Dataset generation for Visual SLAM and Machine Learning

Overview

• Introduction
• Visual SLAM
• Sensor Data Fusion
• Machine Learning
• Conclusions for SciViz

Introduction

• Adam Kalisz
• Media Engineering (Bachelor)
• Computer Science (Master)
• BFCT since May 2016
• PhD Student: Monocular Visual SLAM / Sensor data fusion

Our chair at University

BCon17 Talk

See: https://www.youtube.com/watch?v=PMYET5LA-Kg

Visual SLAM

My research

Motion Tracking (libMV)

The Indirect Method (libMV)

3D Reconstruction (libMV)

Dataset in Blender

Without camera rotation

Paper: "Systematic Analysis of Direct Sparse Odometry", DICTA 2018

With camera rotation

Paper: "Systematic Analysis of Direct Sparse Odometry", DICTA 2018

Sensor Data Fusion

Sensor Fusion in Blender

Paper: "B-SLAM-SIM: A novel approach to evaluate the fusion of Visual SLAM and GPS by example of Direct Sparse Odometry and Blender", VISAPP 2019

Visual SLAM and GPS Plots

Paper: "B-SLAM-SIM: A novel approach to evaluate the fusion of Visual SLAM and GPS by example of Direct Sparse Odometry and Blender", VISAPP 2019

Machine Learning

Dataset generator

Dataset generator output

Conclusions

• Think about the idea of direct methods in Blenders Motion Tracker
• Combine ideas from both rendering and computer vision
• Use synthetic data to train and evaluate Artificial Intelligence

Blender User Group Nuremberg (NuremBUG)

https://www.nurembug.org/

Thank you!