STATIONARY ON-ROAD OBSTACLES AVOIDANCE BASED ON COMPUTER VISION PRINCIPLES

Abstract

In this paper, the classification of the on-road obstacles based on the processing of a sequence of images obtained by a monocular camera embedded on a vehicle as well as the appropriate automatic guidance principle for obstacles avoidance are presented. The typical road scenarios have been used as a testing environment for the overall algorithm. Existing obstacles (vehicles) are classified into three classes: stationary, incoming, and outgoing. The first task in the algorithm consists of obstacles detection over the road background. This is followed by their tracking from one frame to another based on the appropriate selection of features using the SURF method. After that, the obstacles are recognized in a new frame, where it is possible to determine their position from the camera and the relative velocity using projection geometry principles. Then, the polynomial method is used in order to find the path that avoids the obstacles. Synthetic and realistic video sequences are used during the tests.