Lecture 9 - Perception for Social Robots

Question 1

Good Observables

Answer 1

1. Good features ignore all irrelevant variations, accidental, recording specific variations.
2. Good Features capture all distinguishing variations, pairs-of-class specific variations
3. Good sampling provides all relevant group information either by random sampling or stratified sampling
4. Good features have good repeatability and produce little noise themselves.

Question 2

Low Pass Filter

Answer 2

Smooths an image. Retain low-frequency components

Question 3

Gaussian Noise

Answer 3

Variations in intensity drawn from a Gaussian Distribution

Question 4

Moving Average

Answer 4

Replaces each pixel with an average of all values in its neighborhood

• Pixels to be like their neighbors
• Noise to be independent of each pixel

Question 5

Weighted Moving Average

Answer 5

Add weights to the moving average

Question 6

Local Filter - Two types

Answer 6

• Cross-Correlation

- Convolutions

Question 7

Local Filters Characteristics

Answer 7

• Positive Values
• sums to 1 to preserve the brightness of regions
• Removes high-frequency components (Depend of the filter applied)

Question 8

High-Pass Filter

Answer 8

Retains the edges of an image. It highlights high frequency

Question 9

What is an edge in Computer vision

Answer 9

Sharp Intesity changes in images. They correspond to Local extrema or derivative

Question 10

Opposite signs when applying filter

Answer 10

High response in regions with constract. Sharpen an image