Robotik 1 - Kapitel 9 - Bildverarbeitung in der Robotik
A HumanoidRobots lecturer outlines how robot vision turns camera data into usable scene understanding, then introduces image representation and camera calibration for robotic perception.
In this talk
- 0:02
Why robots need image understanding
Vision helps robots perceive people and objects in changing environments, supporting tasks such as visual grasping, adapting motion around obstacles, and working alongside people. The lecturer distinguishes low-level image cleanup, mid-level features and regions, and high-level scene meaning. This lecture focuses mainly on the lower levels, with more advanced semantic interpretation left to later courses.
- 15:18
Robot vision in changing scenes
Robot vision must work with real, dynamic environments, where cameras move, objects become occluded, and 2D images lose depth. What matters in a scene depends on the robot’s task, such as retrieving something from a refrigerator or cutting a cucumber. Lighting and material properties also change the image data, creating challenges for reliable perception.
- 27:10
Representing images and depth
A digital image is a grid of pixels, commonly stored row by row. An 8-bit grayscale pixel ranges from 0 to 255; an RGB image uses three 8-bit channels and can represent about 16.8 million colors. Depth cameras add distance data, while point clouds represent 3D positions with optional color, normals, or labels.
- 54:55
Camera projection and calibration
The pinhole camera model projects 3D points onto a 2D image, so depth cannot be recovered from an ordinary image alone. A calibration matrix captures focal lengths and the image’s principal point, while extrinsic parameters describe the camera’s pose relative to the world. Calibration uses known world-to-image correspondences; the lecture derives a linear system and describes solving it with least squares.
- 20:08
Tools and what comes next
The chapter will cover image filters, morphology, segmentation, edge detection, visual servoing, and point-cloud registration. The lecturer names OpenCV, the Point Cloud Library, and the Integrating Vision Toolkit as available software libraries. The next lecture continues with filter operations.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.