In my thesis I examine radar based object classifiers which are capable of
identifying the participants of common road traffic. Firstly I will present the design of the
classifiers used and I will describe the methodology of designing and tuning the
algorithms, which are necessary to use the classifiers in modern driver assistance systems.
I then introduce the object classes to be determined by the classifiers. Following
this, I discuss the aspects of collecting and preprocessing the input data which have a
strong impact on the classification performance.
To solve the classification problem, I used the Python programming language and
frameworks which support the implementation of modern classifiers (Keras, LibSVM) to
realize the algorithms. While chosing the right parameters, my focus was on the
examination of the effects of the parameters without completeness of performance
Different methods to evaluate a classifier are then introduced. Based on the most relevant
aspects from my point of view, an evaluation and comparison of the realized classifiers
is presented. At the end of my thesis I suggest further improvements which may increase
the effectiveness and usability of the classifiers in driver assistance systems.