117. CNN HAND AND FACE DETECTOR WITH HEAD POSE ESTIMATION ON THE ROAD

Department: Electrical & Computer Engineering
Faculty Advisor(s): Mohan M. Trivedi

Primary Student
Name: Kevan Chun Yiu Yuen
Email: kcyuen@ucsd.edu
Phone: 858-226-4600
Grad Year: 2016

Abstract
In this work, we focus on the interior activities of the driver under real world scenarios. We train a near real-time deep CNN multi-view hand and face detector while simultaneously estimating discrete yaw angle of the head pose. The hand detector may be used for tracking the driver's behaviors to determine if he is driving normally (e.g. hands on wheel) or engaged/distracted in another activity (e.g. texting on a phone, adjusting the GPS/radio/instruments). The face detector and coarse head pose estimation allows for the system to determine in which direction the driver is looking towards and will be able to warn the driver if they are not paying attention to critical areas while driving. Because the system also estimate the head pose of the detected face, it can also be used as an excellent initialization for landmark localization algorithms. We will also examine the effects of training with augmented data with occlusion and harsh lighting conditions in which a driver's face may encounter.

« Back to Posters or Search Results


Contact:   researchexpo@soe.ucsd.edu   (858) 534-6068