A ROI setting method for vehicle detection in urban environment

Abstract

The application of computer vision technology to traffic detecting and information collecting is an important subject in Intelligent Transportation System (ITS). Generally, for the convenience of vision-based vehicle detection and tracking, many algorithms set the region of interesting (ROI) for the image processing manually, that is virtual loops technology. Many factors impact on the ROIs setting, such as the position or the view angle and the focus of the camera, the distance between the camera and the road, etc. Setting the ROIs is by no means a trivial thing, and moreover, these factors may be changed for some reasons. Many potential problems exist in the methods. In this paper, all the vehicles are assumed to be located on the planar region. According to this geometric constraint, a planar region detection method is employed to recognize the road region and the road side. Combined with optical flow-based motion segmentation, the proposed method can differentiate the vehicle motion on the road from other events on the roadside without setting the ROIs. Some experiments are conducted to validate the proposed the method.