This paper presents a vision-based road-barriers detection method. Because horizontal structures are hard to detect by binocular stereovision, object detection methods based on 3D points grouping fail to detect the barriers as obstacles and consequently specific ACC applications as longitudinal control will fail to react when the vehicle path is obstructed by such an object. Therefore the proposed method combines the detection of the horizontal structure of a barrier's boom in the 2D grayscale image space using a Hough based approach followed by a series of 3D validation steps based on dense stereo-vision and derived functions (lane and objects detection) in order to eliminate false positives and to infer the associated 3D information. The detected barrier objects are reported as 3D cuboids which are further tracked along with other objects/obstacles detected through a 3D points grouping method integrated in a multifunction application for vision based driving assistance.