no stereo rig, no LiDAR.

Dense metric depth from a single moving camera — no stereo rig, no LiDAR. A self-supervised network recovers per-pixel depth and the camera's ego-motion from raw video, supervised only by a differentiable photometric reprojection loss.

PyTorch
Self-Supervised
Monocular Depth
OpenCV

Paper→Demo→

Cardiac Gaussian Deformation Field

My bachelor's thesis, graded 6/6: thousands of 3D Gaussians recover the beating heart's deformation and per-point strain from cine-MRI.

My bachelor's thesis, graded 6/6 — the top mark. The same 3D-Gaussian primitive that reconstructs a rover's world, turned inward: thousands of anisotropic Gaussians fit the left-ventricular myocardium from cine-MRI and recover the dense deformation field across the cardiac cycle — radial thickening, longitudinal shortening, and the heart's torsional wring — yielding per-point strain without segmentation masks.