Robotik 1 - Übung 7 - Bildverarbeitung
Markus Kotz, a lecturer with the Robotik 1 course at KIT, leads an image-processing exercise with colleague Christopher Pohl, covering color spaces, filters, morphology, and point-cloud algorithms.
In this talk
- 0:13
Convert RGB colors to HSI
Kotz explains how hue, saturation, and intensity represent color, then works through an RGB-to-HSI example using the values 120, 80, and 210. He also discusses how lighting changes affect RGB and HSI values, and shows how color segmentation can isolate an object for later pose estimation.
- 8:32
Segment colors with OpenCV
Kotz demonstrates converting an image to HSV with OpenCV and using lower and upper color bounds to create a binary mask. He applies the mask to isolate a robot in the image and converts the result back to RGB for display.
- 11:20
Apply filters and morphology
The exercise uses Prewitt filters in the X and Y directions to detect edges, and compares ways to handle image borders, including zero padding, edge duplication, and mirroring. Kotz then walks through erosion followed by dilation for opening, explaining how it removes small regions and thin edges, and describes closing as a way to bridge objects and fill holes.
- 26:43
Register point clouds with ICP
Kotz describes Iterative Closest Point as a repeated process for aligning point clouds, often by estimating a 6D translation and rotation. A triangle example uses a simple gradient method to reduce positional error, while a PCL demonstration shows alignment improving over iterations and notes that ICP may struggle with large initial differences.
- 43:55
Fit shapes with RANSAC
Kotz introduces PCL tools for calculating point-cloud normals and fitting a cylinder model with RANSAC. A noisy scan of two stacked cans shows why RANSAC can be useful, while its random nature means results are not guaranteed to match reality.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.