Robotik 1 - Kapitel 9 - Bildverarbeitung in der Robotik (Teil 2)

In this talk

1. 0:45

   How image filters work

   The lecture explains how a filter mask operates on neighboring pixels: multiply each pixel by its corresponding mask coefficient, then sum the results. It distinguishes correlation from convolution, which rotates the mask by 180 degrees, and describes padding methods such as zero, wrap, mirror, and replicate to handle image borders.

2. 25:11

   Smoothing and edge detection

   Median, mean, and Gaussian filters smooth images, with the median filter especially useful for salt-and-pepper noise and Gaussian smoothing controlled by sigma. Prewitt and Sobel filters estimate directional gradients to detect edges; Laplace filters use second derivatives and zero crossings, while Laplacian of Gaussian smooths first to reduce noise sensitivity.

3. 1:01:42

   Morphological operations for binary images

   After thresholding or color segmentation, dilation expands connected regions and erosion removes isolated pixels and weak connections. Their effects depend on the size and shape of the structuring element. Applying erosion then dilation is opening, which removes small exterior regions; dilation then erosion is closing, which bridges small gaps and fills interior holes.

4. 1:14:55

   Thresholding and color segmentation

   Segmentation assigns pixels to meaningful image regions, supporting tasks such as identifying objects in scenes. The lecture describes thresholding grayscale pixels into black and white, and selecting color ranges in HSV or by distance from a target RGB color. It notes that lighting, shadows, and reflections can make these methods unreliable across environments.