Action and intention recognition of pedestrians inurban  settings  are  challenging  problems  for  Advanced  DriverAssistance Systems as well as future autonomous vehicles tomaintain  smooth  and  safe  traffic.  This  work  investigates  anumber  of  feature  extraction  methods  in  combination  withseveral machine learning algorithms to build knowledge on howto automatically detect the action and intention of pedestriansin urban traffic. We focus on the motion and head orientationto predict whether the pedestrian is about to cross the street ornot. The work is based on the Joint Attention for AutonomousDriving   (JAAD)   dataset,   which   contains   346   videoclips   ofvarious traffic scenarios captured with cameras mounted in thewindshield  of  a  car.  An  accuracy  of  72%  for  head  orientationestimation  and  85%  for  motion  detection  is  obtained  in  our experiments.