Obstacle detection and localization in an automated vehicle using binocular stereopsis and motion field

Abstract

In this work, a modularized obstacle detection system using binocular stereopsis and motion field was implemented for automated vehicles. The module used a Hardkernel Odroid XU4 Single Board Computer and two Leopard Imaging oCam OV5640 USB 3.0 Cameras. The binocular stereopsis algorithm uses Stereo Block Matching Algorithm in order to compute for disparities from spatially adjacent images. The motion field algorithm uses FAST Corner Detector and ORB Key point Descriptors in order to compute for velocity vectors from temporally adjacent images. Results show that the module was able to detect feature-rich obstacles such as vehicles and pedestrians, but it failed when it tried to detect featureless obstacles.