Real-time Monocular Scene-Reconstruction in Public Environments

This video shows our feature-based approach for real-time monocular scene-reconstruction (shape from motion). Our method processes a sequence of images taken by a single camera mounted frontally on a mobile robot. In contrast to other approaches where the camera is moved sidewards our camera moves along its optical axis. This leads to an ill-posed reconstruction problem. However, using a combination of various techniques, we are able to produce a precise reconstruction that is almost free from outliers and can therefore be used for reliable visual obstacle detection and 3D map building.
The upper left image shows the image of the environment as seen be the robot's front camera. The reconstructed features are shown as dots, where their estimated height is coded by different colors. At the end of this video the reconstructed point cloud is used to create a textured 3D surface model using self-organizing feature maps.

References:
Einhorn, E., Schröter, Ch., Gross, H.-M.
Monocular Scene Reconstruction for Reliable Obstacle Detection and Robot Navigation, ECMR 2009

Einhorn, E., Schröter, Ch., Gross, H.-M.
Monocular Obstacle Detection for Real-world Environments, AMS 2009

Category:

Science & Technology

Tags:

• monocular scene reconstruction
• shape-from-motion
• obstacle detection
• 3D mapping
• monocular vision
• robotics

[roidroid]

Do you give preferential assumptions to straight lines when you are reconstructing the 3d surfaces?

Most of the stuff we humans build is made of straight lines, so making that assumption when reconstructing surfaces would increase accuracy yes?