Safety Application Car Crash Detection Using Multiclass Support Vector Machine

In this paper, the application of Support Vector Machine (SVM) on multiple car crash situations for improved decision of saftey applications, e.g. airbag control systems is presented. In general, we should argue from a different perspective: we don't want to show the latest and greatest Machine Learing (ML) results. The intention of the paper is to show how state of the art products use ML for safety critical applications. It is the goal to avoid the deployment of an airbag. Today, saftey applications depend on various detection algorithms and sensor systems. The challenge for these kind of decision problems depend on crucial constraints named decision time and sensor signals which are complicated to distinguish. The system has to decide whether to fire or not to fire an airbag within a few milliseconds. We use a (Multiclass) Support Vector Machine to account for an improved classification with the given constraints. Various multiclass classification methods are rated and the two methods One-Versus-Rest and One-Versus-One are benchmarked in terms of quantities as test error, training time, memory consumption and misclassified crashes. All methods are applied to real measurement data of car crashes for the type of full frontal crashes in various conditions. We will show that One-Versus-One performs best. The method is able to classify car crash situations and improve the detection possibility. This allows for active and passive occupant safety components in the automotive area.