Bavantha Udugama 0000-0002-1932-692X

Projects · Part I

Mono-Hydra

Where it started. The first real-time 3D scene graph built from a monocular camera and an IMU.

2022 to 2023 · published

System design diagram for monocular 3D scene graph construction, showing the perception networks, visual-inertial odometry, and the scene graph layers.
The original Mono-Hydra design: two independent networks for depth and semantics, feeding a robocentric VIO and the Hydra scene graph layers.

01 — The system

What it does

Learned depth and learned semantics run as separate networks, feeding a robocentric square-root VIO and the Hydra scene graph layers: mesh, objects, places, rooms, buildings.

Engineering ownership

My contribution

  • Developed and integrated monocular learned perception, visual-inertial odometry and hierarchical scene graph construction.
System sketch

From camera measurements to a layered map

  1. 01
    MeasureMonocular camera + IMU
  2. 02
    PredictSeparate depth and semantic networks
  3. 03
    Estimate & reconstructRobocentric VIO + semantic mesh
  4. 04
    StructureObjects, places, rooms, building
Simplified architecture: the camera and IMU support visual-inertial estimation; predicted depth and semantics support mapping.

02 — Evidence

System design diagram for monocular 3D scene graph construction, showing the perception networks, visual-inertial odometry, and the scene graph layers.
The original Mono-Hydra design: two independent networks for depth and semantics, feeding a robocentric VIO and the Hydra scene graph layers.

Measured

Key numbers

  1. 0.19 m and 0.21 m ITC mean error
  2. 15 fps at 848x480
  3. RTX 3080 8GB laptop

Every figure here appears in the thesis results registry with its source location. Conditions are stated with the number.

Paper

  1. Mono-Hydra: Real-Time 3D Scene Graph Construction from Monocular Camera Input with IMUPublished, ISPRS Annals 2023

Code

  1. mono_hydrasystem