Real Time 2D Pose Estimation for Pedestrian Path Estimation Using GPU Computing

Future fully autonomous and partially autonomous cars equipped with Advanced Driver Assistant Systems (ADAS) should assure safety for the pedestrian. One of the critical tasks is to determine if the pedestrian is crossing the road in the path of the ego-vehicle, in order to issue the required alerts for the driver or even safety breaking action. In this paper, we investigate the use of 2D pose estimators to determine the direction and speed of the pedestrian crossing the road in front of a vehicle. Pose estimation of body parts, such as right eye, left knee, right foot, etc… is used for determining the pedestrian orientation while tracking these key points between frames is used to determine the pedestrian speed. The pedestrian orientation and speed are the two required elements for the basic path estimation. High Performance Computing (HPC) has recently been considerably improved, for instance GPU Computing has been developed to solve complex problems and transform the Graphics Processing Unit (GPU) into a massively parallel processor. To enhance the performance of our pose estimator we limit the process to a region of interest (ROI) where the pedestrian is initially detected using a GPU accelerated detector.