Methods for Traffic Data Classification with regard to Potential Safety Hazards⁎This work has been supported by the LCM K2 Center within the framework of the Austrian COMET-K2 program. Issue 7 (2021)

• Record Type: Journal Article
• Authors: Obereigner, Gunda; Tkachenko, Pavlo; del Re, Luigi
• Is Part Of: IFAC-PapersOnLine. Volume 54:Issue 7(2021)
• Page Start: 250
• Page End: 255
• Publication Date: 2021
• Subjects: Safety Analysis -- Machine Learning -- Classification -- Datasets -- Automotive; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.08.367
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Traffic data are a key element for setting up scenarios for Advanced Driver Assistant Systems (ADAS) safety and performance testing. Testing will thus reflect in some way the data used. However, there is no clear understanding in which way and how to choose the data so that the evaluation results are reliable and comprehensive. Therefore, the important scenarios in a traffic data set in view of safety analysis have to be determined. The paper presents a method with which traffic situations from a given data set are classified into different safety classes according to easily measurable features. It is shown that taking the Time To Collision (TTC) as a measure of safety and a linear Support Vector Machine (SVM) as a classifier, 64.7% of traffic situations of a validation data set were classified to the correct safety class considering only three measurable features. Thus, traffic situations from a data set can be classified fast into different safety categories, providing information to the ADAS tester if the developed device has been tested in a safe or unsafe environment.