Lernmaterialien für KI in der Fahrzeugtechnik an der Fachhochschule Lübeck

• artificial intelligence that implements a limited part of mind
• only some of the subproblems are implemented

• Def - method of perfoming operations on images
• goal - to get an enhanced image or to extract useful information from it
• operations - computer vision algorithms and mathematical operations

• algorithm that includes all necessary functions to detect edges
• first derivative of Gaussiona approximates the operator, that optimizes the product of signal-to-noise ration and localization
• theoretical model: step-edges corrupted by additive Gaussian noise

• Sensor processing
• Perception
• Path and Behavioural planning
• Control

• Rule-bases, logic, data-based
• machine learning
• deep learning

• easies way of machine leraning
• following a set of rules/questions

• 1 vs. rest
  • evaluate one class against all others, treated as one class
  • output vector gives which class won
  • computational effort 0(IKI)
• 1 vs. 1
  • compare one class to each other class one by one
  • in each comparison all other classes neglected
  • element classified in class decided upon the most
  • computational effort 0(IKI^2)
• K - # of classes

• sample size is not fixed at beginning > sampling can be stopped when signicifcant results are observed
• data are evaluated as they are collected > apply regression during operation of the product
• not enough memory to store all data points

Classify a new object based on it's nearest neighbor

• scalablility
• compactness
• accuracy
• interpretability
• efficiency
• robustness

• compute regression to data with sigmoid function (like linear regression)
• combine with a rule for classification
  • new data point x gives value y via sigmoid function
  • rule for y gives classification (eg for y>0,5 tumor is malignant)
    
    

Perception!

• Extract image shape
• tranform to greyscale
• blur image
• detect canny edges
• mark special regions
• detect hough lines