Left behind occupant recognition based on human tremo detection via accelerometers mounted at the car body

Abstract

The aim is an additional sensor system that is able to detect left behind occupants in parked cars. This should avoid the decrease of fatalities found in unattended or oversight individuals in vehicles. Based on acceleration measurements directly at the car chassis information about the occupancy is extracted. At the beginning of this paper the theory of the signal source, the used car model and the applied classification algorithms is shortly given. Afterwards several measurement results are presented and become analyzed with regard to a following automatic classification. The next step is the evaluation of different classification algorithms and the explanation of the performance on the acceleration datasets that were collected during this research. Many different classification algorithms are available, at this point the support vector machine (SVM), k-nearest-neighbor (k-NN), probabilistic neural network (PNN), decision tree and clustering were observed.