Human Gesture Classification for Autonomous Driving Applications using Radars

Abstract

The age of fully autonomous driving requires the vehicles to be able to fully understand the surrounding environment. Police officers and pedestrians perform different gestures and movements in streets on a daily basis. Detecting these movements and gestures and classifying them are necessary tasks that need to be achieved so as to get ready for autonomous driving. Radars, which are nowadays irreplaceable in the automotive industry, can capture the gestures and their varying signatures with time. Various important traffic scenarios, that occur everyday on the streets are going to be the focus in this paper. A radar-based signal processing chain and classification of the scenarios using convolutional neural networks is going to be presented. Data representation is also introduced to have a better insight for the data distribution.