In today's vehicles, driving support systems are becoming increasingly popular. They help the driver during his trip. Such systems are for example the lane departure warning system, adaptive cruise control, or automatic emergency braking system.
The topic of the dissertation is to evaluate the performance of one of these functions such as the video-based traffic sign recognition system. To do this, a tool chain has been implemented to extract the data required for the function from video sequences. These sequences are files that contain all video and vehicle data. After extracting the needed data, it runs the function, then evaluate the performance with actual values (ground truths).
The main components of the tool chain are: RSF_Dumper, RSF_Validator, and RSF_perf_gen. RSF_Dumper is responsible for extracting the data needed to run the RSF from the video sequences, and then write it into an xml file. RSF_Validator runs the function using these files using several types of parameterization. Its task is to compare the results of the run with the ground truth values for each parameter. RSF_perf_gen then summarizes the results obtained and generates an html file for each parameter that contains the performance results.
The toolchain was tested with eight hours of recorded video sequences. During the run, the effect of each parameter has been tested for performance. Such parameter was the one that determines the minimum quality of the tables we accept.
Most of the development was done in C ++, other parts of the tool chain were implemented in Python.